Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Accuracy

Judge how far to trust HPR Sim’s numbers from its checks against other programs and real flights, model by model, with every known gap.

This page gathers every check HPR Sim has passed so far, and every known gap, in words and numbers. Trusting a simulation explains what they mean for a flier.

When both codes fly the same drag (same-drag), HPR Sim’s whole flights match RocketPy’s in height, speed and time. Without wind they also land where RocketPy’s do. In wind they agree for a rocket that leaves the rail fast. For one that leaves it slowly they differ, in large part because HPR Sim includes a sideways force on the body that RocketPy leaves out.

With each code’s own drag (predicted), HPR Sim’s heights differ from RocketPy’s by −7.280% to +10.302% (report). The larger gaps are where its drag differs most from the example’s.

Against the logs of 12 public real flights, HPR Sim’s apogees miss by 8.88% on average, outside the 5% target (evidence row G24). One by one they miss by −29.16% to +10.40%, and 7 of the 12 read low (real flights, report). Seven were compared as soon as they were flown. Of their five misses past 5%, four are consistent with HPR Sim’s drag, and one with the motor’s impulse. Five were predicted first. Four of those miss past 5%, and no explanation is claimed for them. Seven of the 12 altimeters’ kinds are assumed, and none is calibrated (evidence rows G21 and G39).

Against 60 more flights from a private collection, HPR Sim’s apogees are +8.01% above the logs on average. They miss by 12.09% (mean absolute), also outside the target. OpenRocket’s are +6.72% above (private collection, report).

No flight here is blind yet: all were seen during development. Blind flights will be predicted before their logs are read. Five of the public flights had their predictions merged before they were compared (predicted before compared). They aren’t blind either: their measured apogees were already public.

What has been checked so far:

  • Each model on its own, against exact answers, its published source and, in places, RocketPy, an open-source flight simulator.
  • The descent under a parachute, against RocketPy, for five rockets.
  • Whole flights from the pad to the ground, against RocketPy, for six rockets flown with one declared drag coefficient. One flies past Mach 1.
  • The same flights with each code’s own drag, reported against a target rather than gated, the one past Mach 1 included.
  • Twelve public real flights, ten of them flown in ERA5 weather, nine in their own day’s. Their apogees are compared with their altitude logs, and for seven the climb too (real flights).
  • 60 more real flights of a private collection, flown by HPR Sim and OpenRocket in the weather of their day. They’re compared with their logged apogees as aggregates (private collection, report).
  • The normal force and center of pressure from Mach 0.6 to 4.63, against NASA’s wind-tunnel tests of a sounding rocket. Also against RASAero II, another code (fixture). And from Mach 1.08 to 3.15 against free flights of the Basic Finner, a standard finned test body, in a Canadian aeroballistic range (Aerodynamics).
  • Drag from Mach 0.6 to 4.63 against the same wind-tunnel tests, the forebody only (Aerodynamics). And the drag and pitch damping from Mach 1.08 to 3.15 against the Basic Finner’s free flights (Aerodynamics).
  • A boattail’s own drag and its base pressure, from Mach 0.3 to 3.24, against measured boattails in six NACA and NASA reports (drag fixture).
  • The roll that canted fins give, from Mach 1.5 to 4.63, against NASA’s Arcas Robin wind-tunnel tests. The roll damping from Mach 1.5 to 3, against the Basic Finner’s, a standard finned test body (roll fixture).
  • Drag from Mach 0.1 to 2.0 against the curves labeled RASAero II in RocketPy’s example rockets (Aerodynamics). Drag from Mach 0.5 to 3.2 against a worked example in MIL-HDBK-762, the U.S. Army’s handbook for designing unguided rockets (Aerodynamics).
  • The readings hpr analyze takes from a flight log, against an invented log whose every reading is known (evidence row G10). Each is within the bound its rounding and filter allow (Flight-log readings).

Every number here links to the page or file it comes from. Checking a claim shows how to follow one back to its source and its test, and how the site keeps the two in step.

The census

In short: the census counts the numbers the validation reports hold HPR Sim to, once each. It says what each group was compared with, how, how many flights it holds and how fast they flew. It’s also a check: CI fails when any of those numbers moves, better or worse, until the change is accepted with a written reason.

CI flies the harness’s RocketPy comparisons again on every change. The OpenRocket and real-flight reports need files that CI doesn’t have, so CI holds their committed numbers. A change to the code that would move them is caught only when someone runs them again and commits the result.

compared withkindheld toflights (speed)result
RocketPy 1.13.0: descents under a parachutecode-to-code, same inputsgate: each metric’s tolerance, 0.1% to 3%5 descents30 of 30 gated metrics pass
RocketPy 1.13.0, patched: whole flights on the same dragcode-to-code, same inputsgate: each metric’s tolerance, 0.05% to 3%9 flights (8 subsonic, 1 transonic)142 of 142 gated metrics pass; apogee +0.04% to +1.21%; 11 not scored, each for a written reason
RocketPy 1.13.0, patched: whole flights, each code on its own dragcode-to-code, each code’s own dragtarget: each metric’s tolerance, 0.1% to 3%, reported, not enforced6 flights (5 subsonic, 1 transonic)75 of 102 metrics within target; apogee −7.28% to +10.30%
OpenRocket 24.12: calm flights of OpenRocket’s examplescode-to-code, each code’s own modelno target; an apogee more than 5% off needs a written cause54 flights (53 subsonic, 1 transonic); 3 more not flownapogee −37.93% to +13.80%; apogee within 5% on 45 of 53, largest speed within 5% on 50 of 53, margin within 0.2 calibers on 52 of 53, launch mass within 1% on 54 of 54, mass at rod clearance within 1% on 50 of 54, center of mass at rod clearance within 0.2 calibers on 54 of 54; 3 values withheld by the report
OpenRocket 24.12: calm flights of the private designscode-to-code, each code’s own modelno target; an apogee more than 5% off needs a written cause35 flights (28 subsonic, 6 transonic, 1 supersonic); 2 more not flownapogee −4.84% to +13.60%; apogee within 5% on 34 of 35, largest speed within 5% on 34 of 35, margin within 0.2 calibers on 35 of 35, launch mass within 1% on 35 of 35, mass at rod clearance within 1% on 35 of 35, center of mass at rod clearance within 0.2 calibers on 35 of 35
the public flights’ altimeter logs: real flightsmeasuredtarget: mean absolute apogee error 5%12 flights (5 subsonic, 7 transonic)mean absolute apogee error 8.88% (target 5%, missed); apogee −29.16% to +10.40%; apogee within 5% on 3 of 12, climb’s RMS height error within 3% of apogee on 2 of 7

Reading the table.

  • Compared with names the reference and its version (census): RocketPy, OpenRocket or the public flights’ altimeter logs. “Patched” means RocketPy 1.13.0 runs with two fixes from RocketPy’s own pull requests. They correct the point it takes the burn’s turning moments about (whole flights against RocketPy).
  • Kind says whether both programs flew the same inputs, each flew its own model, or HPR Sim was compared with a measurement.
  • Held to is what each report holds its numbers to (gate and target). A gate fails the run when a number falls outside it. A target is reported: missing it never fails the run. The OpenRocket comparisons have neither, only a line past which an apogee needs a written cause.
  • Flights (speed) classes each flight by its largest Mach number: subsonic below 0.8, transonic from 0.8 to 1.2, supersonic above 1.2 (census). For the harness’s flights that’s RocketPy’s largest Mach number, and for OpenRocket’s flights OpenRocket’s. For the logged flights it’s HPR Sim’s own, since a log has none. A configuration an OpenRocket report lists and HPR Sim doesn’t fly yet is counted as “not flown”, with the report’s reason.

What it counts. Each of these is one row of the census (census):

  • every metric of every case in the harness’s report, and each known gap;
  • each real flight’s apogee, and the RMS of the climb for the seven compared along it;
  • each OpenRocket flight’s apogee, largest speed, stability margin, mass at launch and at rod clearance, and center of mass there, and each configuration not flown.

It leaves out the reports’ figures that explain a difference rather than measure one. A real flight’s apogee on its team’s own drag is one.

The check. The census accepted last is committed, with a page of its own (census). cargo xtask validate --check, which CI runs on macOS, Windows and Linux, holds every row of the committed reports to it. To run it yourself, clone the repository and run cargo xtask validate --check. It fails when:

  • a row’s difference moves by more than its slack, in either direction;
  • a row’s standing changes, for instance a miss that starts meeting its target;
  • a row comes or goes, for instance a flight HPR Sim used to refuse and now flies;
  • a group’s reference changes, such as a new version of OpenRocket.

A row’s slack is 0.1% of its scale. For a harness row it’s at least the harness’s reproduction bound: the larger of 2e-6 in the metric’s own unit and 1e-7 of the value. The scale is the row’s own tolerance where it has one. Otherwise it’s the bar of its kind. That’s 3% of its reference for a harness metric not scored, 5% for an OpenRocket apogee or largest speed and 1% for an OpenRocket mass. It’s 0.2 calibers for a stability margin or center of mass, and 5% for a logged apogee. A logged climb’s RMS gets 3% of the apogee, the bound three RocketPy comparisons hold a flight’s height RMS to (census). For example, a 3% tolerance on a 1,000 m (3,281 ft) apogee allows 30 m (98 ft). The census lets the difference move by 0.03 m (0.098 ft) before it fails.

So the census holds a number to where it was, not to its target. A flight on each code’s own drag, whose bound is only a target, is held that way too (census). That keeps HPR Sim’s own aerodynamics from drifting unseen inside the target.

Accepting a change. cargo xtask census --accept --reason "<why>" writes the new census and the table above. The reason and the list of what changed go into the census page. So a worse number can still merge, but only in writing, in the change that brings it. An improvement has to be accepted too. Otherwise it could slip back later without anyone seeing.

The scoreboard

In short: the scoreboard sets HPR Sim beside OpenRocket and RocketPy, one table per axis the project judges itself by. The axes are:

  • accuracy against logged flights;
  • bias;
  • whether predicted ranges hold real flights;
  • speed;
  • round trips through each tool’s design file;
  • features;
  • steps to a first answer.

Every number is read from a committed report. Many cells aren’t measured yet. Each says so and links the milestone that will measure it. On each row where HPR Sim and another tool both are, it says which is ahead.

It’s also a check. CI fails when one of HPR Sim’s numbers there gets worse than the one main holds, counts fewer flights, or stops being measured. The exception is a cell a decision record accepts for the release the change ships in (how the scoreboard works). The other tools’ numbers, and HPR Sim’s on held-out flights, are shown, not held. cargo xtask scoreboard writes it from the committed reports.

Its steps rows count how many steps it takes to open a design’s .ork file, fly it once and get its landing spread. Commands, script lines, clicks and typed entries are each counted apart. These are counts of documented steps, not timings, and installing isn’t counted (what counts as a step).

  • HPR Sim: 2 commands, or 10 lines of Python, from files a test runs, so every counted step works. It has no window yet (its steps).
  • OpenRocket 24.12: 7 clicks to open and fly a design, from its documentation. It has no Monte Carlo to spread the landing, and no command or documented script that flies a design (its steps).
  • RocketPy 1.13.0: it reads no .ork file. A rocket is typed in as Python instead (its steps).

Reading the evidence

How the numbers are written, and the four kinds of check they come from.

How to read the numbers

  • Powers of ten. Very small and very large numbers are written the way programs print them. The number after the e says how many places the decimal point moves, to the left when it’s negative. So 1e-12 is a millionth of a millionth, a 1 in the twelfth decimal place, and 1e6 is a million. Both appear in the Frames rows of the results table below.
  • Absolute differences carry a unit. They say how far apart two values are, in that unit. Geodesy’s round trips return heights within 2e-8 m, twenty billionths of a meter.
  • Relative differences are marked relative: the difference as a fraction of the value it’s compared with. Gravity within 2e-14 relative means within two parts in a hundred million million. A percentage is a relative difference counted in hundredths.
  • Signs. A signed difference is HPR Sim’s value less the one it’s compared with, over that one. A plus means HPR Sim’s value is the larger in size, a minus the smaller.

Four kinds of evidence

A model can be checked in four ways, from the weakest to the strongest evidence that it matches reality. They’re set out in the project’s validation plan.

kindwhat is comparedwhat agreement shows
AnalyticThe code against exact answers: formulas solved by hand (closed forms), conservation laws, and round trips (converting a value and converting it back)The code computes what its equations say
Published sourceThe code against a source’s printed tables and worked examplesThe code implements the source correctly
Another codeHPR Sim against another simulator, such as RocketPy, flying the same inputsThe two codes agree on the physics; not that either matches reality
Real flightsHPR Sim against measured flightsThe model matches reality, within the flight’s own uncertainty, which here is not measured

The first three check the code. Only the fourth checks the physics against the world.

Twelve public flights have been compared so far, each as a whole. That’s the apogee, and for seven the climb to it, not each model on its own (real flights). Two recovery models are also checked against published drop tests (Recovery), and the normal force against NASA’s wind-tunnel tests of the Arcas Robin (Aerodynamics): measurements, but not flights.

Model by model

Which checks each model has passed, then its headline result.

Where each model stands

A tick means the model has been checked that way; a “no” means it hasn’t yet. In the descents only means RocketPy’s air density and wind were compared with HPR Sim’s at the 23 heights its parachute descents sample. That was part of the descent comparison, and nowhere else (Recovery).

modelanalyticpublished sourceanother codereal flights
Frames✓no✓ RocketPy; the codes of NASA’s check cases 1 to 3 and 6 to 10no
Geodesy✓✓nono
Gravity✓✓✓ RocketPy; the codes of NASA’s check case 1, whose field stops at its second termno
The magnetic fieldno✓nono
Atmosphere✓✓✓ RocketPy, in the descents only; the codes of NASA’s check cases 6 to 10partial: its pressure altitude against two altimeters’ readings of their own pressure, to 0.195 m (0.64 ft) and 1.321 m (4.33 ft, report)
Wind✓no✓ RocketPy, in the descents only; the codes of NASA’s check cases 7 and 8no
Turbulence✓nonono
Design tree✓no✓ RocketPy, mass properties onlyno
Shapes✓no✓ OpenRocket, probe parts and the parts catalogno
Mass properties✓no✓ OpenRocket, the structure without motors, and the parts of its parts catalogno
Solid motors✓no✓ RocketPy, ThrustCurve.org; OpenRocket for a curve file’s own numbersno
Aerodynamics✓✓ Barrowman’s examples; MIL-HDBK-762’s drag example, fins left out: 6 of 12 within 10%, the body reading 6% to 10% low faster than soundpartial: drag and the normal force against RASAero II to Mach 2, the drag with the fins and finish guessed and 5% to 15% low faster than sound; and in whole flights, against a targetpartial: in the seven flights compared first, four of the five apogees past 5% are consistent with its drag (real flights, report); wind tunnel ✓: normal force, drag and boattails; the Arcas Robin’s drag reads high at every speed
Rigid-body flight✓no✓ RocketPy, with the drag given; and on each code’s own drag, against a target; OpenRocket on 53 configurations of its examples (results); the codes of NASA’s check cases 1 to 3 and 6 to 10, on a free body under the cases’ damping and drag (results)partial: 12 public logged flights, as a whole: apogees 8.88% off on average, outside the 5% target (evidence row G24) (real flights, report)
Time integration✓nopartial: NASA’s check case 2, together with the rigid-body equations, against the two of its codes that agree with each otherno
Recovery✓✓✓ RocketPyno (drop tests ✓)
Staging✓ ignition times, the mass step and momentum, against hand sumsno✓ OpenRocket: its two-stage, three-stage, cluster and air-start examples, every flight within 5% (three cluster apogees against OpenRocket’s flight with no parachute, because its parachute opened before apogee) (results)no
Moving mass✓ the mass properties and the static margin against hand calculations, and both momenta kept in free flightnonono
Released mass✓ the mass properties after a release against hand calculations, and mass and both momenta kept across it in free flightnonono
Flight metrics✓ the boost’s peak, the margins of the page’s finless rocket with a boattail (including where a margin is withheld) and the landing placement against hand calculations; max q and top Mach against a 1 ms record; the least margins against 1 ms steps and a scan of every step; the flight margin and the optimum delay against HPR Sim’s own models and a flight with no recoverynonono
Fin flutter✓ the scaling in thickness, stiffness and pressure, and the margin at max q against a 1 ms record✓ Martin’s formula and both of his worked examples reproduced, with his verdicts against his figure 3 band; not compared with flutter datanono
Flight-log readings✓ each reading of an invented flight within the bound its rounding and filter allownopartial: one public altimeter log’s apogee, read at 1,010 ft against the 1,009 ft the altimeter’s software states from the same trace; checked only where the log is fetched, not in CIno
Synthetic flights✓ each noise term on a truth built for it, failing when changedno: the terms are taken from datasheets and public logs, not yet compared with real logs’ statisticsnono
Two logs on one timeline✓ the offset within its bound on all 1,383 two-logger and 135 staged synthetic flights that gave one, largest error 0.21 s; its 90% interval passes the coverage test on both two-logger sets, narrowed, widened and shifted each failingnopartial: one real pair, a weak check: RocketPy’s Prometheus, about 0.08 s from the offset its two boards’ GPS time gives, where a zero offset would miss by 0.13 s; no real staged flightno
A log against its simulation✓ the sign, a barometer’s reading on a warm day and the clock on constructed flights; a log of the simulation’s own flight leaves an RMS under 1 m (3.3 ft)nopartial: each of the 7 public flights compared along the climb is reported, and its simulated apogee, read by the log’s reader, stays within 0.5 m (1.6 ft) of the flight’s own event; this reports residuals, and checks no claim about themno
Ghost files✓ each format read back by outside readers and held to the others within 1 cm (0.39 in)no✓ all 22 real flight logs in RocketPy’s repository with an altitude or a GNSS track: of the 16 logs with a barometer’s or a filtered height, 13 peak within 2% of a recorded apogee in RocketPy’s notebooks and 3 about 0.3% from its simulated one; the two with only a receiver’s altitude 4.9% and 5.9% overno
The least-squares fit✓ each parameter’s 90% interval passes on sets A and B of the synthetic flights, the wind’s on logs with a track, sets seen while its model was chosennonono: no real flight with its inputs measured apart from its log yet (evidence row G38)
The smoother✓ at burnout, the top speed, apogee and a drawn moment, the three sets’ cells hold between 94.2% and 99.4% of the synthetic flights’ truths inside its 90% intervals; two logger cells fall under 90%, 88.7% on set B and 89.2% on set A, which the check’s rule allows; the test sets were seen while its rules were set (how well)nonono: no real log with a truth taken apart from it yet (evidence row G33)
Interpolation✓nonono
Quadrature✓nonono

Results by model

The headline results, one check to a row. Each model page opens with In short. Its verification section lists every test and its tolerance: how far a result may be from its reference and still pass.

modelcompared withhow close
FramesRocketPy’s starting attitude, worked out from the launch rail’s angles, for 8 rail setupswithin 1e-12 rad
Framesan exactly solvable spin whose axis sweeps round a cone (coning), over 1e6 integration stepsattitude within 1e-9 rad
Framesthe codes of NASA’s check cases 1 and 2 (a free body against them, report): the height above the ellipsoid read back from the launch frame along a 4.4 km (2.7 mi) fall down the site’s vertical on the equator, where the ellipsoid’s shape hardly enters, and the Earth’s rate turned into the body’s axeswith the two named model differences taken off each kept code’s largest difference, at most 0.1364 mm (0.0054 in) of altitude and 0.001005 deg/s of inertial rate are left, both inside the codes’ spread; read at every sample, the rates are left just past it (the rigid-body flight row); the attitude is not compared
FramesNASA’s check cases 9 and 10, a sphere with a constant drag coefficient shot from sea level at 1,000 ft/s (305 m/s) up and 1,000 ft/s east or north for 30 s, kilometers downrange, against six codes’ published positions, one left out in case 10 (a free body against them, report)the position ends 1,980 and 1,977 mm (78 in) from the farthest kept code, inside the codes’ spreads of 1,989 and 1,986 mm (78 in); with the gravity and the codes’ own air taken off, 0.03830 mm (0.0015 in) and 0.04217 mm (0.0017 in) from the nearest, but 1,991 and 1,988 mm (78 in) from the farthest, just past the spread, so no cause is claimed
Geodesythe ellipsoid values printed in Table 3.5 of the WGS 84 standardto their printed digits
GeodesyKarney’s published test set of 500,000 WGS 84 geodesics, solved both waysall 500,000 within Karney’s 15 nm (nanometers): distance within 11.18 nm, far point within 14.02 nm, heading there within 13.99 nm; the computed bearing and distance, flown forward from the first place, land within 11.26 nm. CI checks every 500th line and the 21 mirror lines; the whole set where downloaded, measured on macOS
A launch site’s elevation from a fileGDAL 3.12.2’s reading, through rasterio 1.5.2, of seven GeoTIFF files cut from a USGS tile and of the whole tile (the tests)the same corner and pixel size to the last bit, the same sums over all 13 million pixels, and the same pixel and value at all 2,368 of 2,800 places on the seven files (9 of them nodata) and all 1,696 of 2,000 on the tile, the rest off the edge; the whole tile only where downloaded
Geodesyround trips at random points from −10 km (−6.2 mi) to +1000 km (621 mi): latitude, longitude and height to Earth-centered x, y, z, and backlatitude within 1e-14 rad, height within 2e-8 m
Gravitythe WGS 84 formulas, worked to 40 digits by a separate script, at 11 points from the equator to both poles and up to 200 km (124 mi) highgravity’s strength within 2e-14 relative
GravityRocketPy’s gravity formula, at 8 pointsunder 1e-12 relative
GravityNASA’s check case 1, a sphere falling for 30 s from 30,000 ft (9,144 m) with no drag, against five codes’ published histories, a sixth left out (a free body against them, report)the altitude ends 19.51 mm (0.77 in) lower, 137 times the codes’ spread of 0.1424 mm (0.0056 in). The codes’ field stops at its second term; the terms normal gravity holds past it account for 19.50 mm (0.77 in), worked out apart from the flight, which leaves at most 0.1364 mm (0.0054 in) of a kept code’s largest difference, inside the spread
The magnetic fieldNOAA’s test values for WMM2025: the report’s 12 test points and NCEI’s 100 high-precision pointsthe 12 to their last printed digit; the 100 to their last digit in declination, inclination, the east component and four rates. The north component differs at 97 points by up to 7.18e-4 nT, and the horizontal and total intensities by up to as much with it; the down component, moved by the same residue, by up to 2.2e-6 nT, and the rates of the north, horizontal and total components by up to 1.5e-6 nT a year. A test places the north component’s difference in NCEI’s file; the rates’ difference has no known cause
Atmospherethe tables of the 1976 standard atmosphere, at 32 heights from −2 to 86 km (−1.2 to 53 mi)every value within 0.1%
AtmosphereCIPM-2007 (Picard et al., 2008), a published reference formula for the density of humid air that treats air as a real gas, from 15 to 27 °Chumid-air density within 0.047%
AtmosphereRocketPy’s air density, at the 23 heights its descents sample (Recovery)within 3.7e-4 relative
AtmosphereNASA’s check case 6, a sphere with a constant drag coefficient dropped for 30 s from 30,000 ft (9,144 m) through still air, against five codes’ published positions, a sixth left out (a free body against them, report)the position ends 102.6 mm (4.0 in) from the farthest kept code, past the codes’ spread of 85.03 mm (3.3 in). The codes don’t fly the same air: one’s density runs up to 1.35e-3 above the 1976 standard atmosphere’s. The gravity past the codes’ second term and each code’s own density, flown apart from the flight, leave at most 77.57 mm (3.1 in), inside the spread; that code is brought inside, not explained
WindRocketPy’s wind, at the same heights (Recovery)each component within 1e-9 m/s
Windthe drift of RocketPy’s four descents with wind (Recovery)the distance drifted within 0.28%
WindNASA’s check cases 7 and 8, case 6’s sphere in a steady 20 ft/s (6.1 m/s) wind and in one that grows with height to 70 ft/s (21 m/s) at the start, against five codes’ published positions, a sixth left out (a free body against them, report)the position ends 102.8 mm (4.0 in) and 105.0 mm (4.1 in) from the farthest kept code, past the codes’ spreads of 85.24 mm (3.4 in) and 87.74 mm (3.5 in); the same model difference as case 6’s leaves at most 77.63 and 77.68 mm (3.1 in), inside, bringing SIM 1 inside without explaining it
Turbulencethe Dryden gust spectra, published formulas for how gust strength spreads over wavelength, over 2²⁰ random samples (about a million)within 4 standard errors (the scatter expected by chance) in every octave band (a range of wavelengths spanning a factor of two): ±1–3% in the wide bands. Unvalidated for rockets (evidence row G12)
Design treea rocket worked by hand, loaded, burning and burnt outmass within 1e-12 kg, center of mass within 1e-12 m, inertia within 1e-12 relative
Design treeRocketPy, for twelve cases of its example rockets, at the times its equation solver computed the burning grains (up to 60 per case; the thirteenth case, Prometheus’s, has a motor with no such times)mass, center of mass and inertia within 8.0e-10 relative (the center as a fraction of the rocket’s length)
Design treeRocketPy, for all thirteen cases, at 103 even times through the burn and after it, where RocketPy interpolates between its solver’s timesmass within 1.4e-5 relative, inertia within 2.6e-5 relative
Design treethe propellant mass left in the grains, at both sets of timeswithin 2.5e-9 of the initial propellant mass at the solver’s times, and 6.4e-5 between them
Shapesexact formulas (closed forms) for filled noses and transitionswithin 1e-10 relative
Shapesseparately computed high-precision integrals, for 22 noses and transitionswithin 1e-12 relative
Shapesthe same kind of integrals, for 20 hollow shells of a given wall thicknesswithin 1e-10 relative
ShapesOpenRocket 24.12’s mass of three walled probe parts: a conical and an ogive nose and a conical transition (probes)within 3.93e-5 relative (the test holds 5e-5)
Mass propertiesa cone, a tube, four fins and an off-axis payload, added up by handwithin 1e-11 relative
Mass propertiesfin cross-sections, against exact numerical integrationwithin 1e-13 relative
Mass propertiesmaterial densities, converted from the units their sources printthe sources’ values, such as white ash at 678 kg/m³
Mass propertiesOpenRocket 24.12’s structure (every stage, no motor), on 71 compared designs in the current scratch-excluding surveymass within 1% on 70 and center of mass within 1% of length on 70, the one file outside with a named cause: airfoil fins OpenRocket weighs by a factor, a cause named only when weighing them its way brings the design within both thresholds; pitch inertia within 1% on 63, the 8 outside with no named cause yet but the two copies of OpenRocket’s tube fin example (below); roll inertia a median 1.619% apart, which OpenRocket’s shortcut for fins accounts for, and on the cluster designs its stacking of their tubes on the axis (clusters): with the shortcut in HPR Sim’s place the median is 0.000%, and 7 files (4 distinct designs) remain outside 1% with a named cause.
Mass propertiesthe same survey, tube fins and airfoil finsA ring of tube fins departs in both inertias, kept on purpose: OpenRocket’s roll inertia for it is more than any mass inside the ring could have, and its pitch inertia leaves out how far the tubes sit from the axis, which accounts for the tube fin example’s pitch inertia being 1.98% below OpenRocket’s (tube fins). HPR Sim’s airfoil fins are 19.4% lighter than OpenRocket’s, a departure kept on purpose (fins).
Mass propertieswhat a file leaves unsaid (a wall of no thickness, no material) and which override wins, held to OpenRocket’s on 32 probe designs, and each fin section and each kind of part alone on 50 more, packed parts among them (packed parts)On those that ask what a file leaves unsaid: mass within 0.001%, center of mass within 0.001 mm (0.000039 in), bar an elliptical fin’s 0.18%, and an attached tube that writes no thickness (−2.4% on the committed probe; measured in the .ork conventions decision). On the override probes: two rules kept as measured departures, and flags that disagree (4.7 mm (0.19 in)) (probes); a stage’s override on the lower of two stages, the rocket’s inertia across the axis 4.95% and 3.33% over, and 1.6% and 1.1% under with a motor aboard (worked by hand) (decision)
Mass propertiesfin fillets, on 9 of those probes (fillets, decision)agree to 1e-15 in the fillets’ mass and center of mass (the test holds 1e-12), the whole probe’s pitch inertia up to 0.64% apart
Mass propertiesan automatic radius inside a nose cone, on 14 more probes (13 of which HPR Sim flies)puts every part inside but one packed mass component (#186) at OpenRocket’s mass to 1e-14 and station to 1e-15, as the test holds them; the nose cones’ and the transition’s own walls keep their earlier gaps, up to 3.93e-5 of mass and 2.5e-6 m (the format guide)
Mass propertiesa tube fin set whose radius OpenRocket works out from the body, on 19 more probeshas OpenRocket’s radius and wall within 1e-15 and every part’s mass within 1e-14 but the one nose cone among those probes, within 5e-5 (the format guide)
Mass propertiesOpenRocket 24.12’s mass and center of mass of the parts in its parts catalog that the builder makes (all 3,449 but four it refuses) (test, The builder)tubes, rings, bulkheads, lugs, parachutes and streamers within 1e-14 of the mass; nose cones and transitions within 1e-3 of the mass, and their center of mass within 1e-3 of the part’s length (largest 6.3e-4 and 9.7e-4); but four blunt hollow nose cones, up to 4.8e-3 heavier here and 1.7e-3 of their length apart in center, where the two codes define a wall differently; masses stated in ounces, 8.8 parts in 10 billion apart (OpenRocket’s rounded ounce); and one streamer, whose stated mass OpenRocket ignores. A hollow part’s shoulder weighs nothing in OpenRocket and has the part’s wall here, and is taken out before comparing
Solid motorsThrustCurve.org’s own statistics code (total impulse, burn time, average and peak thrust), on the 412 files that pass the 1% check; the 1,289 that fail have only that check (the whole catalog)within 3e-15 relative, reading two or more samples at the same time as a step. ThrustCurve.org’s code averages such samples, and on 6 files that moves total impulse, burn time, average or peak thrust, by up to 6.5% (a peak thrust) (validation)
Solid motorsOpenRocket 24.12’s own reading of the same 32 curve files (total impulse, peak thrust, the 5%-of-peak burn-time window and the curve’s duration)every one bit for bit equal. Its average thrust divides the window’s own impulse by the window where HPR Sim divides the whole curve’s, so HPR Sim’s is +0.0107% to +0.3147% higher (median +0.0965%). Nothing else in the motor model is compared with OpenRocket
Solid motorsRocketPy’s solid-motor model, on three bundled motors, at 203 times eachtotal mass and inertias within 7.9e-5 relative; the propellant’s own mass and inertias within 1e-4 of their values at ignition
AerodynamicsBarrowman’s five worked examples, at Mach 0 (low speed): each rocket’s normal-force slope and center of pressureevery center of pressure within 1%. Every slope within 1% too, except the six-fin Recruiter’s: +2.87% (+3.42% on its fins alone)
Aerodynamicsdrag curves labeled RASAero in RocketPy’s examples, at Mach 0.3, with the fins and surface finish guessed because the curves don’t record themwithin 10% in two of seven cases; two boattailed ones −10.3% and −11.5%; −18.3% for Cavour power-on (motor burning), cause open
AerodynamicsValetudo’s drag table, which is 1.44 times the drag in the OpenRocket export for the same rocket−47.0% power-off and −50.4% power-on. Against the OpenRocket export, HPR Sim is 23.5% under as designed here, and 1.9% under with the export’s own surface finish and launch lugs
Aerodynamicsthe same curves every 0.05 from Mach 0.1 to 2.0, as far as each reaches; Calisto’s, the one real RASAero II export, to Mach 2Calisto within 10% at 15 of 15 subsonic Mach numbers, 5 of 7 transonic and 8 of 17 supersonic, where HPR Sim reads −14.9% to −5.1%, lowest at Mach 2 (−29.8% to −24.4% before the boattail’s supersonic wave drag) (evidence row G26). Other plausible fins put 14 to 17 of the 17 within 10%, though none puts every row within it, so most of what is left is within the unrecorded inputs; part of it is HPR Sim’s body, which reads low faster than sound against MIL-HDBK-762 too
AerodynamicsMIL-HDBK-762’s worked drag example, a rocket whose every term the handbook calculates, from Mach 0.5 to 3.2: a calculation with every input known, not a measurement. Its fins are sharp-edged wedges, which HPR Sim can’t represent, so their pressure drag is left out on both sides6 of 12 within 10%. From Mach 0.9 to 1.2, +12.3% to +31.9%, mostly the nose and the base; from Mach 1.6, −6.0% to −9.6%, friction and the base
Rigid-body flightthe exact motion of a tumbling, spinning rocket in a vacuum, over 22 sthe center of mass within 1.7e-6 m of the exact parabola
Rigid-body flightNASA’s check case 2, a brick of three unequal moments of inertia tumbling for 30 s with no aerodynamics, against four codes’ published histories, a fifth left out (a free body against them, report)the inertial body rates differ by up to 0.01077, 0.01633 and 0.005500 deg/s in roll, pitch and yaw, past the codes’ spreads of 0.0008991, 0.001242 and 0.0003532 deg/s. HPR Sim’s rotational equations leave the Earth’s rate out (#1521); that accounts for all but 0.0008818, 0.001005 and 0.0001268 deg/s at each code’s largest difference, each inside the spread. Read at every sample instead, what is left is past the spread by up to 0.0000001701 deg/s
Rigid-body flightNASA’s check case 3, case 2’s brick with a constant damping coefficient on each axis, applied by the test, against four codes’ published histories, a fifth left out (a free body against them, report)the inertial body rates differ by up to 0.004873, 0.005633 and 0.005102 deg/s in roll, pitch and yaw, past the codes’ spreads of 0.003609, 0.003792 and 0.003503 deg/s. The Earth’s rate left out of the rotational equations, with the same damping on both sets, accounts for all but 0.003176, 0.001123 and 0.0007793 deg/s, each inside the spread; at every sample roll and yaw stay inside and pitch comes to the spread
AerodynamicsNASA’s wind-tunnel tests of the half-scale Arcas Robin and a longer version, Mach 0.6 to 4.63: normal-force slope and center of pressure, 22 readings at 12 Mach numbers (fixture)from Mach 1.5, every slope within target: +8.8% to −3.3% (short) and +9.4% to −2.3% (long), since the committed designs fly the shock-expansion method to their base; the center of pressure within 0.53 calibers, which misses the half-caliber target on the long model at Mach 1.8 and 2.3, where that model’s body alone reads 15% to 19% high (fins off, the short model’s reads as much as 38% high at Mach 1.5: the crossflow excess sized at every speed); from Mach 0.8 to 1.2, 2 of 9 within 15% and half a caliber
AerodynamicsRASAero II’s normal-force slope and center of pressure for Calisto, Mach 0.1 to 2.0 (fixture)within 15% and half a caliber at 11 of 15 Mach numbers; HPR Sim’s slope rises with Mach through subsonic flow where RASAero II’s stays flat (+21.9% at Mach 0.9), and from Mach 1.5, where its von Kármán nose flies the shock-expansion method behind a Newtonian cap, reads +8.3% to +12.9%
Aerodynamicsthe body faster than sound, by a method no flight uses yet: the tables of NACA TN 3527, its source, for 144 cone- and ogive-cylinders at Mach 3 to 6.28, both its authors’ own values and their wind-tunnel measurements of the normal-force slope and center of pressure (fixture)against the measurements, 117 of 120 slopes within ±0.2 per radian (−0.278 to +0.251) and 109 of 120 centers of pressure within 0.2 calibers (−0.540 to +0.328); against the authors’ values, 102 of 144 slopes within 0.05 (−0.134 to +0.146) and 125 of 144 centers of pressure within 0.1 calibers (−0.670 to +0.257), so the targets are not met, where a separate implementation of the same equations, an uncommitted script, agrees with HPR Sim outside the 12 rows at the method’s limit (#81)
Aerodynamicsthe same method on the Arcas Robin’s nose and cylinder, against its measured body alone, Mach 1.5 to 4.63, 11 readings; no target, since the measurement includes the boattail and crossflow (fixture)the short model within 5% from Mach 1.8 to 2.96, +16.4% at 1.5, −15.0% and −18.7% at 3.96 and 4.63; the long model −13.7% to −26.4%
Aerodynamicsa blunt or vertical nose tip’s Newtonian cap ahead of the same method: NASA TN D-4865’s sphere-cone (a nose radius of 0.175 diameters on an 11.5° cone), its measured normal-force slope and center of pressure at Mach 1.50 to 4.63, 6 readings; and the Arcas Robin’s committed power-series nose on its cylinder and boattail, the lip left off, against its measured body alone, 11 readings; no target (fixture)the sphere-cone like for like (at its plotted angles, body lift included) −1.2% to +32.1%, high from Mach 2.96, where the report’s own method reads −3.3% to +13.6% overall, and the center of pressure within 0.06 diameters; the Arcas Robin like for like −4.8% to +37.2%, below the fitted secant ogive’s +3.4% to +41.0% at every Mach number and nearer the tunnel at nine of its eleven rows. No measurement checks a tip that isn’t spherical, and a nose that is nearly a cone but for a vanishing tip carries an unmeasured bias (issue #101)
Aerodynamicsthe lip at the Arcas Robin’s base, a reflexed flare behind its boattail, which HPR Sim gives no normal force faster than sound: what the measured fins-off pitching moment says the lip’s share is, 11 readings at Mach 1.5 to 4.63; no target (fixture)one share fitted to every row is +0.021 ± 0.019 per radian, so slender-body theory’s 0.178 sits 8.4 standard errors above it; fitted to each model alone it is −0.016 ± 0.022 (short, whose six rows disagree among themselves, χ2 per degree of freedom 6.4) and +0.108 ± 0.034 (long, whose five agree, 0.9). The number blames the lip for every miss in the center of pressure, so it rejects slender-body theory’s 0.178 but cannot separate zero from Seiff’s embedded Newtonian bound (0.044 down to 0.014)
AerodynamicsNASA TN D-4865’s model 2, the one flared body in the sources whose normal force and pitching moment are printed: a blunt 2.75° cone with an 18.5° flare, Mach 1.50 to 4.63, six readings, three of them with its boundary layer separated; no target (fixture)like for like (at its plotted angles, body lift included) the normal-force slope −1.9% at Mach 1.90, +7.0% at 2.30, +13.4% at 2.96 and +51.5% and +50.4% at 3.95 and 4.63, with the center of pressure within 0.05 calibers through Mach 2.96 and 0.088 at 3.95.
Aerodynamicsthat flare’s separated boundary layer, against the unflared model 1 (fixture)The report’s shadowgraphs show that flare’s boundary layer separated from Mach 2.96 up, though the cost only shows in the two fastest rows: the unflared model 1 reads +29.7% and +32.1% there against +12.5% at 2.96, so the flare itself adds 21.7 and 18.3 percentage points at 3.95 and 4.63 and between −1.9 and +0.9 below.
Aerodynamicsa flared body below about Mach 1.5289there is no marched reading at all and a flared body falls back to slender-body theory, carried up over 0.3 Mach by the join: drawing the flare out to the steepest turn its shock holds lands past the steepest the march itself takes
Aerodynamicsthe Arcas Robin’s body alone, fins off, against the 15% target the milestone set before the work began, 11 readings at Mach 1.5 to 4.63, the same bodies as the row above, with the lip in place, which moves them under a point; the boattail cap does not reach these, whose boattail is 15° (fixture)not met on six rows: the short model +37.7% at Mach 1.5, +25.9% at 1.8, +16.9% at 2.3 and 2.96, and the long +19.4% at 1.8 and +15.5% at 2.3; at Mach 3.96 and 4.63 both are within 5%.
Aerodynamicswhere the miss sits, which can only be told so farthe measurement’s own fit trades its slope at α→0 against its curvature at a correlation of −0.96. At α→0 HPR Sim is within 1.5 standard errors on every row outside the target, and on five of the six most of the gap is in the curvature body lift adds (1.3 to 2.8 times the measured, and ×12.20 at Mach 1.8 where the tunnel’s curve barely bends); on the sixth, the short model at Mach 2.96, 77% of the gap is HPR Sim’s own α→0 slope
AerodynamicsNASA’s wind-tunnel tests of the same two models, Mach 0.6 to 4.63: drag on the forebody (the models’ bases sat on a sting), fins on and off, 44 readings (Aerodynamics)2 of 44 within 10%, and every reading high. With the fins off, +13.5% to +24.1% from Mach 1.5 and +12.0% to +54.1% below, most of it the models’ 15° boattail, which HPR Sim over-predicts in a thick boundary layer. With the fins on, from Mach 1.5, +39.4% to +154.0%, where HPR Sim’s fins’ drag stays near 0.30 and the measured falls to 0.046. HPR Sim’s base drag behind a plain cylinder is not measured by the tunnel and has been checked at no speed faster than Mach 0.3
Aerodynamicsthe Basic Finner’s normal-force slope and center of pressure at a small angle of attack, from fits of free flights in the range of Canada’s DREV, Mach 1.08 to 3.15, six Mach numbers; no target. It leaves out the fins’ wedge section, the spin of the models whose fins were canted, and the table’s two rows past Mach 3.5 (report)the slope low on every row, where the fits’ probable errors are 1% to 4%: 11.2% and 16.1% low from Mach 1.5 to 2.5, 24.9% low at both Mach numbers below 1.5, and 9.3% and 13.0% low from Mach 2.5 to 3.5. The center of pressure forward of the measured one on every row, which understates stability: 0.34 and 0.18 calibers from Mach 1.5 to 2.5, 0.07 and 0.03 below, 0.27 and 0.23 above. At each fit’s root-mean-square yaw, about 1°, the slope’s miss is 0.8 to 2.4 percentage points smaller. The cause is not found
AerodynamicsNASA’s wind-tunnel tests of the two Arcas Robin models, Mach 1.5 to 4.63: roll forcing, the rolling moment per degree of cant, 11 readings (roll fixture)from Mach 2.3, all 8 within 5.3%; at Mach 1.5 and 1.8, +14.3% to +47.8%
Aerodynamicsthe Basic Finner’s zero-lift drag and pitch damping, against the same fits of DREV’s free flights, Mach 1.08 to 3.15, six Mach numbers; no target. The fins’ wedge section, which the design can’t draw, is modeled from published figures, a band. It leaves out the fins’ bases beside the body’s, the canted models’ roll, and the drag’s growth with angle (report)the drag 16.5% and 20.6% low from Mach 1.5 to 2.5, 21.3% and 19.3% low from Mach 2.5 to 3.5, below every shot of its row and below DREV’s even with the fins’ bases at a vacuum; below Mach 1.5, 10.1% high at Mach 1.077 (band 9.5% low to 17.1% high) and 3.5% low at 1.293, inside the shots’ spread. The cause is not found. The pitch damping 24.7% and 26.2% low from Mach 1.5 to 2.5, 39.6% and 30.4% low from 2.5 to 3.5, and 16.8% low at Mach 1.293: less damping than measured, the side that does not flatter. At Mach 1.077 it is 25.4% high, the side that does
AerodynamicsThe Basic Finner’s roll damping measured in the US Naval Ordnance Laboratory’s (NOL’s) wind tunnel, Mach 1.53 to 2.76 (roll fixture); in the free-flight range of Canada’s DREV, Mach 1.83 to 3.15, from cached readings (DREV statistics); and Barrowman’s computed value at Mach 0.07+11.7% to +27.5% against NOL’s tunnel, most at Mach 1.53: HPR Sim overstates the damping. Against DREV, 8.8% high on average and 12.4% at most in the core band (Mach 1.83 and 2.38); 5.0% on average and 8.7% at most in the extended band (2.72 and 3.15). −2.0% against Barrowman’s computed value (a theory curve, not a measurement; it confirmed how HPR Sim reads his method). Barrowman’s Fig. 5-7 points, once scored here (HPR Sim 5.9% to 16.2% below them), sit 1.27 times BRL 799’s range data (Aerodynamics) and are no longer scored
Aerodynamicsconical boattails measured in six NACA and NASA reports, jet off, Mach 0.3 to 3.24: the boattail’s own pressure drag and the base pressure behind it (drag fixture)attached boattails of 3° to 10° from Mach 1.2: −21.9% to +28.3%, within 0.0123 (58 readings); from Mach 1.0 to 1.1, −18.2% to −5.4% (4), and −46.2% to +60.0% at points near Mach 1 their report calls questionable (27); through the rise from Mach 0.85 to 0.95, −77.5% to +7.6% (28); under Niskanen’s rule to Mach 0.8, −100% to −83.5% (58). 16° in a boundary layer a fifth of the diameter thick, +26.4% to +54.2% (9). Separated 30° and 45°, −2.8% to +6.6% (3, which set the separation angles).
Aerodynamicsthe base drag behind themwithin 0.0102 of the measured on the cylinder’s area, though behind small bases that is up to about 40% of the base’s own drag (8 of the 12 set the ratio below Mach 2.5)
AerodynamicsOpenRocket 24.12’s normal-force slope and center of pressure for tube fins, on 14 probe designs and its Tube fin rocket, at Mach 0.05 to 0.75 (the decision that measured it); a code-to-code comparison. Tube fins as a whole are unvalidated (evidence row G27): no measurement of a tube fin’s lift or center was foundnot within the quarter caliber that the tube-fin lesson asks: HPR Sim’s center of pressure is 0.42 to 3.0 calibers forward of OpenRocket’s on every probe up to Mach 0.5, and 1.07 on the Tube fin rocket. From Mach 0.6, where OpenRocket moves the tubes’ center to their leading edge, 5 of 28 are within it. OpenRocket’s tubes lift 1.26 to 1.86 times HPR Sim’s, and up to Mach 0.5 it puts their center at a quarter of their length, ahead of which HPR Sim’s lies. All 70 probe gaps, and the Tube fin rocket’s, are pinned by a test
Rigid-body flightRocketPy’s whole flights from the pad to the ground, for six rockets, one past Mach 1, both codes flying one declared drag coefficientheights, speeds, times and accelerations within 3% (below), the largest +1.783% in the report; the path too, except the drifts of Juno III, Bella Lui and Prometheus 2022 in wind and NDRT 2020’s apogee drift, reported, not scored, as measured differences between the models (the decision)
Rigid-body flightOpenRocket 24.12’s calm flights of the 54 configurations of its examples that HPR Sim flies, each figure by OpenRocket’s own definition: apogee, largest speed, and the stability margin at rod clearance (report); a code-to-code comparison, with no target set. OpenRocket aborts one, so the figures are over the other 53, and that one is compared at two points up to the abort (flights OpenRocket aborted)margin within 0.016 calibers on 41 of 53, taken with the air along the axis as OpenRocket’s is; the Tube fin rocket’s is 1.08 calibers below OpenRocket’s (tube fins), the three-stage example’s three are 0.039 to 0.058 below, cause not yet sized (#185), and the five of Pods--airframes and winglets are 0.071 to 0.076 above, the flattering side (#325 and #326, two open causes in the fins), and the three of Pods--powered with recovery deployment are 0.070 above, cause not yet traced.
Rigid-body flightthe same flights, apogee and largest speedWith no named cause, apogee −4.34% to +1.95% (36 flights), and largest speed −0.69% to +6.10% (51 flights). The two-stage, three-stage, cluster, air-start and parallel-booster examples are within 5% in apogee and largest speed on every flight, the bar the staging milestone set before measuring; three cluster apogees and two three-stage ones are compared with OpenRocket’s flight with no parachute, since its parachute opened before apogee.
Rigid-body flightthe six apogees one named cause moves by more than 5%OpenRocket’s parachute opening while the rocket still climbs (HPR Sim’s flight compared here flies none), up to +13.80%. OpenRocket flying those six again without it brings four within 5% (sizing the causes).
Rigid-body flightthe other two of those six, Base drag hack (short-wide) on a D12-3 and an E12-4whose part set to no drag now flies as OpenRocket flies it (a part’s drag override): against OpenRocket’s flight with nothing deployed they read +5.84% and +8.92%, and within 0.10% when HPR Sim flies OpenRocket’s own drag, so their gap is the drag coefficient, by OpenRocket’s breakdown the very blunt nose’s. With recovery in both, the basis the E12-4’s milestone judged, the D12-3 is within 5% (+4.52%) and only the E12-4 is over (+7.80%). HPR Sim keeps a value for that nose read between Hoerner’s measured round heads, so the gap stays (the decision, #177).
Rigid-body flighta seventh, the Tube fin rocketreads +6.95%, and within 0.03% when HPR Sim flies OpenRocket’s own drag, so the gap is HPR Sim’s tube-fin drag, which probably reads low (#228).
Rigid-body flightthe flight OpenRocket aborts, Pods--powered with recovery deployment’s first configurationwithin 5% at both points (height −1.17% and −1.18%, speed −0.04% and +4.02%), weaker evidence than an apogee (its milestone)
Rigid-body flightOpenRocket 24.12’s calm flights of the 35 configurations of 11 private designs that HPR Sim flies, published only as differences under anonymized ids (report); a code-to-code comparison, with no target setcovers 11 of the 12 private designs. Apogee −4.84% to +13.60%. One is more than 5% off: the one supersonic flight, C06/1, +13.60% in apogee and +7.98% in largest speed, whose cause is the drag coefficient: flown on OpenRocket’s own drag, HPR Sim reads +1.11% and +1.04%; the supersonic pressure drag (mostly wave drag) and HPR Sim’s base drag under power, checked against a measurement only below Mach 0.8, are open (#222: HPR Sim’s supersonic pressure drag is about twice OpenRocket’s) (evidence row G14). The rest: largest speed −0.65% to +2.28%.
Rigid-body flightthe same flights’ stability margin (report)−0.0166 to +0.1108 calibers. HPR Sim calls four designs clearly more stable than OpenRocket does: C09 by 0.0350 to 0.0478 and C03 by 0.0564 to 0.0730 on every flight, cause not yet traced (#172); C08, a two-stage design, by 0.1108, a center-of-mass gap with a lead but no measured cause (#186); and C06/1 by +0.0604, most of it a center of mass 0.0447 calibers forward of OpenRocket’s, with a lead but no measured cause: its airfoil fins, which OpenRocket weighs by a factor of 0.85 (the mass survey’s fin-section cause, sized on the structure alone).
Rigid-body flighta fifth design, C12, launched from a tilted rod: its margin0.0125 to 0.0366 calibers, most of it OpenRocket’s rocket clearing the rod at an angle of attack: at that angle HPR Sim reads 0.0033 to 0.0074 (a tilted launch rod)
Time integrationa separate line-by-line transcription of DOPRI5, the published Fortran integrator by Hairer and Wanner that HPR Sim’s Dormand-Prince stepper follows, on the problem Hairer’s own example program for DOPRI5 solves: the Arenstorf orbit, the closed, looping path of a small body pulled by two large ones that circle each otherthe same step counts
Time integrationa vertical flight with drag that has an exact solutionapogee, deployment and landing times within 1.5e-8 s
Time integrationNASA’s check case 2 again, the model difference taken off, at the default tolerances and at 1e-12 (a free body against them, report)the body rates sit 0.0000005866, 0.0000009229 and 0.0000002790 deg/s from the code nearest them at the default tolerances (SIM 1 or SIM 4, which agree with each other; the other two kept codes are up to 0.00004975 and 0.001242 deg/s away), and 0.000000002126, 0.000000003051 and 0.000000001089 deg/s at 1e-12: the first three are step error, read from HPR Sim’s own two flights. Not a check of the steps alone
RecoveryRocketPy’s descents under a parachute, for five rocketsevery descent metric within 3% (below)
Recoverythe same descents: the heights where the later parachutes fire, and the descent rate under the droguewithin 0.17% and 0.01%
Recoverypublished drop tests of five small models falling with nothing deployed (tumbling)descent rate −10 to +19% off
RecoveryKidwell’s streamer drop tests (2001)descent rate +9% fast for his one flat streamer, and +58% fast for one folded into pleats, which HPR Sim doesn’t model
Interpolationthe exact formula of a smooth curve (a spline) through three points, y=3x/2−x3/2matched
Interpolationproperty tests, which check a rule on many randomly generated tablesevery table passes exactly through its own points
Quadraturepolynomials, whose integrals are known exactlyexact up to degree 22
Quadraturesix test integrals with known answers, one of them infinite at an endwithin 1e-11 relative

Against RocketPy

The descents and whole flights compared with RocketPy, on the same drag and on each code’s own.

The descent under a parachute, against RocketPy

This is one of the two comparisons the validation harness runs. The other is whole flights. The harness is the program behind cargo xtask validate. It flies every validation case and writes the committed report.

Five of RocketPy’s example rockets are flown down in both codes, set up the same way:

  • the same starting state near apogee, with the first parachute opening at once;
  • the same drag areas, deployment triggers and wind;
  • the random noise RocketPy can add to each parachute switched off, so its runs repeat exactly;
  • RocketPy’s gravity formula and its way of interpolating the wind (by its east and north components), in place of HPR Sim’s defaults.

The committed validation report gives the six numbers below for each descent. All were scored, and all are within tolerance. The largest difference is +2.865%.

Each metric must agree within 3% of RocketPy’s value (landing speed 0.1%, the tighter gate; most others 0.3%, a tighter gate), with no absolute floor. Each case file argues why, for example NDRT’s. A floor would be a fixed allowance, in meters or seconds, that passes any smaller difference. The metrics:

  • descent_time_s: the time from the shared start to landing.
  • impact_speed_m_s: the vertical speed at landing. The wind adds to the speed over the ground.
  • mean_descent_rate_m_s: the start’s height over the descent time, so it repeats the descent time in another form.
  • drift_m: how far the rocket lands from where the descent started, not from the pad. drift_east_m and drift_north_m are its east and north parts. A drift to the south or west is negative.

A difference is HPR Sim’s value less RocketPy’s, over RocketPy’s. A positive one means HPR Sim’s value is larger in size, in the same direction. NDRT drifts south, and HPR Sim carries it further south.

Every result of the report, as HPR Sim’s difference from RocketPy:

casedescent time, % (descent_time_s)impact speed, % (impact_speed_m_s)mean descent rate, % (mean_descent_rate_m_s)drift, % (drift_m)east drift, % (drift_east_m)north drift, % (drift_north_m)
descent-calisto-tests-motor-at-minus-1.373+0.077%−0.029%−0.077%+0.075%+0.074%+0.080%
descent-valetudo−0.019%+0.004%+0.019%−0.889%−0.889%−1.766%
descent-ndrt-2020-nose-to-tail+0.705%+0.012%−0.700%+0.276%+0.214%+2.865%
descent-prometheus-2022-generic-motor+0.083%−0.030%−0.083%+0.083%+0.086%+0.082%
descent-juno-iii−0.018%−0.008%+0.018%−0.018%−0.018%−0.018%

Each case is named after the RocketPy example it flies. Three names carry more:

  • descent-calisto-tests-motor-at-minus-1.373 is Calisto as RocketPy’s own tests build it. They put the motor at −1.373 m (−4.50 ft) in RocketPy’s coordinates. RocketPy’s getting-started notebook puts it at −1.255 m (−4.12 ft, notes on RocketPy’s example rockets).
  • descent-ndrt-2020-nose-to-tail is the NDRT 2020 rocket, which RocketPy’s example measures from the nose toward the tail (notes on RocketPy’s example rockets).
  • descent-prometheus-2022-generic-motor is Prometheus 2022, whose motor RocketPy describes with its generic-motor model. That model treats the propellant as a solid cylinder (notes on RocketPy’s example rockets, Recovery).

The Valetudo case flies RocketPy’s own example, not the flight on Getting started. From the shared start, 800 m (2,625 ft) above the ground, it falls in still air under the example’s one drogue. With RocketPy’s drag area of 0.4537 m², it lands at 17.627 m/s (57.83 ft/s). Getting started flies the same airframe from the pad. That flight has its own drogue and a main parachute, in a 5 m/s (16 ft/s) wind (case file, Recovery).

In still air, Valetudo’s drift, 0.19 m (0.62 ft), comes only from the Earth’s rotation (the Coriolis acceleration). Its north part is 19 µm. So a small difference there is a large fraction (Recovery).

The largest gap, NDRT’s north drift, most likely comes from added mass. RocketPy counts the air a canopy drags along: 15.9 kg for NDRT’s main, against the rocket’s 20.8 kg. HPR Sim has no such term, so the two respond differently as the canopy opens (Recovery). That explanation fits the size of the gap, but no test has isolated it yet.

What this shows: the two codes agree on the physics of a descent. It says nothing about whether either matches a real parachute on a real day.

Whole flights against RocketPy

Six of RocketPy’s example rockets are flown from the pad to the ground in both codes. They’re Calisto, Valetudo, NDRT 2020, Juno III, Bella Lui and Prometheus 2022. They’re set up the same way (the decision on what’s compared):

  • One declared drag coefficient, a constant 0.5 (case file), on the same reference area.
  • The example’s launch rail, site, parachutes and motor. The thrust curve is flown as measured, with no correction for the site’s thinner air, as RocketPy’s examples fly it.
  • RocketPy’s gravity formula, standard atmosphere and frictionless rail, and a declared wind.
  • The random noise RocketPy can add to each parachute switched off.

This is same-drag mode. It checks the equations of motion, the motor and the air, not the drag. HPR Sim’s own drag is compared below.

In short: how high, how fast and how long agree. So does where the rocket goes, except for rockets that leave the rail slowly in a wind. One of the six flights passes Mach 1, and three fly again in calm air. In all nine, the heights, speeds, times and accelerations agree within the 3% of each case’s gate (case file). The largest difference is +1.783% (report).

Still air means an example flown with no wind (Valetudo’s). Calm air is a windy case flown again with its wind switched off. The apogee and landing points agree too, within 2.2%, in every flight without wind and for Calisto in wind (the wind-path decision).

Juno III and Bella Lui leave the rail slowly in the wind, at a steep angle to the airflow. There HPR Sim’s body lift, which RocketPy leaves out, moves their drifts. So do its later release from the rail and, for Juno III, its simpler fin model. Together they put the drifts 10.195% to 38.158% from RocketPy’s (report).

Prometheus 2022’s drifts differ by −7.292% and +4.505%, from body lift and the rail release. NDRT 2020’s apogee drift differs by −4.333%, mostly from the rail release. These seven drifts are reported, not scored. So the validation milestone’s landing offset is met except where the two codes’ models differ.

Each of fifteen numbers per flight must agree within 3% of RocketPy’s, with no absolute floor. Rail-exit speeds must agree within 0.05%, landing speeds within 0.1%, burnout and top speeds and six burnout heights within 0.2%, six apogees within 0.35%, rail-exit times within 0.5%, and six apogee times within 1% (the tighter gates, more, the speeds, the apogees). The six are the flights with no written model difference. No same-drag apogee may read below RocketPy’s by more than 0.001%, since an apogee too low could wrongly clear a waiver; all nine read high today. Each case file argues why, for example Juno III’s. A number left unscored has its reason in its case file. Two more numbers compare the whole trace, explained after the tables.

The numbers are measured as RocketPy defines them, with one exception. Each code’s solver advances the flight in time steps and keeps the state at each step’s end. RocketPy takes each maximum (top speed, top Mach, top acceleration) only at those step ends. HPR Sim also searches between its own step ends for the true peak. So its number doesn’t depend on where its solver happened to step.

That search can only raise a maximum. Against HPR Sim’s old step-end readings it raised them by at most 6.3e-5 of themselves (NDRT 2020’s top speed). How much RocketPy’s own step ends miss wasn’t measured. Either way it’s far inside the 3% gate (the decision that also sets how peaks are found in both modes).

The definitions:

  • apogee_agl_m and apogee_time_s: the highest point, and when.
  • flight_time_s: the time from ignition to landing.
  • max_speed_m_s and max_mach: the top speed over the ground, and the top Mach number.
  • rail_exit_speed_m_s and rail_exit_time_s: when the forward rail button reaches the top of the rail, and the speed then. HPR Sim’s own rail-exit event waits for the last button, so the comparison finds RocketPy’s instant instead.
  • burnout_altitude_agl_m and burnout_speed_m_s: at the end of the thrust curve.
  • impact_speed_m_s: the vertical speed at landing.
  • max_acceleration_power_on_m_s2: the largest acceleration while the motor burns.
  • max_acceleration_m_s2 and max_acceleration_time_s: the largest over the whole flight, and when. For NDRT 2020 that’s its main parachute opening, not a flight load (case file).
  • apogee_drift_m and landing_drift_m: how far from the pad, along the ground, the apogee and the landing point are.

Speeds and accelerations are those of the rocket’s center of dry mass, the point RocketPy’s flight follows. Heights are measured from where that point starts, as RocketPy’s are. HPR Sim’s own output follows the loaded rocket’s center of mass.

The nine flights are the six, and Juno III, Calisto and Bella Lui again in calm air. The validation report scores all but eleven of their numbers, and all the scored ones are within tolerance. The eleven are measured and reported but not scored, each for a reason in its case file (below). Every result of the report, as HPR Sim’s difference from RocketPy:

caseapogee, % (apogee_agl_m)apogee time, % (apogee_time_s)flight time, % (flight_time_s)top speed, % (max_speed_m_s)top Mach, % (max_mach)
flight-calisto-tests-motor-at-minus-1.373+0.052%+0.103%+0.123%+0.015%−0.120%
flight-valetudo+0.112%+0.238%+0.233%+0.035%−0.026%
flight-ndrt-2020-nose-to-tail+0.064%+0.157%+0.643%+0.031%+0.003%
flight-prometheus-2022-generic-motor+1.208%+0.652%+0.830%−0.006%−0.235%
flight-juno-iii+0.641%+0.376%+0.475%+0.051%−0.221%
flight-bella-lui+0.342%+0.201%+0.229%+0.016%−0.071%
flight-juno-iii-calm+0.085%+0.073%+0.066%+0.015%−0.123%
flight-calisto-tests-motor-at-minus-1.373-calm+0.045%+0.099%+0.118%+0.022%−0.105%
flight-bella-lui-calm+0.038%+0.020%−0.012%+0.020%−0.018%
caserail-exit speed, % (rail_exit_speed_m_s)rail-exit time, % (rail_exit_time_s)burnout height, % (burnout_altitude_agl_m)burnout speed, % (burnout_speed_m_s)impact speed, % (impact_speed_m_s)
flight-calisto-tests-motor-at-minus-1.373−0.010%−0.072%+0.019%+0.019%−0.020%
flight-valetudo−0.002%−0.100%+0.048%+0.042%+0.009%
flight-ndrt-2020-nose-to-tail−0.009%−0.085%−0.004%+0.043%+0.022%
flight-prometheus-2022-generic-motor−0.014%−0.033%+0.600%−0.007%−0.013%
flight-juno-iii−0.005%−0.142%+0.269%+0.056%−0.003%
flight-bella-lui−0.013%−0.029%+0.141%+0.018%+0.021%
flight-juno-iii-calm−0.002%−0.137%+0.042%+0.020%+0.001%
flight-calisto-tests-motor-at-minus-1.373-calm−0.001%−0.074%+0.023%+0.025%−0.003%
flight-bella-lui-calm−0.000%−0.032%+0.012%+0.024%+0.021%
casepeak acceleration, burning, % (max_acceleration_power_on_m_s2)peak acceleration, % (max_acceleration_m_s2)peak acceleration time, % (max_acceleration_time_s)apogee drift, % (apogee_drift_m)landing drift, % (landing_drift_m)
flight-calisto-tests-motor-at-minus-1.373+0.099%+0.099%−96.811%−0.902%+1.257%
flight-valetudo+0.248%+0.248%+0.006%−0.931%−1.811%
flight-ndrt-2020-nose-to-tail−0.020%+83.059%+0.197%−4.333%+1.627%
flight-prometheus-2022-generic-motor+0.003%+19.062%+1.094%−7.292%+4.505%
flight-juno-iii−0.197%−0.197%+0.001%−38.158%+36.678%
flight-bella-lui+1.783%+1.783%+0.001%−10.195%−21.686%
flight-juno-iii-calm−0.012%−0.012%+0.046%−1.750%−1.807%
flight-calisto-tests-motor-at-minus-1.373-calm+0.108%+0.108%−96.811%−0.246%−0.415%
flight-bella-lui-calm+1.778%+1.778%+0.001%−1.155%−1.791%

The last two numbers compare the whole trace, not one point of it. The series height RMS (series_height_rms_m) is the root mean square of HPR Sim’s height less RocketPy’s. To get it, square each difference, average the squares and take the square root. The series speed RMS (series_speed_rms_m_s) is the same for speed. Both follow the center of mass without propellant, the point RocketPy’s series records.

Both are taken at RocketPy’s 120 series times, from ignition until HPR Sim lands. Both codes’ clocks start at ignition on the rail, so no time shift is fitted. A fitted shift would hide a real difference in the burn or on the rail (case file). Every time counts the same, so the long descent weighs most. A difference during the burn shows in the burnout and top-speed numbers instead.

Exact agreement would give 0, so these two are in meters and meters per second, not percentages. Each is held to 3% of RocketPy’s apogee (for height) or top speed (for speed), or 1% and 0.5% (more). That’s the validation milestone’s 3% for one number, applied to the whole trace (case file).

All nine flights pass, each well inside its bound (the table below). The largest height RMS is Prometheus 2022’s, 35.4 m (116 ft), about a third of its 110.3 m (362 ft) bound. Its apogee is also the furthest off, +1.208%. Body lift accounts for that too. RocketPy flown with HPR Sim’s body lift and rail release reaches 3,723.8 m (12,217 ft), against HPR Sim’s 3,723.6 (case file). Juno III’s height RMS is 14.6 m (47.9 ft), about a fifth of its 78.4 m (257 ft). The other heights are under three tenths of theirs, and each speed RMS is under a third of its bound (report).

caseheight RMS, m (series_height_rms_m)height bound, mspeed RMS, m/s (series_speed_rms_m_s)speed bound, m/s
flight-calisto-tests-motor-at-minus-1.373+1.8426.1+0.05801.21
flight-valetudo+2.127.7+0.1720.56
flight-ndrt-2020-nose-to-tail+2.4312.1+0.1250.91
flight-prometheus-2022-generic-motor+35.4110.3+1.5510
flight-juno-iii+14.678.4+0.8106.7
flight-bella-lui+1.6615.9+0.2742.9
flight-juno-iii-calm+2.0026.2+0.06581.13
flight-calisto-tests-motor-at-minus-1.373-calm+1.8226.1+0.05151.21
flight-bella-lui-calm+0.08975.4+0.02270.49

What the two codes still do differently, and what it moves:

  • In wind: body lift, the rail release and Juno III’s fins. A rocket that leaves the rail slowly in a wind meets the airflow at a steep angle. Juno III leaves at 18 m/s (59 ft/s) in an 8.5 m/s (28 ft/s) wind, 26° off it (the wind-path decision). Three things differ there.

    • HPR Sim’s normal force includes body lift, which grows with the square of that angle. RocketPy’s doesn’t. Much of it acts ahead of the rocket’s center of mass, the nose’s above all. So it moves the center of pressure forward and weakens the moment that turns the rocket into the wind. HPR Sim turns into it less.
    • HPR Sim keeps the rocket guided until its last rail button leaves the rail. RocketPy frees it at the first.
    • Juno III’s example gives its fins an airfoil lift curve, which RocketPy uses and HPR Sim can’t model. RocketPy’s fin slope is 7.6% steeper than HPR Sim’s flat-plate one (the wind-path decision).

    Juno III’s apogee is 245.3 m (805 ft) from the pad in HPR Sim and 396.6 m (1,301 ft) in RocketPy (−38.158%). Adding HPR Sim’s choices to RocketPy one at a time moves RocketPy’s apogee drift (the wind-path decision, measured by wind_response.py). It’s 360.7 m (1,183 ft) with HPR Sim’s rail release and 286.5 m (940 ft) with its body lift too. With its fin slope as well, it’s 248.3 m (815 ft). Every windy drift lands within 1.3% of HPR Sim’s the same way. Bella Lui’s drifts are −10.195% and −21.686%. Prometheus 2022’s are −7.292% and +4.505%: within 0.1% of HPR Sim’s once RocketPy has its body lift and rail release (case file). NDRT 2020’s apogee drift is −4.333%, mostly its rail release. These seven are reported but not scored, as measured differences between the models. Every other drift is scored and passes. That’s Calisto’s in wind, Valetudo’s in still air, NDRT 2020’s landing and all six in calm air (report, case file).

  • RocketPy’s own equations, corrected. RocketPy 1.13.0 takes the turning moments during the burn about the wrong point. That point is as far in front of the rocket’s center of dry mass as the real center of mass is behind it. That makes its rockets too stable while the motor burns, so they turn into the wind too far. The fix is a pull request to RocketPy, now merged. It builds on an earlier one, merged too. No release carries either yet. RocketPy 1.13.0 as installed still has the error, and the comparison applies both fixes (the wind-path decision). Without them, Juno III’s apogee drift was 582.4 m (1,911 ft). HPR Sim’s drifts in wind were up to −60.8% short of RocketPy’s at apogee and +151% beyond it at landing (the question of issue #50).

  • On the rail, HPR Sim keeps the terms for the center of mass moving inside the body as the propellant burns. RocketPy’s rail equation leaves them out. At a sharp ignition spike, with thrust and mass the same to five digits, HPR Sim’s acceleration is 1.2 to 1.3 m/s² higher. That’s Bella Lui’s +1.783%, whose peak is 7 ms after ignition (report, case file).

  • Calisto’s two peaks. Calisto’s acceleration peaks twice, 0.9% apart: on the rail at 0.05 s and at 1.568 s. The same rail terms make HPR Sim’s first peak the higher. So max_acceleration_time_s moves from one peak to the other (−96.811%), and it’s reported but not scored. The peak’s size is scored, but its +0.099% compares two instants. At 0.05 s HPR Sim is 1.05% higher (report, case file).

  • The main opening. RocketPy adds the air a canopy drags along (added mass), and HPR Sim has none. So NDRT’s peak deceleration as its main opens is +83.059% in HPR Sim, and Prometheus 2022’s is +19.062%. Both are reported but not scored. Their times are scored, since both codes put them where the main opens (report, case file).

Prometheus 2022 flies through Mach 1. Both flights peak just above Mach 1, HPR Sim’s 0.235% below RocketPy’s. Until the normal force through Mach 1, HPR Sim stopped any flight at Mach 1, and the case was a known gap. Now it flies on its drag table to the ground (Aerodynamics). Its scored numbers agree within 1.208% (report). This flight is a light test of the transonic normal force. No committed check measures its angle of attack there. A local probe found it below 0.11 degrees from Mach 0.8 to 1.2 (case file).

What this shows: where rockets leave the rail slowly in a wind, the models differ, and which code is nearer the truth is for real flights to say. The real flights below so far compare heights, not drift. This says nothing about HPR Sim’s own drag, which the next section compares, or about a real flight. The comparison with OpenRocket is on its own page (HPR Sim’s flights against OpenRocket’s).

Whole flights with each code’s own drag

The same six flights fly again, in predicted mode. HPR Sim flies its own drag, from each design’s shape, where same-drag mode gives it the declared one. Both modes use HPR Sim’s own normal force, so only the drag differs. Everything else is set up as in the same-drag flights above (the predicted-mode decision).

RocketPy flies the drag each of its examples ships, as RocketPy 1.13.0 flies the example. That’s a drag curve for Calisto, Valetudo and Juno III, and a function of Mach for Prometheus 2022 (case file). It’s a constant for NDRT 2020 (0.44, case file) and Bella Lui (0.43, case file).

In short: HPR Sim’s heights are within 3% of RocketPy’s for Calisto (+0.601%), Bella Lui (+1.278%) and Juno III (+2.951%). They’re well above for Valetudo (+10.113%) and NDRT 2020 (+10.302%), where HPR Sim’s drag is well below the example’s. They’re below for Prometheus 2022 (−7.280%), where its drag is above the example’s through the coast (report). These are results, not a pass or fail.

Neither code’s drag is the truth. Each example’s drag came from RASAero, OpenRocket or its team’s own estimate. So each number is compared with the same-drag flights’ 3% (the validation milestone), but only as a target. The rail-exit and landing speeds are compared with 0.1% and the rail-exit time with 0.5% (the tighter targets). A miss doesn’t fail the test suite. It’s reported and explained in its case file (Valetudo’s and NDRT 2020’s, for example). The report is committed, so any number that moves shows up in review.

Every predicted result of the report, as HPR Sim’s difference from RocketPy:

caseapogee, % (apogee_agl_m)apogee time, % (apogee_time_s)flight time, % (flight_time_s)top speed, % (max_speed_m_s)top Mach, % (max_mach)
predicted-calisto-tests-motor-at-minus-1.373+0.601%+0.209%+0.411%+0.500%+0.371%
predicted-valetudo+10.113%+6.179%+8.916%+2.292%+2.237%
predicted-ndrt-2020-nose-to-tail+10.302%+6.534%+6.870%+1.378%+1.351%
predicted-prometheus-2022-generic-motor−7.280%−4.920%−4.978%+1.280%+1.069%
predicted-juno-iii+2.951%+1.584%+2.424%+1.064%+0.795%
predicted-bella-lui+1.278%+0.732%+1.011%+0.273%+0.185%
caserail-exit speed, % (rail_exit_speed_m_s)rail-exit time, % (rail_exit_time_s)burnout height, % (burnout_altitude_agl_m)burnout speed, % (burnout_speed_m_s)impact speed, % (impact_speed_m_s)
predicted-calisto-tests-motor-at-minus-1.373−0.009%−0.071%+0.328%+0.729%−0.020%
predicted-valetudo+0.029%−0.109%+1.326%+2.883%+0.009%
predicted-ndrt-2020-nose-to-tail+0.000%−0.087%+0.723%+1.532%+0.022%
predicted-prometheus-2022-generic-motor+0.001%−0.036%+1.227%+1.778%−0.013%
predicted-juno-iii−0.001%−0.143%+0.799%+1.195%−0.003%
predicted-bella-lui−0.010%−0.029%+0.288%+0.336%+0.021%
casepeak acceleration, burning, % (max_acceleration_power_on_m_s2)peak acceleration, % (max_acceleration_m_s2)peak acceleration time, % (max_acceleration_time_s)apogee drift, % (apogee_drift_m)landing drift, % (landing_drift_m)
predicted-calisto-tests-motor-at-minus-1.373+0.190%+0.190%+0.002%−0.644%+1.597%
predicted-valetudo+0.250%+0.250%−0.030%+12.215%+10.437%
predicted-ndrt-2020-nose-to-tail+0.656%+83.059%+9.646%+12.259%+12.648%
predicted-prometheus-2022-generic-motor+0.848%+19.062%−6.903%−17.642%−6.167%
predicted-juno-iii+0.119%+0.119%+0.002%−35.049%+41.058%
predicted-bella-lui+1.797%+1.797%−0.029%−8.940%−20.255%

The whole-trace numbers, in meters and meters per second, are defined as for the same-drag flights. Valetudo’s, NDRT 2020’s and Prometheus 2022’s height RMS are outside the target. So is NDRT 2020’s speed RMS. They miss for the reason their apogees do: HPR Sim’s own drag differs from those examples’ drag. Each bound is 3% of that case’s own RocketPy apogee or top speed, so it differs from the same-drag bound. Juno III’s height RMS, 78.4 m (257 ft), is inside its 83.9 m (275 ft) here, at about nine tenths of it (report).

caseheight RMS, m (series_height_rms_m)height bound, mspeed RMS, m/s (series_speed_rms_m_s)speed bound, m/s
predicted-calisto-tests-motor-at-minus-1.373+12.384.6+0.3407.4
predicted-valetudo+75.420.9+2.793.2
predicted-ndrt-2020-nose-to-tail+11638.1+6.785.5
predicted-prometheus-2022-generic-motor+264128.8+7.2010.3
predicted-juno-iii+78.483.9+1.136.8
predicted-bella-lui+7.0716.2+0.3082.9

Why the misses, largest first:

  • Valetudo and NDRT 2020 fly high: the drag. HPR Sim’s drag coefficient at Mach 0.3 is −47.0% from Valetudo’s table. That table is hand-edited, at 1.44 times the drag of the OpenRocket export for the same rocket (Aerodynamics). For NDRT 2020 it’s 0.318 against the example’s constant 0.44 (case file). These drags are compared at Mach 0.3 only. Less drag also means a later apogee, a longer descent and further to drift. That moves their times and drifts too.
  • HPR Sim’s drag here is for the design as transcribed. Where RocketPy’s examples say nothing, the designs’ fin thickness and edges and their surface finish are placeholders. These results compare HPR Sim’s drag for those designs, not for the rockets as built. For Valetudo, the rocket’s own OpenRocket finish and launch lugs take HPR Sim’s drag coefficient from 0.5566 to 0.714 (Aerodynamics). So a miss here isn’t a gap for HPR Sim to close toward the example’s drag.
  • The drifts of Juno III and Bella Lui in wind: HPR Sim’s body lift and rail release, and Juno III’s fin slope. They act as in same-drag mode (the wind-path decision).
  • Prometheus 2022 flies low: the drag again, the other way. Since the drag through Mach 1, it passes Mach 1 on HPR Sim’s own drag. It peaks at Mach 1.060 against RocketPy’s 1.048. Its coasting drag rises toward Mach 0.8, where the example’s falls. So HPR Sim peaks −7.280% low and sooner, and its drifts and times follow (report, case file).
  • NDRT 2020’s and Prometheus 2022’s peak deceleration at their main openings, +83.059% and +19.062%, are the added-mass difference explained above. Their times move +9.646% and −6.903% with the apogee (report, case file).

What this shows: with its own drag, the larger gaps are the two drags differing, not the flight. It doesn’t say which drag is right. The real flights below start to, one rocket at a time.

Real flights

In short: HPR Sim has flown 12 public real flights and compared each with its altitude log. Read the way each log’s altimeter reads the air, HPR Sim’s apogees miss the logs’ by 8.88% on average. That’s outside the 5% target (evidence row G24). With its sign the mean is −4.64%, and 7 of the 12 read low (report).

Seven are RocketPy’s logged flights, flown in ERA5 weather and compared as soon as they were flown (the decision). Their mean is 6.06%, and +0.03% with its sign. Five of them miss by more than 5%. Each has an explanation checked against the report’s numbers: four are consistent with HPR Sim’s drag, and one with the motor’s impulse. The other five flights were predicted first. They miss by 12.82% on average, and no explanation is claimed for them (report).

Seven of the 12 altimeters’ kinds are assumed and none is calibrated (evidence rows G21 and G39). In the ten designs built from RocketPy’s examples, a fin’s edge shape is a placeholder where no airfoil file or published section gives it (the predictions’ note on what is traced). So read the numbers as those 12 flights’ (report), not as a bound on every rocket. The rest of this part is about the seven compared first. The five predicted first have their own part.

What is flown. RocketPy’s documentation flies more than a dozen rockets against their teams’ logs. HPR Sim flies ten of them the way a user would: the seven here, and three of the flights predicted first. The design is built from the example’s masses and shapes, with HPR Sim’s own aerodynamics. The thrust is the example’s own file, read as RocketPy reads it, to the same total impulse.

It flies from the example’s launch rail and site. The weather is the launch hour’s, from the example’s ERA5 file, a reanalysis of the atmosphere (ERA5 weather files). Genesis and Lince were added to the public designs for this.

What is left out, and why. The sample is the documented flights HPR Sim can fly today:

rocketswhy not yet
Camões, Halcyon, Hedythey fly their teams’ own motors, outside HPR Sim’s scope of commercial motors (decision)
Valetudo, Defiancetheir logs give an apogee, not a climb (decision)
Valkyrieits inputs are only in a data file (decision)

Astra, Andromeda and Erebus 11 are no longer left out: they are three of the five predicted first. Erebus 11 flies a COTS motor, a Cesaroni Pro54 K, most likely the 2437K660. Its thrust file gives 2,418 N·s over 3.657 s against that motor’s 2,437 N·s and 3.69 s (decision).

Juno III flies its team’s own motor too, but its design was already public. HPR Sim flies the thrust file as it would any other: nothing of the motor is modeled (decision).

The private designs add none. None is the rocket of a flight in the private collection of logs, so none has a log to compare with (the decision). A private flight collection added on October 4, 2026 pairs designs with their flights’ logs. Its 60 compared flights are below.

Reading the logs. Every log here comes from a barometric altimeter, or is assumed to (evidence rows G21 and G39). Such an altimeter turns pressure into the standard atmosphere’s altitude. On a day warmer than the standard it reads less than the height climbed, and more on a cold one. It reads 6.5% less on a day 20 K warmer (worked example).

Two logs record their pressure, and their heights are that reading, to 0.195 m (0.64 ft) and 1.321 m (4.33 ft) (report). So HPR Sim’s height is read the same way, from the pressure of the air it flew in at its center of mass. The table gives HPR Sim’s apogee both ways. The reading moves it from −7.9% (Juno III, in June at Spaceport America) to +1.7% (NDRT 2020, in February) (report, decision).

The kind of four of these seven altimeters is assumed rather than known (evidence row G21), and so is the kind of the three EuRoC 2022 ones among the flights predicted first (evidence row G39). Each row of the report gives its evidence. Reading all 12 logs as heights would give a mean of 7.90%, but five logs are known to be barometric. Reading only the seven assumed ones as heights gives 9.01%: the target is missed either way. Read that way, among the seven compared first, Lince, read as a height at 3,898.7 m (12,791 ft) against its log’s 3,587.7 m (11,771 ft), would miss by more than 5% with no checked explanation, and Genesis wouldn’t (report, decision).

Two logs also carry satellite (GNSS) heights, a geometric reference. Their barometric apogees are 0.943 and 0.935 of the satellite ones. HPR Sim’s conversion makes its own apogee 0.932 and 0.921 of its height (report). That’s the same direction and nearly the same size, 1 to 2 points lower on both, on the side of both flights’ misses. Juno III’s log is cut up to 20 m (66 ft) below its apogee, which would put its own ratio as high as 0.941 (decision).

Four of the seven logs are cut by hand just before a transient at their apogee, where the reading jumps. Juno III’s rises 62 m (203 ft) in 0.3 s as it levels off. The team’s reported apogee is that spike. At the cut its own velocity column still reads 17.5 m/s (57 ft/s) up. So its apogee may be 10 m (33 ft) to 20 m (66 ft) low (report). Cavour’s is read to its apogee event, just after its coast reached its peak of 2,785 m (9,137 ft). Two rows later its filtered height reads 2,789 m (9,150 ft), the team’s official apogee: a rise of 8 m (26 ft) in 0.2 s from a fall of 5 m/s (16 ft/s), which the rocket can’t fly. It comes about 2 s before the drogue charge most likely fires (decision).

What is compared.

  • The apogee: the log’s highest reading, against HPR Sim’s highest point read the same way. A plus means HPR Sim flies higher.
  • The climb: the root mean square (RMS) of HPR Sim’s height less the log’s over the ascent, the report’s trace RMS. A log’s clock starts when its altimeter says so, not at ignition. So both clocks are set to zero where each first reaches 30 m (98 ft). The RMS runs from there to the first of the two apogees (report). The descent isn’t compared: which parachute opened, and when, was the team’s.

The logs, thrust files and weather files are other people’s data, so they stay out of the repository. cargo xtask real-flights reads them from pinned copies of RocketPy and of the Macie XL repository (fetched with cargo xtask refs fetch rocketpy rocketry-files) and commits only the numbers (report).

flightlog apogee (m)HPR Sim apogee (m)apogee errorHPR Sim as a height (m)climb RMS (m)error on the team’s drag
Bella Lui, EPFL, 2020459.0464.0+1.10%464.73.7+0.24%
NDRT 2020, Notre Dame1,320.41,457.6+10.40%1,432.889.9+0.17%
Prometheus, Western Engineering, 2022 ¹3,895.83,549.3−8.90%3,809.9190.9+0.45%
Juno III, Projeto Jupiter, 20233,151.52,950.2−6.39%3,203.1219.2−9.05%
Cavour, Politecnico di Torino, 20232,785.02,946.0+5.78%3,052.1124.6−2.15%
Genesis, EuRoC 20232,916.72,746.0−5.85%2,875.395.2−0.08%
Lince, EuRoC 20233,587.73,732.5+4.04%3,898.769.9−12.20%

¹ Flown in the weather of the same day a year later, 2023, as RocketPy’s example flies it. No file of the flight’s day is available (report).

Over these seven the mean absolute apogee error is 6.06%, and over all 12 it is 8.88%, against the 5% target of the validation plan (gate and target). The largest climb RMS is Juno III’s, 6.95% of its apogee (report). No target is set on the climb. The census counts each climb against a bar of 3% of its apogee, the bound three RocketPy comparisons hold a height RMS to. Bella Lui and Lince are within it (census).

The last column is a diagnostic, not a prediction. It flies each flight again on the drag the example’s RocketPy notebook specifies, in place of HPR Sim’s. That drag is the team’s estimate, a table, a curve from RASAero II or a CFD analysis, or a constant given with no source. When the error falls inside 5%, the miss is consistent with HPR Sim’s drag. Neither drag is the truth. A team may have tuned its drag to this very flight. On the teams’ drag the seven’s mean absolute error is 3.48% (report).

A notebook can reshape its thrust file to a stated burn time and total impulse, as Juno III’s does. Where one does, a second diagnostic flies the file as recorded.

The five of the seven past 5%. Each explanation is a claim the report checks against its own numbers. A change to HPR Sim that makes one false fails cargo xtask real-flights --check, which runs only where the pinned copies are. CI checks each claim against the committed numbers (decision).

  • NDRT 2020 is +10.40%, consistent with HPR Sim’s drag. On the team’s constant drag coefficient, 0.44, the apogee is +0.17% (report). HPR Sim’s own drag is lower, as it was against RocketPy flying the same constant.
  • Prometheus is −8.90%, consistent with HPR Sim’s drag. On the team’s drag table the apogee is +0.45% (report). The barometric reading uses the temperatures of 24 June 2023, a year after the flight. Against the satellite heights, HPR Sim’s conversion reads 1 to 2 points below the altimeter, both here and on Juno III. Juno III flew in its own day’s weather, so the wrong day’s share of the miss can’t be told apart.
  • Juno III is −6.39%, consistent with the motor’s impulse. The notebook reshapes the team’s own motor curve to 8,800 N s, 4.9% less than the file. On the file as recorded (9,251.7 N s, its negative end read as zero) the apogee is +2.16%. On the team’s drag it’s −9.05% (report).
  • Cavour is +5.78%, consistent with HPR Sim’s drag. On the curves the team labels RASAero II the apogee is −2.15% (report). HPR Sim’s drag is below those curves, 8.3% at Mach 0.3 with the motor off and 18.3% with it burning (Aerodynamics).
  • Genesis is −5.85%, consistent with HPR Sim’s drag. On the team’s curves the apogee is −0.08% (report).

These are consistent explanations, not proofs. The teams’ drags, and the impulse a notebook sets, are estimates too. Lince, inside the target on HPR Sim’s drag, is −12.20% on its team’s. That isn’t investigated (report).

Predicted before compared

In short: five more public flights were predicted before they were compared, and are now compared. Three are RocketPy’s logged flights from EuRoC 2022: Astra, Andromeda and Erebus 11. Two are flights of Macie XL, an OpenRocket design in a public repository. Those two are flown as the simulations saved in that design file, in standard air: no record gives the day’s weather. The five predicted apogees were merged, that is, committed to the project’s public history, before any of them was compared with its log (the predictions; decisions: the three, the two). Against their logs they miss by −6.53%, +4.14%, −8.05%, −16.20% and −29.16%. That is 12.82% on average, and −11.16% with its sign (report, the decision). One prediction was revised after it merged, Erebus 11’s, by a change to the base drag; both numbers are kept.

The seven flights above were compared as soon as they were flown. A number published only after its comparison could have been tuned to its own miss, even without meaning to. These five went the other way round. Their inputs, the rule each log’s apogee is read by, and HPR Sim’s predictions went onto main first. The comparison came in a later change, which prints each prediction below beside its error (twelve public flights, predictions first).

flightimpulse flown (N·s)merged prediction (m)flown today (m)log apogee (m)apogee erroras a height climbed (m)every input traced
Astra (Cesaroni 4263L1350-P)4,280.23,037.13,037.13,249.5−6.53%3,155.4yes
Andromeda (its team’s thrust file)9,866.73,581.03,581.03,438.5+4.14%3,729.5no
Erebus 11 (a Cesaroni Pro54 K, the example’s thrust file)2,418.22,714.12,760.13,001.8−8.05%2,856.5yes
Macie XL, flight 1 (AeroTech I284W)599.5706.2706.2842.8−16.20%706.3no
Macie XL, flight 2 (Cesaroni 601I350-16A)600.8692.6692.6977.8−29.16%692.7no

The error is of the apogee flown today, which for four of the five is the merged prediction to the digit printed. Worked from the printed numbers, the errors are −6.54%, −16.21% and −29.17% for Astra and the two Macie XL flights, and the same for the other two. Erebus 11 flies 46.0 m (151 ft) higher than its merged prediction since ADR-362 relieved the base behind a boattail below Mach 0.8; its merged prediction missed by −9.58% (report).

What the comparison shows. Four of the five read low, and four miss by more than 5%. The report claims no explanation for them. The two diagnostics it flies among the seven, each team’s drag and Juno III’s thrust file as recorded, aren’t flown for these five, and their climbs aren’t compared (the open issue). What the report does give is each flight’s tags, written before the comparison.

Astra and Erebus 11 are the two tagged as having every input traced. Over those two the mean is 7.29%, and −7.29% with its sign. The seven compared first carry no tag, because they were compared before tags existed, so they are counted in neither tagged group. Over the 10 flights flown in ERA5 weather the mean is 6.12%, and −1.03% with its sign (report, the decision).

How each log is read. By the rule merged with its prediction, which a test holds the reader to. The three EuRoC 2022 logs give the highest reading of their height column, with the log read whole. The two Macie XL logs give the apogee in the altimeter’s own header, never the highest sample. On flight 1 a spike of one sample lifts the highest sample to 1,052.8 m (3,454 ft), which is 210.0 m (689 ft) above the header’s apogee (report). Each log must be the file whose digest was committed with its prediction.

A prediction that changes. A merged prediction is never edited. If a later change to HPR Sim or to an input moves one, cargo xtask real-flights fails until the code lists a revision for it: the number flown now, and the reason. The report then prints both numbers and both errors, and gives the means over the merged predictions too. One has been: Erebus 11, by ADR-362. On the predictions as first merged, the 12 flights miss by 9.01% on average, and by −4.76% with the sign (report).

The predicted apogee is what a barometric altimeter would read in the air the flight is flown in: the day’s for the three EuRoC 2022 flights, the stored simulation’s standard atmosphere for the two Macie XL flights. Each log is taken to be a barometric altimeter’s. For the three EuRoC 2022 flights that is assumed: no notebook says how its altimeter makes a height (evidence row G39). Astra’s reading is 3,037.1 m (9,964 ft), against a predicted climb of 3,155.4 m (10,352 ft) (the predictions). The reading is the lower of the two because an altimeter turns pressure into height with the standard atmosphere, and the air over the site that day was warmer than that (pressure altitude, the weather’s record). The comparison uses the reading, as it does for the seven.

Traced means the flight’s own record, here RocketPy’s example notebook, states every input that moves the apogee, bar what no RocketPy example records: the surface finish, the rail buttons’ size, the fins’ thickness where no airfoil file gives it, and their edge shape where no airfoil file or published section does. Those are placeholders in the designs built from the examples, the seven’s too, and they move the drag (the seven’s fin edges are evidence row G21’s guesses). So a traced flight’s inputs are stated, not exact.

How the three EuRoC 2022 flights are flown. By the seven’s method: the design built from the example’s masses and shapes, on the example’s own thrust file, from its rail and site, in the launch hour’s ERA5 air. The three share one weather file in a newer format. A committed script converts it and checks that the four fields’ values survived bit for bit (ERA5 weather files).

Andromeda is tagged not traced. Two things that move its apogee aren’t settled by its record. It flew air brakes, and nothing public says whether, when or how far they opened. If they did, a flight with none, like this one, reads high by an unknown amount. Its thrust file also holds 9,866.7 N·s, more than the 8,428.7 N·s on the bundled certified curve of the Cesaroni M2020 it is named for. On that curve the same design, at the same mass, is predicted at 3,050.3 m (10,008 ft). That figure is shown for scale and enters no average of errors. Against the log it reads −11.29% (report). The counted prediction is the team’s file’s, as every RocketPy flight here flies its example’s own thrust file. It is also the one prediction of the five that passes Mach 1, at Mach 1.019 (the predictions, Andromeda’s thrust curve).

The two Macie XL flights have no example notebook. Their record is a public repository (MIT): one OpenRocket design file, a page about the rocket and each flight’s logs. HPR Sim reads the file whole, as OpenRocket 24.12 saved it. Its stage’s mass override adds to its parts (a stage’s added mass). The file holds a stored simulation named for each flight: an OpenRocket run saved inside the design file, with its motor and its launch conditions. That is what is flown: its motor, from its vertical 1 m (3.3 ft) rod, in its steady 2 m/s (6.6 ft/s) wind and the standard atmosphere, at sea level (the Macie XL predictions). Four things are left out: the stored wind’s turbulence, the parachutes (the file opens one at apogee and the other lower, on the way down, so neither moves the apogee), the day’s weather and each pad’s height above sea level.

ThrustCurve.org can hold several data files for one motor, and HPR Sim bundles them. Each motor’s thrust comes from the bundled file that the design names by its digest, the file holding the curve OpenRocket flies. For the I284W that is not the first file of the bundle, whose thrust is as far from OpenRocket’s curve as 0.372989 of that curve’s peak thrust (the decision). The I284W’s masses stay the bundle’s for the motor, as every bundled motor’s do: 313.6 g of propellant and 555.52 g loaded, where the digest’s file and OpenRocket give 310.016 g and 555.072 g. Both logs are a PerfectFlite altimeter’s, which is barometric (.pf2 files). At these apogees, in the standard atmosphere, the reading and the climb agree within 0.1 m (0.33 ft).

Both are tagged not traced (the decision). For these two the record is the repository: its design file, its page and its logs. It gives no date, weather or rail. Of the launch site it gives only each log’s pad height above sea level, 3.7 m (12 ft) for flight 1 and 11.3 m (37 ft) for flight 2. The stored simulation’s conditions are flown, its sea level among them; all nine simulations in the file carry the same ones. Nothing says the rocket was weighed as flown. And HPR Sim’s .ork reader warns on both curve files. The I284W file’s impulse is 1.28% from ThrustCurve.org’s figure for the motor. The 601I350-16A file’s loaded mass is 616.3 g, against ThrustCurve.org’s record of 782.0 g. The file the design names is flown all the same, and each warning is kept with its prediction.

The design file also keeps the apogees OpenRocket’s own runs of the two simulations reached, 704.6 m (2,312 ft) and 691.6 m (2,269 ft) (the decision). They are another simulator’s numbers, not measurements, and are shown beside the predictions without being counted. Where the two programs agree, that is one program agreeing with another on the same file. It says nothing about how close either is to what flew.

Against the logs those two runs read −16.40% and −29.26%, beside HPR Sim’s −16.20% and −29.16% (report). So both programs miss these two flights by about as much, on a file whose record leaves the day’s air, the rail and the flown mass untraced. The report doesn’t say which input, if any, is behind the miss.

The choice to fly each stored simulation’s own conditions, and the tags, were written after HPR Sim had first flown the file at hpr sim’s defaults, and that run’s apogee had been seen (the decision). That order is the maintainer’s statement; no check can show it.

RocketPy and the Macie XL repository had published the measured apogees before the predictions were made. Each log was also looked at, for its columns and for a spike near its peak, before its rule was fixed. The code keeps the merged apogees as first merged, to the digit the page prints, and Erebus 11’s revision beside its own (report). A test that CI runs fails a committed report whose prediction has moved from them with no revision listed.

Flying them again needs RocketPy’s files and the Macie XL repository’s, which CI doesn’t have. After cargo xtask refs fetch rocketpy rocketry-files and the weather file’s conversion, cargo xtask public-predictions --check fails when a prediction flown now differs from the committed one. cargo xtask real-flights --check flies all 12, reads the five logs again and fails when the committed report doesn’t reproduce. The list of merged apogees is in the code, so a change to it shows in the code’s history.

Real flights of the private collection

These are a private collection’s logged apogees (the decision), rechecked against the collection’s catalog (the recheck). In short: HPR Sim and OpenRocket flew 60 real flights in the weather of their day. Both over-predict: HPR Sim’s apogees are +8.01% above the logs on average (mean absolute 12.09%), OpenRocket’s +6.72% (11.26%). The two codes agree within 5% on 59 of the 60 (report). Neither meets the 5% target.

Where the two codes differ from each other, it is mostly in drag: drag is the largest part of the gap on 61 of the 68 flights both fly (report; the gap between the two codes, by cause).

The report’s histogram gives the low side, the one a waiver depends on: its bins below zero hold the flights whose simulated apogee is below the logged one.

The gap to the logs is mostly not a difference between the two codes, and what causes it isn’t shown here. The flights’ owners allow only aggregate statistics, so no flight is named here.

Held out. A hash of each flight’s id holds out a third in expectation, here 23 of the 60 (decision). On them HPR Sim misses by 16.34% (mean absolute) and OpenRocket by 14.98%; on the other 37, 9.45% and 8.95% (report). Fits across flights, the scoreboard’s check included, leave the 23 out. All 60 were seen before the split, so neither half is blind. A flight added later is compared only once its prediction, made with no logged apogee, has its hash merged first.

What is flown. The collection pairs each design as flown with its own flight’s log, the motor, date, site and the ERA5 weather of the day (decision). Of its 89 best-documented flights (tier A, a first grading), 83 have an OpenRocket .ork design (report). Its catalog, a second reading of each flight’s sources made later, proposes a tier for each again (recheck):

  • A: the design, motor, log, day and site are each tied to the flight;
  • B: one of them is inferred, most often the motor;
  • C: the design isn’t shown to be the rocket flown.

HPR Sim flies the whole ERA5 profile of the launch hour: temperature, pressure and wind at every level. OpenRocket 24.12 can take only a temperature and pressure at the pad, with the standard atmosphere above. So it gets those, and the same wind every 50 m (164 ft) (report). Both fly OpenRocket’s curve for the motor the file names, from the same rail, with no parachute opened. A third row flies HPR Sim in OpenRocket’s air, which shows how much of the difference between the codes is the atmosphere.

What each log says. Every log here is a barometric altimeter’s. Such an altimeter turns pressure into the standard atmosphere’s altitude and subtracts the pad’s. So on a warm day it reads less than the height climbed (worked example).

The comparison turns each reading back into a height through the day’s ERA5 air. Across these flights the heights came out 1.0366 times the readings on average, from 0.9601 to 1.0777 times (report). Where a log’s highest value was an ejection charge’s pressure spike, the reading before the charge was taken. Of the 83 readings, 24 differ from the collection’s own figure by more than 0.5%, each with its reason recorded (decision).

over 60 flightsmeanmedianmean absoluteflights within 5%flights within 10%
HPR Sim+8.01%+7.60%12.09%1231
OpenRocket 24.12+6.72%+6.08%11.26%1331
HPR Sim in OpenRocket’s air+8.15%+7.60%12.19%1230
HPR Sim against OpenRocket+1.18%+0.95%1.73%5960

Each error is the simulated apogee less the logged one, over the logged one, in percent. So a positive error is a prediction above the log. The mean absolute error averages the errors without their signs.

The last row is HPR Sim’s apogee less OpenRocket’s, over OpenRocket’s. A positive value means HPR Sim flies higher. “Within 5%” and “within 10%” count the flights whose error is strictly smaller than that, either way (report).

The gap between the two codes, by cause

The last row above is over the 60 flights compared with a log. Both codes fly 68: those 60 and the 8 left out of the errors above, which the split can use because it needs no log. Over the 68, HPR Sim’s apogee is +0.98% above OpenRocket’s on average (mean absolute 1.68%, report). The split (decision) sizes how much of that gap each input carries. HPR Sim flies each flight again with OpenRocket’s own inputs in place of its own, one after another, each added to the ones before:

  1. Thrust: the thrust OpenRocket flew, against time.
  2. Mass: OpenRocket’s mass and center of mass, against time.
  3. Drag: OpenRocket’s drag coefficient along its own flight, against Mach number, one curve while the motor burns and one after.
  4. Air: OpenRocket’s air, the standard atmosphere through the pad’s temperature and pressure, in place of the day’s profile.

Each swap moves HPR Sim’s apogee, and that move is the swap’s part: the apogee before the swap less the apogee after it, over OpenRocket’s apogee. A positive part means the input as HPR Sim has it flies higher than OpenRocket’s does. So a positive drag part means HPR Sim’s drag is the lower of the two over that flight. Other is everything after the third swap, in two pieces. The air is the fourth swap’s part. The rest is what remains once all four inputs are OpenRocket’s: HPR Sim’s apogee on them less OpenRocket’s own. Thrust, mass, drag and other sum to the whole on every flight, and the run that writes the report stops on a flight where they don’t.

In the table, a part dominates a flight when its absolute value is the largest of the four. The last column counts the flights where a part’s absolute value is over 1% (report). A cell reading n/a is not counted. The rows below the four parts are pieces of other (the air and the rest) or checks on the parts (the whole, and drag swapped first), so none of them dominates a flight. Of those rows only the rest counts its flights over 1%, for the reading of the mass swap below.

over 68 flightsmeanmedianmean absoluteflights it dominatesflights where it is over 1%
thrust+0.00%+0.00%0.00%00
mass+0.10%+0.00%0.10%11
drag+0.85%+0.76%1.64%6139
other+0.03%+0.01%0.26%64
other: the air−0.15%−0.17%0.25%n/an/a
other: the rest+0.18%+0.19%0.33%n/a2
the whole+0.98%+0.85%1.68%n/an/a
drag, swapped first+0.85%+0.76%1.64%n/an/a

All 68 flights are split. None lacks an input in OpenRocket’s record, none is refused by HPR Sim, and none fails the check on the thrust and mass swaps (report, thrust and mass from another program).

What the rows say:

  • Drag is most of the gap. It is the largest part on 61 of the 68 flights, and its mean absolute value, 1.64%, is nearly the whole’s 1.68% (report). Its sign varies from flight to flight: its mean is +0.85%. Swapped first instead of third, its mean, median and mean absolute value are the drag row’s to the report’s two decimals. The report puts the drag swap in only these two places, first and third.
  • Thrust is nil. Both codes fly the same thrust curve for each motor. Swapping in the thrust OpenRocket made of it moves the apogee by 0.00% on average, to the report’s two decimals (report).
  • Mass is small on most flights. Its median is 0.00%, and it is over 1% on one flight, so its mean, +0.10%, isn’t typical of the flights (report).
  • The air is small. With thrust, mass and drag already OpenRocket’s, HPR Sim flies 0.15% lower on average in the day’s profile than in OpenRocket’s air (mean absolute 0.25%, report).
  • The rest holds what no swap replaces. With all four inputs OpenRocket’s, HPR Sim still flies 0.18% above OpenRocket on average (report). What stays HPR Sim’s through every swap is in it: the normal force, the center of pressure, the damping of pitching and yawing motion, the inertia, the steps in time, gravity, and what the drag table leaves out. None of those is measured apart here.

The mass swap and stability. After the mass swap HPR Sim flies OpenRocket’s center of mass against its own center of pressure. Where the two codes put the center of mass apart, that moves HPR Sim’s stability margin. What it does to the apogee is in the mass row, and no later swap undoes it: the flight on all four swaps keeps it, and the rest is measured from that flight to OpenRocket’s own. So the rest can take back what the mass row holds. On the one flight where the mass part is over 1%, the rest is over 1% too, with the opposite sign, which fits that reading and doesn’t prove it (report). The report counts such a flight by the two signs and the 1% floor alone, not by how much of the mass part the rest takes back. An open issue follows that flight (#1486): the lead is that the two codes weigh the rocket the same but put its center of mass apart.

As a worked example on a design anyone can run, take mach-probe, an invented test rocket that both codes fly faster than sound in still standard air. HPR Sim’s apogee is 3.06% above OpenRocket’s, and the drag part is 3.03%: over that flight HPR Sim’s drag is the lower. Thrust and mass are each under 0.005%, the rest is under 0.05%, and the air’s part is zero, since both fly the same air. On the slower pods-none the whole is 0.81% and the drag part 0.79%. A test holds each of those (the test, the_public_flights_split). From a clone, cargo test -p xtask the_public_flights_split runs it. It needs neither OpenRocket nor the private flights: OpenRocket’s record of both designs is committed.

The split has limits. The drag swap is a table in Mach number read from OpenRocket’s own flight, so it leaves out how OpenRocket’s drag changes with the Reynolds number and the angle of attack away from that flight’s path. The air’s part is taken with that table in place, so it holds the air’s density and speed of sound, not what the air does to friction. The mass swap keeps HPR Sim’s inertia and can change its stability, as above. And the split doesn’t say which part of the drag differs. One part is sized by itself, below: the pressure drag of tangent-ogive noses above Mach 1. And the drag part is split by Mach band.

Early deployments. On 19 of the 60, a parachute charge fired or the rocket separated before the highest reading, while it was still climbing (report). The readers’ notes put each at under 1.5 s before the peak and a few meters of climb. So the apogee barely moves (decision). Without them, over 41 flights, HPR Sim is +3.70% (mean absolute 9.46%) and OpenRocket +2.41% (8.81%). The two sets differ in which rockets they hold more than in what the early charges cost.

The recheck. The catalog flagged 68 of the 83 flights where its reading and this comparison’s differed. Each flag was settled against the flight’s log and sources (decision). Most changed nothing: a simulator’s miss, a spin or a recovery fault after apogee stays in the comparison. Four flights changed, and every flight gained its proposed tier:

  • one launch day moved to the event’s date;
  • one logged apogee dropped 0.23%, to the reading before a drogue charge (decision);
  • one flight left the errors because it tumbled through its climb, which the drawn design, at over 3 calibers of margin, can’t show;
  • one left because its posted design is the flier’s later, heavier rebuild.

Nothing in HPR Sim changed at the recheck; the comparison’s flights did. The recheck lowered HPR Sim’s mean error over the compared flights, and the tumbling flight explains more than all of the fall: its error was the largest in the set by a wide margin, and leaving it out lowered the mean by 1.28 points. Leaving out the rebuild raised it by 0.19 (decision). Later, HPR Sim came to read a fin tab as its file states it, even past the root’s ends (decision), and so flies seven more of the flights (#357), which gives the 60 compared above (report).

The catalog proposes 15 flights below tier A (decision). Over the 49 compared flights it keeps in tier A (report):

over 49 flightsmeanmedianmean absoluteflights within 5%flights within 10%
HPR Sim+7.30%+7.10%11.60%926
OpenRocket 24.12+6.09%+5.84%10.84%1026

What is not flown, and why. Of the 83, HPR Sim flies 68, and each cause of the rest has an open issue (report). OpenRocket flies 79, missing the motor in one (#362). For HPR Sim, most are small reading problems in its .ork importer that OpenRocket accepts:

  • a fin outline with a repeated point (3, #358);
  • rail buttons spaced forward (1, #359);
  • unread surface finishes (3, #360);
  • two override flags read as one (1, #361);
  • a part in an off-axis inner tube (2, #181).

Two fly motors whose curve OpenRocket’s database lacks (#362). In three more the collection itself falls short: two designs don’t hold the flown motor (#363) and one flight’s day has no weather file (#364).

The six tier-A flights in other formats wait until HPR Sim reads RockSim, RASAero and RocketPy files. The climb traces wait for their logs to be read.

Sources. The flights are published by university rocketry teams, rocketry courses and hobbyists on GitHub, team websites and The Rocketry Forum. The weather is ERA5, generated using Copernicus Climate Change Service information 2026, and contains modified Copernicus Climate Change Service information 2026. Neither the European Commission nor ECMWF is responsible for any use made of it. The report gives the full attribution and the ERA5 citations.

The tangent-ogive nose’s line

This part goes back to the gap between the two codes and sizes one difference inside its drag part: the pressure drag of a tangent ogive nose above Mach 1 (M2.3c13, the nose’s line, decision). Pressure drag is the drag from the air’s pressure on the nose, mostly wave drag past Mach 1. It is for anyone comparing the two codes on a rocket that flies faster than sound. Like the split, it is code against code, and here over 5 flights only. None of the 5 passes Mach 1.3, so the line sizes how the two noses differ between Mach 1 and 1.3 (report). The public probe covers that band too: there OpenRocket 24.12’s nose coefficient is below HPR Sim’s at each of the 31 Mach numbers from 1 to 1.3, in steps of 0.01, by 0.155 to 0.163 on the nose’s base area (the probe).

From Mach 1.3 up, OpenRocket 24.12 gives a tangent-ogive nose almost no pressure drag. HPR Sim gives it the drag of the cone with its length and base. On an invented rocket at Mach 1.5 the nose’s coefficient is 0.00073 in OpenRocket and 0.13028 in HPR Sim, both on the nose’s base area (the probe). Between Mach 1 and 1.3 each code joins its value at Mach 1 to its value at Mach 1.3. HPR Sim’s rises to 0.23428 at Mach 1.1, and OpenRocket’s to a hump of 0.07551 there, falling to 0.00098 at Mach 1.3 (the probe). A maintainer of OpenRocket confirmed the defect and merged a fix, which no release carried by November 14, 2026 (the probe’s section).

The line is taken inside the drag step. With thrust and mass already OpenRocket’s, HPR Sim flies once more with one thing changed: above Mach 1, the nose’s pressure drag is the one OpenRocket reports for that nose, at each Mach number. Everything else in the drag stays HPR Sim’s. On the probe, whose tube is as wide as the nose’s base, at Mach 1.5 that takes 0.13028 off the rocket’s drag coefficient and puts 0.00073 back (the probe’s table). The flight after that is the drag swap as before, on OpenRocket’s whole drag.

  • The nose’s line is the apogee before the nose’s swap less the apogee after it, over OpenRocket’s apogee. A negative line means HPR Sim climbs lower on its own nose drag than on OpenRocket’s.
  • The rest of the drag is the apogee after the nose’s swap less the apogee on OpenRocket’s whole drag.

The two sum to the flight’s drag part, and the run that writes the report stops on a flight where they don’t.

The line’s flights are those both codes fly whose own HPR Sim flight passes Mach 1 with a tangent-ogive nose: 5 of the 68. All 5 are swapped (report).

over the 5 flights swappedmeanmedianmean absolute
the nose’s pressure drag above Mach 1−1.28%−1.33%1.28%
the rest of the drag−0.43%+0.90%2.02%
the drag part−1.71%+0.51%2.40%

What the rows say:

  • The nose’s line is small, and negative. On its own nose drag HPR Sim climbs 1.28% lower on average than on OpenRocket’s. The mean absolute value is the same to two decimals, so the line is negative on every one of the 5, to that rounding (report).
  • These flights only just pass Mach 1. Their top Mach numbers, by HPR Sim’s own flights, average 1.06, and none passes Mach 1.3 (report). That is the band from Mach 1 to 1.3, where the probe’s runs from Mach 1 to 1.3 set the two noses side by side (the probe). So the line sizes how the two codes’ noses differ just past Mach 1, not the near-zero drag OpenRocket gives from Mach 1.3 up. It says nothing of a faster flight.
  • The rest of the drag is larger in mean absolute value, and of either sign. It is 2.02% against the nose’s 1.28%, with a mean of −0.43% and a median of +0.90% (report). In the means the nose is most of the drag part, −1.28% of −1.71%, because the rest’s positive and negative flights partly cancel.

A flight of the line that can’t be swapped is counted under a reason. The column here is OpenRocket’s pressure drag for the nose against Mach number, as recorded for that design. The reasons: more than one nose cone; a flight not split; no column in OpenRocket’s record; a nose whose drag the design states, itself or through a stage or parent that covers it; or a flight that HPR Sim fails with the column in or that leaves the Mach numbers the column covers. Each counts 0 flights (report). A test holds the rule for the line’s flights and each reason on public designs (the tests); from a clone, cargo test -p xtask nose_line runs them without OpenRocket or the private flights.

The flight this difference was first found on is not in the line. It is in the design library’s report, not this one, and that report holds no record of OpenRocket’s thrust and mass against time, which the line is taken on (#222, the supersonic drag gap to OpenRocket). The library’s report swaps the nose on it instead, on HPR Sim’s own thrust and mass. The swap alone moves the flight’s apogee from +13.60% to +31.73% of OpenRocket’s. With OpenRocket’s base rule under power as well, it comes to +0.45% (a supersonic flight).

The drag part by Mach band

This part takes the drag difference between HPR Sim and OpenRocket in the gap between the two codes and splits it by the speed the drag acts at, over the same 68 flights of the private collection (report; M1.14i1, the drag part by Mach band, decision). It is for anyone asking where the two codes’ drag differs, and it set what the project works on next. By mean absolute value, the largest share is below Mach 0.8. Like the split, the parts are code against code: they don’t say which code is nearer a real flight, and the flights are private, so no table by flight is shown. A second table sets the bands beside the logs.

The bands are in Mach number: below 0.8, from 0.8 to 1.2, and above 1.2 (report). The report’s data names them subsonic, transonic and supersonic (as data); the first stops at Mach 0.8, short of the speed of sound, where the second begins (transonic and supersonic). A Mach number at an edge is in the faster band (the tests, a_mach_number_s_band). With thrust and mass already OpenRocket’s, HPR Sim flies three more times. First OpenRocket’s drag coefficient replaces its own below Mach 0.8 only. Then it does below Mach 1.2. Then at every Mach number, which is the split’s drag swap. Outside the bands swapped the drag stays HPR Sim’s own.

  • A band’s part is the apogee before its swap less the apogee after it, over OpenRocket’s apogee. A positive part means HPR Sim’s drag in that band is the lower of the two over that flight.
  • A flight has a part in a band when that part is not zero: the swap changed the flight. A swap that changes no drag the flight reads leaves it the same to the last digit, so its part there is exactly zero. The flights with a part are the ones that reach the band, or come close enough that one of the solver’s time steps reads the drag inside it.
  • A band dominates a flight when its part’s absolute value is the largest of the three. The table’s last column counts the flights where that absolute value is over 1% (report).

The three parts sum to the flight’s drag part, and the run that writes the report stops on a flight where they don’t. All 68 flights split are split by band (report).

over 68 flightsflights with a part in itmeanmedianmean absolutemean absolute over those flightsflights it dominatesflights where it is over 1%
below Mach 0.868+0.39%+0.29%0.94%0.94%5024
Mach 0.8 to 1.228+0.24%+0.00%0.48%1.16%119
above Mach 1.27+0.23%+0.00%0.23%2.21%77
the drag part68+0.85%+0.76%1.64%

The mean, the median and the mean absolute value are over all 68 flights, zeros included, so the two faster bands’ medians are +0.00%: most flights have no part there. The next column is over the band’s own flights only (report).

What the rows say:

  • Below Mach 0.8 is the largest band. Its mean absolute part is 0.94% of the apogee, against 0.48% and 0.23%, and it dominates 50 of the 68 flights (report). Every flight has a part in it.
  • The faster bands are larger where they act. Over the flights with a part in them, the mean absolute part is 1.16% from Mach 0.8 to 1.2 and 2.21% above Mach 1.2 (report). They rank lower over the collection because their swaps change fewer flights: 28 and 7.
  • Below Mach 0.8 the sign varies from flight to flight. The mean is +0.39% against a mean absolute 0.94% (report).
  • Above Mach 1.2 HPR Sim’s drag is the lower. Over the 7 flights with a part in that band, its mean and its mean absolute value are both 2.21%, and each of the 7 is over 1%. So its part is positive on each of the 7 (report, as data).

Each swap changes the speed the next band is flown at, so a band’s part depends on the swaps before it. To check that, each band is also swapped alone, first: on HPR Sim’s own thrust and mass, with its own drag in the other two bands. Over the collection the two ways give sizes within 0.01 points of each other: alone, the three bands’ mean absolute parts are 0.94%, 0.48% and 0.24% (report).

Beside the logs. Four kinds of flight are set beside the logged apogees of the 60 flights compared: HPR Sim’s own, HPR Sim’s with one band swapped alone, HPR Sim’s on OpenRocket’s whole drag, and OpenRocket’s own. Each row is the apogee less the logged one, over the logged one. A row moves a flight when its apogee is not that of HPR Sim’s own flight. The last two columns count the moved flights that come nearer their log than HPR Sim’s own flight, and the ones that go farther (report). The first row is HPR Sim as it flies, and the last is OpenRocket’s own flight. The four between fly HPR Sim’s own thrust and mass, and none of them is a model HPR Sim could fly: the coefficient is read, by Mach number, along OpenRocket’s own flight of the same design.

HPR Sim’s drag, over 60 flightsflights movedmeanmedianmean absoluteflights nearer their logflights farther from it
its own+8.01%+7.60%12.09%
OpenRocket’s below Mach 0.8, its own above60+7.45%+6.77%11.68%3822
OpenRocket’s from Mach 0.8 to 1.2, its own elsewhere25+7.68%+7.46%11.81%169
OpenRocket’s above Mach 1.2, its own below7+7.72%+6.66%11.91%61
OpenRocket’s at every Mach number60+6.85%+6.33%11.27%3822
OpenRocket’s own flight, for comparison60+6.72%+6.08%11.26%4020
  • On OpenRocket’s whole drag, HPR Sim’s mean absolute error is OpenRocket’s to 0.01 points: 11.27% against 11.26%. Its mean stays above OpenRocket’s, +6.85% against +6.72%, and so does its median, +6.33% against +6.08% (report).
  • The band below Mach 0.8 moves the score the most. Its swap alone takes HPR Sim’s mean absolute error from 12.09% to 11.68%. The band from Mach 0.8 to 1.2 takes it to 11.81%, and the one above Mach 1.2 to 11.91%. Part of that is reach: the first swap moves all 60 flights, the others 25 and 7 (report).
  • Not every flight comes nearer. With OpenRocket’s drag below Mach 0.8, 38 flights come nearer their log and 22 go farther (report).

As a worked example on designs anyone can run, take the two public test rockets of the split. pods-none stays below Mach 0.8, so its drag part of 0.79% is all in the slowest band, and its parts in the other two are exactly zero. mach-probe passes Mach 1.2. Its three parts are −0.08%, +0.26% and +2.85%, which sum to its drag part of 3.03%: nearly all of that flight’s drag difference is above Mach 1.2. A test holds each of those (the tests, the_public_flights_split_by_band). From a clone, cargo test -p xtask drag_bands runs it, with neither OpenRocket nor the private flights.

What it decided. The band below Mach 0.8 was worked on next: M1.14i, drag below Mach 0.8 (the decision that orders the work). Over these flights it has the largest mean absolute part, 0.94% against 0.48% and 0.23%, and it dominates the most of them (report). The faster bands are larger where they act, but on 28 and 7 flights; they keep the steps they had. The steps the accuracy work takes by Mach band began at Mach 0.8, so this band needed one of its own. That step first splits the band’s part into friction, pressure and base drag. It then looks for a measurement of the largest of the three, from a wind tunnel or a flight, to check HPR Sim’s model against; if it finds none to add, the step ends there with the gap written down. It does not make HPR Sim’s drag OpenRocket’s.

The slowest band, by term

This part takes the band below Mach 0.8 and splits its share of the drag difference by the term of the drag it comes from: friction, pressure and base drag, over the same 68 flights (report; M1.14i2, the part below Mach 0.8 by term, decision). It is for anyone asking which of HPR Sim’s drag models to check first. The largest term is the base drag, four fifths of it while the motor burns; the pressure term, of the other sign, cancels part of it. Like the bands, the parts are code against code. A measured reference of that term is then the check (below).

The terms, set against each other. OpenRocket’s flight data splits its drag coefficient into friction, pressure and base. The two codes don’t file every part of the drag under the same name, so a small probe flies one public design four more times in OpenRocket, each time with one thing added, and records which of OpenRocket’s terms moves (record, written by drag_terms.py):

  • A launch lug or rail buttons raise OpenRocket’s pressure term by 0.012 to 0.018 and leave its base alone. HPR Sim files them as parasitic drag, so its pressure term here is its pressure plus its parasitic drag.
  • A drag coefficient stated for a part shows up in none of OpenRocket’s three terms: the coefficient less the three is the stated value, 0.3 on the body tube and three times 0.3 on a three-fin set (the probe). So the fourth term, the rest, is a stated coefficient in both codes.

Each code’s four terms sum to its whole coefficient, under power and coasting, and a test holds both sums and the probe’s record (the tests, each_codes_four_terms_sum_to_its_whole, openrocket_books_lugs_and_buttons_as_pressure_and_stated_drag_in_the_rest).

The split. With thrust and mass already OpenRocket’s, HPR Sim flies four more times. OpenRocket’s friction replaces its own below Mach 0.8, then its pressure too, then its base, then its rest. Above Mach 0.8 the drag stays HPR Sim’s own. A term’s part is the apogee before its swap less the apogee after it, over OpenRocket’s apogee. A positive part means HPR Sim’s term is the lower of the two over that flight. The last of the four flights is the band’s own flight below Mach 0.8, to the last digit, so the four parts sum to the band’s part, and the run that writes the report stops on a flight where they don’t. All 68 flights split by band are split by term (report).

over 68 flightsflights with a part in itmeanmedianmean absoluteflights it dominatesflights where it is over 1%
friction68−0.17%−0.16%0.17%00
pressure, launch lugs and rail buttons among it68−0.64%−0.11%0.66%1712
base68+1.20%+0.91%1.20%5133
the rest: stated drag coefficients0+0.00%+0.00%0.00%00
the part below Mach 0.868+0.39%+0.29%0.94%

A term dominates a flight when its part is the largest of the four in absolute value, and the last column counts the flights where its part is over 1% of the apogee (report).

What the rows say:

  • The base is the largest term. Its mean absolute part is 1.20% of the apogee, against 0.66% for pressure and 0.17% for friction, and it dominates 51 of the 68 flights (report). It is positive: HPR Sim’s base drag is the lower of the two.
  • Most of the base difference is the base under power. On a flat base, coasting, the two codes’ base drag agree: 0.13170 against 0.13170 at Mach 0.3 on the public test rocket, a test’s pinned numbers (the tests, both_codes_terms_are_pinned_at_mach_0_3_and_0_6). While a motor burns, HPR Sim takes the burning motor’s cross-section off the base, and OpenRocket 24.12 keeps the whole base. Behind a boattail below Mach 0.8, HPR Sim keeps dB/d1 of the flat-base value and OpenRocket all of it, since ADR-362 (below). Swapped alone, the base moves HPR Sim’s apogee by a mean absolute 1.17%; swapped only while a motor burns, by 0.94% (report).
  • Pressure and the base partly cancel. The pressure term is negative, HPR Sim’s the higher. Its mean, −0.64%, is far from its median, −0.11%, and it is over 1% on 12 flights. That is why the band’s mean, +0.39%, is smaller than its mean absolute part, 0.94% (report).
  • Friction is small. Its part never dominates a flight (report). On the public test rocket HPR Sim’s friction is the higher: 0.30690 against OpenRocket’s 0.30310 at Mach 0.3 (the tests, both_codes_terms_are_pinned_at_mach_0_3_and_0_6).
  • No flight states a drag coefficient in either code, so the rest is zero on every flight.

Beside the logs. As with the bands, each term is also swapped alone, on HPR Sim’s own thrust and mass, and set beside the logged apogees of the 60 flights compared. The columns are the bands’ (report).

HPR Sim’s drag, over 60 flightsflights movedmeanmedianmean absoluteflights nearer their logflights farther from it
its own+8.01%+7.60%12.09%
OpenRocket’s friction below Mach 0.860+8.19%+7.74%12.18%1545
OpenRocket’s pressure below Mach 0.860+8.63%+8.23%12.21%2337
OpenRocket’s base below Mach 0.860+6.67%+6.63%11.49%4515
OpenRocket’s rest below Mach 0.80+8.01%+7.60%12.09%00
OpenRocket’s whole drag below Mach 0.860+7.45%+6.77%11.68%3822

On OpenRocket’s base alone, HPR Sim’s mean absolute error falls from 12.09% to 11.49%, more than on OpenRocket’s whole drag below Mach 0.8, and 45 of the 60 flights come nearer their log (report). Its friction and its pressure alone each take HPR Sim farther from the logs. That doesn’t show OpenRocket’s base rule is the truer one: both codes climb above the logs on average, so more drag from any cause moves the mean toward them.

As a worked example anyone can run, take pods-none, which stays below Mach 0.8 and burns there. Its band’s part of +0.79% splits into −0.37% friction, +0.04% pressure, +1.12% base and 0% rest (the tests, the_public_flights_split_by_term; from a clone, cargo test -p xtask drag_terms).

The base under power, against a measured flight

This part sets HPR Sim’s base drag below Mach 0.8, coasting and with the motor burning, beside a measured flight (the rows are in the fixture report). It is for anyone weighing the term above against something other than another code. It stops at Mach 0.8 because that is the band the term split covers.

The reference. A 1955 NACA report gives free flights of rocket-propelled models of the Hermes A-3A from Wallops Island, with the sustainer motor burning and then coasting, and the pressure on the base measured during both (Jackson 1955, NACA RM SL55F15, a US government work in the public domain). Its figure 20 gives the base drag coefficient on the model’s largest cross-section, from Mach 0.6 to 1.8, with the motor burning and (from about Mach 0.65) coasting. The report believes the error of these faired curves to be less than ±0.005 (printed p. 10). The curves are read off the figure by pixels to ±0.001 more (the reading script, the readings and the model’s sizes). At Mach 0.7 the report’s own base pressures, in its figure 22, read by eye (a research note), give base drag 0.0042 above figure 20 under power and 0.0057 above it coasting, about the stated error. The figures don’t settle which is right (the readings’ notes).

The model is 8 in across, with a 4.5° cone narrowing to a 4.8 in base. Its motor, a nominal 5 in case, sits forward in the body, and a blast tube carries the jet to a 3.0 in nozzle exit at the base. The jet leaves below the outside air’s pressure, past fixed vanes in it (the model’s sizes).

HPR Sim’s rule and OpenRocket’s. Below Mach 0.8, HPR Sim’s base drag coefficient on the base is 0.12+0.13M2 (Drag), coasting on the whole base, and behind a boattail it keeps dB/d1 of that, the base’s diameter over the boattail’s fore diameter (the largest, where the boattail starts at it): a conical afterbody’s base drag scales with that ratio cubed on the largest cross-section where the rule on the base’s area scales with its square (Stoney, NASA TR R-100, 1961, p. 11, from tunnel data; ADR-362, which this reference decided). In a flight, with the motor burning, HPR Sim takes the motor case’s area off the base, down to zero. Here the case is wider than the base, so as HPR Sim flies it there is no base drag under power. Given only the nozzle exit’s area, which is what this base holds of the motor, the same rule leaves the ring of base around the exit, the area the report measured on: that tests the coefficient, not the area HPR Sim flies. OpenRocket 24.12 keeps the whole base under power, on the flat-base rule with no relief behind a boattail: 1/0.6 times the table’s last column, which is the whole base on HPR Sim’s coefficient. As a worked example at Mach 0.7: 0.12+0.13×0.49=0.1837 on the base. The base is (4.8/8)2=0.36 of the largest cross-section, and behind the cone it keeps 4.8/8=0.6 of the rule, so coasting HPR Sim gives 0.1837×0.36×0.6=0.0397. Under power the ring is (4.82−3.02)/82=0.219 of it, so given the nozzle exit HPR Sim gives 0.0242, and with the case off 0. The flight measured 0.0105 coasting and 0.0244 under power (fixture report). Before ADR-362 the base kept the whole rule behind a boattail, 1/0.6 times as much coasting.

Machmeasured base drag coefficient, power offHPR Sim’s base drag coefficient, power offmeasured base drag coefficient, under powerHPR Sim’s base drag coefficient, under power, case offHPR Sim’s base drag coefficient, under power, nozzle exit offthe whole base’s drag coefficient, under power, on HPR Sim’s coefficient
0.650.01270.03780.02700.00000.02300.0378
0.700.01050.03970.02440.00000.02420.0397
0.750.00960.04170.02210.00000.02540.0417
0.790.00860.04340.02040.00000.02650.0434

(fixture, written by cargo xtask aero; the same rows are in the fixture report)

What the rows say. Which way each rule errs, past the stated error and the reading, is held by a test (the tests, the_power_off_rule_reads_high_the_flown_power_on_rule_low, with two mutation probes in ADR-362):

  • Under power, HPR Sim’s rule as it flies reads low, by 0.0204 to 0.0270: it leaves no base drag where the flight measured some. Too little drag gives too high an apogee: the optimistic side for a flier aiming at a target altitude, not the side that flatters a flight near a waiver’s ceiling (which error flatters; fixture report).
  • Given the nozzle exit instead, the same rule comes within 0.007, from 0.0040 low to 0.0061 high: within the stated error and the reading together, ±0.006, at Mach 0.65 to 0.75. The whole base kept under power reads high by 0.0108 to 0.0230 on HPR Sim’s coefficient; on OpenRocket 24.12’s, the flat-base rule with no relief behind the boattail, by 0.036 to 0.052 (the fixture’s flat-base column less the measured; fixture report).
  • Coasting, HPR Sim reads high, by 0.025 to 0.035: its value is 3.0 to 5.1 times the measured. Before ADR-362 it read high by 0.050 to 0.064, 5.0 to 8.4 times: the mean absolute error over the four rows fell from 0.0574 to 0.0303, and no row’s grew. Hoerner’s rule, which also lowers the base for a thicker boundary layer, reads high by 0.012 to 0.016 (fixture report; the rules side by side).
  • The jet raised the measured base drag, to 2.1 to 2.4 times its coasting value, as the report says in its own words: “more than twice”. HPR Sim’s rule given the nozzle exit lowers it, to 0.61 of coasting, and the whole base keeps it. Neither moves the way the flight did. Kept on the measured coasting value instead, the whole base would read low, by 0.0118 to 0.0143 (fixture report).

What it leaves out: one vehicle and one motor, whose jet left through a blast tube, past fixed vanes, below the outside air’s pressure. How far the result carries to another motor and base is not measured. The base model was changed on this reference for a boattailed base, power off (M1.14i3, ADR-362), and what that moved is above (the slowest band, by term). Nothing here covers the jet (evidence row G14), and no hobby rocket’s base behind a boattail has been measured (evidence row G41).

Known gaps

These are the largest known differences and missing pieces. Each model page’s In short lists the rest.

  • HPR Sim’s own drag in a whole flight. Its heights differ from RocketPy’s by −7.280% to +10.302%, where its drag differs from the examples’ (report). Against the seven real flights compared first, four of the five apogees more than 5% from their logs are consistent with HPR Sim’s drag (real flights, real-flight report).
  • Four of the five public flights predicted first miss by more than 5%, and no explanation is claimed for them. The five miss by −29.16% to +4.14%, four of them low (predicted before compared, real-flight report).
  • Drag faster than sound reads high against NASA’s wind tunnel, above all with fins. With the fins on, it’s +39.4% at Mach 1.5 to +154.0% at 4.63. The fins take a blunt leading edge’s formula. Nothing models a thin, sharp fin’s own wave drag. Niskanen’s cone, which ogives share, reads 45% to 105% above a measured cone through the rise near Mach 1 (Aerodynamics).
  • A steep boattail’s drag reads high in a thick boundary layer. Boattails of 16° read +26.4% to +54.2% from Mach 1.0 to 1.28. The Arcas Robin’s 15° boattail puts its forebody, fins off, +13.5% to +24.1% high from Mach 1.5. From Mach 1.0 to 1.2 it’s +20.3% to +50.8%. No cited correction exists in the sources used (issue #72; Aerodynamics).
  • Drag past Mach 1.6 reads low against RASAero II’s Calisto, to −14.9% at Mach 2, with the fins and finish the Mach 0.3 check declares. Part of it is HPR Sim’s body, which reads 6% to 10% low faster than sound against MIL-HDBK-762’s worked example as well (Aerodynamics).
  • Drag and pitch damping read low on the Basic Finner faster than sound. Against its free flights from Mach 1.83 to 3.15, with its wedge fins’ drag modeled from published figures, HPR Sim’s drag reads 16.5% to 21.3% low and its pitch damping 24.7% to 39.6% low; at Mach 1.077 the damping reads 25.4% high, the side that flatters. The drag’s cause is not found (#1513). A square fin’s trailing edge may also take too little drag faster than sound (#1514; Aerodynamics).
  • Roll: the forcing high near Mach 1.5, the damping high, nothing committed below Mach 1.5. The roll forcing from canted fins reads +47.8% at Mach 1.5 and +14.3% to +17.8% at 1.8 against NASA’s wind tunnel. There, linear theory’s load climbs toward Mach 1 faster than the fins’ does. The roll damping reads 11.7% to 27.5% high against the Basic Finner’s in NOL’s wind tunnel, Mach 1.53 to 2.76, and against DREV’s free-flight range 8.8% high on average in the core band and 5.0% in the extended band. No committed comparison checks either below Mach 1.5, where uncommitted measured sets put the damping 24.5% to 200% high (Aerodynamics).
  • A flared rocket’s supersonic normal force rests on one measured flare. A conical flare now flies the shock-expansion method (the change). On a test rocket with a 10° flare, its center of pressure moves 0.089 to 0.238 calibers forward of the old model’s (evidence row G28). The move grows with Mach number. The one measurement beside it is NASA TN D-4865’s model 2, an 18.5° flare on a 2.75° cone. It isn’t that rocket (Aerodynamics). No other flare angle has been checked. The reading a flare steeper than its corner’s limit gets has been checked against nothing at all. A flare shallow enough to reduce its own element (a few thousandths of a degree on that rocket) takes the older generalized method instead. No measurement checks that either. It leaves a step where the corner’s pressure crosses its tangent cone’s (Aerodynamics). On a body with a short shoulder that crossing isn’t near-flat at all (0.7° to 4.6°). There the step reaches +4.3% of the body’s normal force and 0.19 calibers. Its exact size for any conical flare is on that page.
  • Some shapes fall back to slender-body theory faster than sound, and read more stable. Past Mach 1.2, four shapes keep slender-body theory for the whole body. Two are a step in a body’s radius (#87) and a flare behind a boattail too long for its wake (#120). The others are a pointed tip steeper than 30° (#121) and a vertical tip steeper than its cap’s handover. The sizes are measured at Mach 3 and 4° on two test bodies. In normal force they’re −8.65%, −27.5%, −7.7% and −7.0%, with the center of pressure 1.03, 0.29, 0.81 and 0.64 calibers aft. So the body reads more stable than it is. The flare’s is measured on a finless body, so its share is of a body’s normal force alone. The other three are on a straight rocket with four fins, so the two kinds of share don’t compare (Aerodynamics). No size inside the core band, below Mach 2.5, is pinned yet. They’re first in supersonic accuracy, and in the extended band above Mach 2.5.
  • The normal force near and far past Mach 1. Against NASA’s wind tunnel, between Mach 0.8 and 1.2 HPR Sim’s slope runs up to +27.1% high. Its center of pressure is up to 2.36 calibers off. From Mach 1.5 up its slope holds to within 9.4% (fixture, Aerodynamics). Its center of pressure holds to 0.53 calibers, missing the half-caliber target on the long model (Aerodynamics: body lift). On another body, the Basic Finner, free flights from Mach 1.08 to 3.15 show HPR Sim’s slope 9.3% to 24.9% low and its center of pressure 0.03 to 0.34 calibers forward (Aerodynamics).
  • Tube fins’ center of pressure is unmeasured, and OpenRocket’s differs. HPR Sim’s is 0.42 to 3.0 calibers forward of OpenRocket 24.12’s on 14 probe designs up to Mach 0.5. On OpenRocket’s Tube fin rocket it’s 1.07, and HPR Sim gives the margin as 0.79 calibers to OpenRocket’s 1.87. Neither code has been checked against a measured tube-fin rocket. HPR Sim leaves out the body’s interference, which slender-body theory predicts (#234). Its tube-fin drag probably reads low (#228; tube fins). Check a tube-fin design in both programs and treat the smaller margin as the more cautious estimate, not a bound.
  • Drag against the RASAero curves at Mach 0.3 is within 10% in two of seven cases, with the fins and finish guessed, as the curves don’t record them. Two boattailed cases, Juno III and Calisto’s getting-started fins, are −10.3% and −11.5%. Cavour power-on is −18.3%, cause open. Valetudo’s −47.0% and −50.4% are against a table 1.44 times the OpenRocket export’s (Aerodynamics).
  • Six fins. The normal-force slope of Barrowman’s six-fin Recruiter is +2.87% above his printed value, mostly from HPR Sim’s different six-fin rule (Aerodynamics).
  • Tumbling is −10 to +19% off its source’s own drop tests. It’s used far outside the fit behind it. That fit comes from small models falling at 5.0 to 6.6 m/s (16 to 22 ft/s). If Valetudo came down tumbling, with nothing deployed, HPR Sim would bring it down at 36.4 m/s (119 ft/s). The default streamer model reads +58% fast on a pleated streamer (Recovery).
  • Opening loads, the force on the rocket as a canopy opens, are no safe bound either way. With a filling time, HPR Sim leaves out the brief rise of drag above its steady value while the canopy fills. Opening at once, it ignores how a light rocket slows while the canopy fills. The deployment speed can itself read high or low: a separated body falls with no drag until its device opens (Recovery).
  • No added mass under a canopy. It’s the likely cause of the 2.86% drift difference above (Recovery). It also causes NDRT’s +83.059% and Prometheus 2022’s +19.062% peaks as their mains open in the whole flight (report, case file).
  • Body lift in wind. A slow rocket leaves the rail at a steep angle to a crosswind. There HPR Sim’s body lift, which RocketPy leaves out, is the largest reason its drift differs. How much body lift a rocket body makes is itself uncertain. HPR Sim takes it from Jorgensen’s crossflow term (Aerodynamics), whose factor is about 0.9 for these rockets at low speed. Flown in RocketPy with HPR Sim’s rail release and fins, that puts Juno III’s apogee 248.3 m (815 ft) from the pad. HPR Sim’s own is 245.3 m (805 ft, case file). The same runs set Galejs’s constant K, which HPR Sim used before, across its source’s range. It gives 240.2 m (788 ft) at 1.0, 231.1 m (758 ft) at 1.1 and 194.1 m (637 ft) at 1.5. With no body lift at all, the drift would be 328.0 m (1,076 ft) (the decision). Only real flights can say which is right. The real flights so far compare heights, not drift.
  • Airfoil fins. HPR Sim’s fins use the flat-plate lift slope. It can’t model an airfoil lift curve such as the one Juno III’s example gives its fins, which makes RocketPy’s fin slope 7.6% steeper (decision).
  • Turbulence is an aircraft model, unvalidated for rockets, and no flight uses it yet (Turbulence) (evidence row G12).
  • Wall and fin mass may follow different conventions from OpenRocket’s, which its documentation doesn’t state. Six are measured so far (Mass properties). OpenRocket’s nose cone walls are measured across each station, HPR Sim’s square to the surface: the same on a cone, HPR Sim’s up to 0.48% heavier on three blunt noses (Shapes).
FusionSpace HPRVersion 0.1.0Issued 2026-11-21In preparationAbout HPR Sim