Flight metrics
Learn how HPR Sim measures the numbers a flight is judged by, such as apogee, top speed, max q, stability margin and landing point.
In short
- What it models: the numbers a flight is judged by. These are the apogee, the top speed and
Mach number, and the peak dynamic pressure (“max q”). Then
come the boost’s peak acceleration and, kept apart, the opening shock. The metrics also give the
stability margin from the rail exit to apogee or the first
deployment. They give the largest
angle of attack that counts toward the envelope’s
high-angle flag (below). They say whether the rocket is unstable
while a motor burns (below). They give the
ejection delay that would fire the charge at apogee, and each
landing’s latitude and longitude. Anything that didn’t happen is
None, never a zero. - Sources: the margin is Barrowman’s center of pressure (to Mach 0.8; faster, as Aerodynamics describes) against the center of mass (the CG). It is defined as RocketPy defines its static margin and stability margin. The peak search is Kiefer’s golden-section search (References). Latitude and longitude come from HPR Sim’s WGS 84 conversions (Geodesy).
- How well it is validated: each number is only as good as the flight it comes from. The flight is checked against RocketPy and OpenRocket (Accuracy), and the metrics add no physics of their own. Tests check each against a hand calculation or against the simulator’s own models evaluated directly (Tests). In a test, the margin in the weakest plane is checked against the simulator’s own scan of 3,600 planes. It is also checked against one OpenRocket run that is not committed. None is validated against a real flight.
- What it leaves out:
- Fin flutter, which has its own page: Fin flutter. Its margin comes from the max q found here.
- File exports: a later part of the same milestone, M1.10c (CSV, JSON, KML and GeoJSON files).
- The descent of a separated body, such as a dropped booster. It gets a landing, but no peaks.
- A damping ratio: how fast a wobble dies out. Both margins here are static quantities.
- The margin at the flight’s angle of attack. Both margins take the air along the rocket’s axis, in the weakest direction it can cross (Stability margins). A rocket meets the air at an angle when it leaves a rail in wind. In the simulator’s model its margin there can be smaller or larger than these, depending on where its body’s lift acts. Stability margins says why they leave it out.
Purpose and use
Why the metrics are built so, and how to ask for them.
Why these rules
Loft, this project’s predecessor, got four things wrong that these metrics are built to avoid:
- It read peaks off a table of recorded rows and missed them, and it counted the opening shock as the boost’s peak acceleration (L34).
- It printed zeros for things that never happened, and never said which height an apogee was counted from (L35).
- It published margins of ±12 to 15 calibers for rockets where a margin has no meaning (L33).
- Its optimum ejection delay depended on the delay flown (L94).
Each is a numbered lesson with a test (Tests).
Getting the numbers
Fly a Simulation with a
FlightMetrics watching it. Then ask the
watcher for a FlightSummary:
let mut metrics = FlightMetrics::new();
let flight = simulation.run(&mut metrics)?;
let summary = metrics.summary(&flight, simulation.environment())?;
let best = hpr_sim::metrics::optimum_delays(&simulation)?;
A watcher keeps one flight. Call metrics.clear() before it watches another; summary refuses a
flight unless the watcher saw each of its steps once.
The example program
flight_metrics.rs
does this for Valetudo, the rocket from Getting started. It flies with a
5 m/s (16 ft/s) wind from the west and one 1.5 m (4.9 ft) parachute. The charge fires 6 s after
burnout (the end of the thrust curve), and the canopy is open 0.5 s
later. Run it with cargo run --example flight_metrics -p fusionspace-hpr-sim. It prints:
Valetudo on a K400C, a 1.5 m parachute fired 6 s after burnout, open 0.5 s later
Not yet validated: see the Accuracy page before trusting these numbers.
Heights are the center of mass's above the launch site; it starts 0.94 m up.
Apogee: 714.0 m above the site (713.0 m of climb) at 11.16 s
Top speed: 112.3 m/s at 3.00 s, 188 m up
Top Mach number: 0.337 at 3.02 s, 190 m up
Max q: 6667 Pa at 2.99 s, 186 m up
Boost acceleration: 47.2 m/s² at 0.03 s, 1 m up
Opening shock: 144.0 m/s² at 9.76 s, 698 m up
Rail exit: 16.2 m/s at 0.37 s, 17.2° off the oncoming air
At rail exit (0.37 s): static margin 3.09 cal, flight margin 3.09 cal at Mach 0.051
Least static margin: 3.09 cal at 0.37 s, 4 m up
Least flight margin: 3.09 cal at 0.37 s, 4 m up
Optimum delay: 10.6 s after burnout at 3.26 s (flown: 6.0 s), for an apogee of 778.7 m at 13.83 s
Landing: 32.990000° N, 106.967084° W: 272.6 m east and 0.0 m north of the site, at 11.6 m/s, at 78.67 s
The 6 s delay is too short:
- The charge fires at 9.26 s, and the canopy opens at 9.76 s: about 4 s (13.83 − 9.76) before the apogee it would have reached.
- The rocket still rises 1.4 s more, to an apogee of 714.0 m (2,343 ft).
- With the charge held, it would have coasted to 778.7 m (2,555 ft) at 13.83 s. That is the apogee of the Getting started flight, whose drogue fires at apogee.
- So the optimum delay is 10.6 s. The opening shock, 144.0 m/s², is three times the boost’s 47.2 m/s², and it is not counted as the boost’s peak.
- Don’t size a shock cord from the 144.0 m/s²; it is no bound either way. The canopy reaches full drag at line stretch, 0.5 s after the charge. It has no filling time (the simulator’s default), so the rocket doesn’t slow while it fills. The shock reads high. The simulator also leaves out a canopy’s drag overshoot near the end of filling: that reads low (The opening load).
Top speed comes at 3.00 s, before the 3.26 s burnout. That is because in the thrust curve’s last moments the motor pushes less than drag and gravity pull back.
Heights and peaks
Which heights a summary gives, and how each peak is found between the integrator’s steps.
Heights
Every height is of the center of mass: its
ellipsoidal height above the launch site’s, the same as
a flight’s height_above_ground_m. The center of mass starts above the site, because the rocket
stands on the rail: 0.94 m (3.1 ft) for Valetudo. So the summary gives both:
apogee.height_above_ground_m: 714.0 m (2,343 ft), above the site.apogee.gain_m: 713.0 m (2,339 ft), the climb from where the center of mass stood at launch. OpenRocket’s altitude counts this way. A flight started in the air (Simulation::run_free) has no launch height, and no climb:None.
After a separation with nothing left to burn, the summary’s apogee
is the apogee of the part that keeps the nose (body 0). If an
ejection parts the rocket, or that part, while it still climbs, the
summary has no apogee (None): what the charge leaves at the nose is lighter, and coasts to a
height of its own. Ejected pieces gives each case.
A flight ends when its center of mass comes back down to the site’s height, so a landing is at height 0 by this measure.
The rail exit speed is the center of mass’s speed over the ground when the aft edge of the aft-most rail button or lug passes the rail’s top, or the rocket’s aft end does with no guides (Flight). On the rail the rocket doesn’t turn, so every point of it moves alike, but the center of mass also slides forward as the propellant burns. So the printed speed reads a little higher than the guides’.
Peaks
Each peak comes with its time and its height. The watcher looks at every step the integrator takes (Time integration and events), not at a table of recorded rows:
- It evaluates the equations of motion at each step’s start, middle and end.
- It fits a parabola through the three values. When the parabola bends down with its top inside the step, the peak may lie inside. A golden-section search then looks for it on the step’s dense output (the integrator’s smooth curve through the step). The search runs to a billionth of the time since launch, or of 1 s early in the flight.
- It keeps the largest of what the search finds and the three samples.
Thrust-curve knots, the times where a motor’s tabulated thrust changes slope, end steps. So a spike in the thrust curve is a step’s end, and is never averaged away. Loft took its peak acceleration from a finite difference of the recorded speed, and read it low. Valetudo in a vacuum shows how much. The watcher’s peak matches the hand value from the motor and the masses within the test’s 1e-6. A finite difference of the speed recorded at 100 Hz reads it more than 1% low.
| Peak | What it is |
|---|---|
| Top speed | The center of mass’s speed relative to the ground, after liftoff |
| Top Mach number | The airspeed over the local speed of sound |
| Max q | The largest dynamic pressure, on the airspeed |
| Boost acceleration | The largest acceleration of the nose tip, from liftoff until a recovery device opens |
| Opening shock | The largest specific force while a device is open: the drag and any thrust over the mass, the opening load over the mass, without gravity. It follows the simulator’s inflation model (Recovery) |
The boost acceleration is of the nose tip, which is the body’s origin in the simulator’s frames. It differs from the center of mass’s only by the rocket’s turning and by the center of mass’s slow drift forward as propellant burns. The turning is small in a straight boost. It is relative to the launch frame and straight from the equations of motion, so it includes gravity’s pull, as a trajectory’s acceleration does.
The opening shock leaves gravity out. Under a canopy the motion’s acceleration is the drag and the
weight over the mass. Falling, the drag points up and the weight down, so the motion’s
acceleration is the drag over the mass less g, and it would read the load low by up to g. The
opening shock is the specific force instead, the drag over the
mass (with a motor still burning, the drag and the thrust together). That is what an accelerometer
reads, and with no motor burning it is the opening force over the mass (Knacke, NWC TP 6575,
p. 5-50, gives the opening force as ). In a canopy’s steady fall it is
1 g, so it reads at least about that once the canopy reaches its steady fall. A canopy opened
slower than its steady fall reads less if it doesn’t get there while it is counted: the flight
reaches the ground first, or an ejection parts the stack or the part
that keeps the nose, which ends the count. After a separation with
nothing left to burn, the part that keeps the nose (body 0) is counted until it ejects a piece
itself or lands, and the part dropped aft never is
(the opening load).
Each separated body has its own figure beside the flight’s, body_max_descent_acceleration_m_s2,
in m/s²: the same specific force over that body’s own flight and mass
(how to read it).
In the example above the canopy opens while the rocket still climbs, with the drag and the weight both pointing down, so the motion’s acceleration there (the
plot’s spike) is close to g above the opening shock of 144.0 m/s².
A test opens a canopy 5 s after apogee, falling, and finds the shock equal to at the
opening to 1e-9. At the ground, in the canopy’s steady fall, the specific force is g and the
motion’s acceleration nearly nothing
(the_opening_shock_is_the_drag_over_the_mass_without_gravity).
Nothing is kept while the rocket sits on the pad. A rocket that never lifts off has no peaks at
all: None, not zero.
Stability
The margin, the largest angle of attack, and the unstable-under-power flag.
Stability margins
The margin is how far the center of pressure lies behind the center of mass, in calibers of the reference diameter :
Here is a station measured aft of the nose tip, and is component ’s normal-force slope. The simulator gives two margins at each instant:
- Static margin: the center of pressure at zero angle of attack and Mach 0, against the center
of mass of that instant. This is RocketPy’s
static_margin. It changes only as propellant burns. - Flight margin: the center of pressure at the flight’s own Mach number, still with the air
along the axis, against that instant’s center of mass. It is defined as RocketPy’s
stability_marginis.
RocketPy’s min_stability_margin takes the least over its whole flight, on the rail and in the
descent too. It reads it at its solver’s steps, so it can differ from HPR Sim’s least below. No
page compares HPR Sim’s margins with RocketPy’s yet.
Against OpenRocket, HPR Sim’s margin at the flight’s Mach number is within 0.016 calibers on 41 of
the 53 flights of OpenRocket’s own examples. That is in calm air at rod clearance, where a rocket
is still slow. HPR Sim’s margin in this comparison has the air along the rocket’s axis (the 0°
plane), as OpenRocket’s does. It isn’t the margin in HPR Sim’s weakest plane (below). On
[C6-7; B6-0] of Pods--powered with recovery deployment, HPR Sim’s margin at rod clearance in
the report is +3.74 calibers along the axis. But hpr sim prints a least margin of −4.21 in the
weakest plane. At rod clearance in the 0° plane, HPR Sim’s margin differs from OpenRocket’s on:
- the Tube fin rocket: 1.08 calibers below, tube fins;
- the three-stage example’s three flights: 0.039 to 0.058 below, cause not yet sized, #185;
- the five of Pods--airframes and winglets: 0.071 to 0.076 above, the flattering side, #325 and #326;
- the three of Pods--powered with recovery deployment: 0.070 above, also not yet traced;
- 35 flights of private designs: from 0.017 lower to 0.11 higher, cause not yet traced (Accuracy).
The Mach number moves the center of pressure. On a rocket with fins only at the tail it generally moves aft until supersonic theory starts for the fins, Mach 1.2 or later. Their slope grows, and from Mach 0.8 their own center moves aft too. Past that their slope falls and the body’s lift can grow, and the center of pressure mostly moves forward again (Your rocket’s center of pressure). Valetudo is slow: at its rail exit, at Mach 0.051, both margins read 3.09 calibers. Near Mach 1, HPR Sim puts the Arcas Robin’s center of pressure up to 2.36 calibers behind the wind tunnel’s (Normal force through Mach 1). So there its flight margin reads high.
Both margins are the weakest direction’s. The air can cross a rocket from any direction around
its axis, and for most rockets the margin is the same in every one. A fin set of one or two fins
breaks that. It carries of its force in the plane at roll angle , for fins at
angles . The share is Niskanen’s (2009, eq. 3.51, as in Fins). So a rocket with one has a margin for
each direction, and the simulator gives the least, the direction it was found in as roll_rad.
Until #329, the simulator’s margins were the 0°
direction’s alone, which can be the strongest. A rocket whose fin sets each have three or more fins,
or that flies a normal-force table, keeps its one margin, bit for bit.
A worked case: the sustainer of OpenRocket’s Pods--powered with recovery deployment example has
a single fin at 0° and a two-fin strake set at 90°. At Mach 0.2, after the booster drops, its
margin is +1.86 calibers with the air crossing at 0°. There the single fin lies edge-on and the
strakes carry the force. At 45° it is +0.17. At 90° the strakes lie edge-on and only the single
fin works: −3.44, unstable. OpenRocket 24.12 was asked for that sustainer at 90° in a scratch run (not
committed). It puts its center of pressure at 0.2954 m (0.969 ft) with 3.49. HPR Sim’s are 0.2993 m (0.982 ft) and
3.47. For one of its staged configurations, [C6-7; B6-0], hpr sim’s least margin was +1.56 calibers and is now −4.21, at 0.86 s. That is the sustainer’s margin at the split itself, the moment
the booster drops. The sustainer’s own motor is just lit and still full, and the air is at Mach 0.07. The −3.44
is a later instant, at Mach 0.2. In the conditions of OpenRocket’s record, both programs’ flights of
that configuration turn over before apogee. OpenRocket’s record has its site at 28.61° N, where the Earth’s rotation tips
HPR Sim’s flight. HPR Sim’s angle of attack passes 90° at 2.25 s. At hpr sim’s default site, on the
equator, nothing tips HPR Sim’s flight: it stays upright and reaches 251.3 m (824 ft)
(the format guide).
The simulator finds the least in closed form, not by a scan
(weakest_margin). Each part’s slope varies as
, so the net slope and its moment do too.
Three directions, 0°, 60° and 120°, give each sum’s , and . The center of pressure
is foremost where
with
and and the coefficients of and . The margin’s conditioning (below) is checked at its own worst direction the same way. Each root and the 0° direction are evaluated through the model, and a direction where no margin can be given wins over any number.
Both margins leave out the angle of attack. In the simulator,
body lift grows with the angle and acts at each body’s side-view centroid, and
the center of pressure moves toward it. Where that centroid lies ahead of the zero-angle center of
pressure, the margin at an angle shrinks. On a long body with small fins it can lie behind, and the
margin grows. No test pins either direction. But the angle is not a steady property of
the rocket. Valetudo leaves its rail 17.2° off the oncoming air in the example’s 5 m/s (16 ft/s) crosswind.
Near apogee the angle swings toward 90° as the rocket slows and tips over. If the least margin
followed the angle, it would land wherever the simulator chose to stop counting large angles. The
simulator models no fin stall that could say where that is. For the
margin at a given angle, call margin with
Flow::new(mach, angle_rad, roll_rad).
The watcher keeps both margins in FlightMetrics::stability(). The series starts at the rail exit
and ends at apogee or when a recovery device opens, whichever comes first. Before the rail exit the
rail holds the rocket, so its margin says nothing about how it flies.
The summary’s least margins cover the same span. Each is looked for inside every step, as a peak is (Peaks). Wherever the margin is defined at a step’s start, middle and end, a parabola goes through the three values. When it bends up with its bottom inside the step, a golden-section search finds the bottom. A later least replaces an earlier one only when lower by more than rounding, so a flat least keeps its first time. A margin with a sharp corner near a step’s end can still hide from the parabola.
- On four test flights, and in the example, both leasts come at the rail exit. There the rocket is heaviest and its center of mass furthest aft. Flown in steps of at most 1 ms, those flights give the same leasts to a millionth of a caliber.
- On a two-stage test flight the least flight margin comes inside a step, after the split. There it matches a scan of every step at 201 points to 1e-7 calibers, and is never above it.
After a powered separation the margins are the sustainer’s, in its own diameter. The split has two entries in the series: the stack’s, then the sustainer’s.
When there is no margin
As the net slope goes to zero, the air’s loads become a pure couple: a turning moment with no line of action. The quotient then runs away. That is how Loft came to publish margins of ±12 to 15 calibers.
The simulator judges the net slope against the sum of its terms’ sizes:
Each part the model adds up acts at a station on the rocket, within its length . The center of pressure then lies within of every station. So a fractional error in one part’s slope moves the center of pressure by at most about (to first order). A part that is a pure couple on its own (below) has no station, and doesn’t count it.
- A rocket whose parts all push the same way has . A 1% error then moves its center of
pressure by at most 1% of its length. All 13 designs in
validation/designs/stay below 1.5. That holds at Mach 0, 0.3, 0.8, 1.2 and 2 and at angles of attack of 0°, 5°, 10° and 20°. - At , a 1% error in one slope can move the center of pressure by a tenth of the
rocket. Past that, or when the net slope is not positive, the simulator gives no margin and no
center of pressure:
None. The limit is a chosen bound on that sensitivity, not a measurement.
The pitch-moment slope about the center of mass is always given, because it stays finite:
A negative turns the nose back into the wind. With a margin defined, .
A worked case, with no fins, pinned by a test:
-
The rocket has a conical nose 0.3 m (0.98 ft) long of radius = 0.05 m, then a 0.5 m (1.6 ft) tube. Last comes a conical boattail 0.4 m (1.3 ft) long, narrowing to a radius . So = 0.1 m.
-
Barrowman gives the nose a slope of 2 at 0.2 m (0.66 ft). The boattail gets , at a distance in meters of
-
So , with . The margin is given for , an aft radius of 34.66 mm (1.4 in) or more.
-
The center of mass is at 0.5 m (1.6 ft).
| Boattail’s aft radius | Net slope, per rad | ratio | Margin | Pitch-moment slope , per rad |
|---|---|---|---|---|
| 40 mm (1.6 in) | 1.28 | 2.13 | −7.46 cal | +9.55 |
| 35 mm (1.4 in) | 0.98 | 3.08 | −11.2 cal | +10.98 |
| 34.5 mm (1.4 in) | 0.952 | 3.20 | none (the quotient: −11.7 cal) | +11.11 |
| 21.5 mm (0.85 in) | 0.370 | 9.82 | none (the quotient: −37.1 cal) | +13.72 |
| 5 mm (0.20 in) | 0.020 | 199 | none (the quotient: −741 cal) | +14.82 |
With no fins, every one of these rockets is unstable: its center of pressure lies ahead of its center of mass, and is positive. The cut is not about how large the margin is. The 35 mm (1.4 in) and 34.5 mm (1.4 in) rows have margins near −11 calibers; what differs is whether the quotient can be trusted.
A component can be a pure couple on its own. A step down in radius followed by a flare back up cancels its own slope and still turns the rocket. keeps its moment, which a test also checks.
The largest angle of attack
max_angle_of_attack_rad is the largest angle between the rocket’s axis and the air flowing past
it, in radians. Its span is the one the envelope’s high-angle flag watches
(Validation plan: the operating envelope;
ADR-179):
-
from more than 1 s after the rail exit, as the first second off the rail is the expected transient in a crosswind;
-
to apogee or the first deployment, the span of the least margins above;
-
only at instants where a 15° angle would give a normal force of at least a fifth of the rocket’s weight:
with the dynamic pressure, the reference diameter and the flight margin’s normal-force slope.
The angle is read at each step’s start, middle and end. Unlike the peaks, it has no search between them, so an excursion inside one step can be missed. Above 15°, the flight’s
envelope_flags include high_angle_of_attack.
The floor is there because near apogee every flight whose path turns over passes 15°. There the air
is too weak to turn it: gravity does. Valetudo, one of RocketPy’s examples, launched off an 84° rail
in calm air, passes 15° about 0.9 s before apogee. It is then at 16.7 m/s (55 ft/s) through the air, a little faster
than its 16.2 m/s (53 ft/s) rail exit. There a 15° angle would give a normal force of 1.0% to 1.5% of its
weight. So an error in the aerodynamics at that angle barely moves it. A wind layer met at speed is
different. Valetudo climbing into 30 m/s (98 ft/s) of wind 300 m (984 ft) up passes 15° where that force is 62% of its
weight. The flag is raised. On the tests’ RocketPy rockets, turn-overs give 0.7% to 5.0% and
wind layers met at speed 62% to 141%. The floor of a fifth, chosen rather than measured, sits between. The
tests a_tilted_calm_climb_passes_15_degrees_only_with_too_little_force,
a_wind_layer_met_at_speed_raises_the_high_angle_flag and
a_wind_layer_met_in_a_fast_coast_raises_the_high_angle_flag (in the flight engine’s metrics.rs) fly
these cases.
Unstable under power
A flight that is unstable while a motor burns raises one of two flags, and its apogee is not a prediction (#335). Unstable means the air turns the rocket away from its path instead of back. Where it ends up then depends on small disturbances, and on aerodynamics that hold only at small angles of attack. The simulator still flies it, and the flag changes no number: it marks the flight’s path and apogee as not a prediction. A flight raises at most one of the two flags.
unstable_under_power: the static margin falls below zero while a motor burns. The center of pressure is then ahead of the center of mass.unstable_without_margin: no static margin can be given while a motor burns (When there is no margin), and the pitching moment’s slope is above zero. It is raised only when the first flag isn’t.
min_powered_static_margin_cal is the margin the first flag reads. It is the least static margin
in the weakest plane (Stability margins), over the moments a motor burns.
The span runs from the rail exit to apogee or the first deployment. The margin is found inside steps, as the other least
margins are. A step counts when the thrust at its middle is above zero. The simulator ends a step at
every ignition, thrust-curve point and burnout. So a motor that burns at a step’s middle burns
through all of it, and the margin at the burnout itself counts. A margin of exactly zero raises
nothing; any margin below it does.
max_powered_moment_slope_per_rad is what the second flag reads. The simulator gives no margin where the
parts’ normal-force slopes nearly cancel, or their sum is not positive. There the center
of pressure would be noise. The moment about the center of mass is still well defined there:
with and the normal-force slope and station of part , their sum, and the center of mass. Every station is aft of the nose tip, and is the reference diameter. A negative turns the rocket back toward its path; a positive one turns it away. The simulator keeps the largest at each powered step’s start, middle and end where the margin is undefined, without searching between them. A of exactly zero raises nothing; any above it does. Without this flag, a rocket so unstable that its margin is undefined would raise no warning at all, while a milder one would.
Only the moments under power count. A rocket that turns then is driven along its new heading by
its own motor, so the path flown follows the turn. A rocket that is unstable only in the coast,
after its motors burn out, raises neither flag. It has no thrust to carry it off its path. The
least static margin over the whole span, min_static_margin_cal, still prints its margin
there.
For example, OpenRocket’s example Pods--powered with recovery deployment, flown on [C6-7; B6-0]
or [C6-7; 2× A3-4, B6-0], separates at 0.86 s. The sustainer left has a static margin of −4.21 calibers with its C6 still burning. On [C6-7; B6-0]:
static margin 2.05 calibers off the rail; least -4.21 calibers, at 0.86 s, before apogee
apogee 251.3 m above the site at 5.74 s, not a prediction: unstable under power
warning: unstable: the static margin falls to -4.21 calibers at 0.86 s while a motor burns. The rocket is unstable under power: the air turns it away from its path rather than back, and its apogee is not a prediction
For the second flag, take RocketPy’s Valetudo with its fins taken off. Give it a conical tail that narrows from the body’s 40.45 mm (1.6 in) radius to 22 mm (0.87 in) over 0.08 m (0.26 ft). At the rail exit its net slope is 0.59 per radian against a sum of sizes of 3.41. That is too small for a margin, but is +43.9 per radian:
apogee 399.1 m above the site at 8.46 s, not a prediction: unstable under power
warning: unstable: while a motor burns, the pitching moment turns the rocket away from its path (C_mα +43.9 per radian at 0.25 s). HPR Sim can give no static margin then. The rocket is unstable under power, so its apogee is not a prediction
With the tail ending at 30 mm (1.2 in) instead, the margin is defined, −34.37 calibers, and the first flag
fires in its place. These figures are from a run on October 6, 2026; the test
(sim_flags_a_rocket_unstable_where_it_has_no_margin) pins only the flag and a slope above 1.
hpr sim prints the warning: unstable: line and marks the apogee for either flag. Its JSON
lists the flag under flags. The library has it in FlightSummary::envelope_flags, and Python in
Flight.envelope_flags. When this was written (October 6, 2026), none of the 25 public .ork files
and 12 validation designs that hpr sim flies offline raised either flag. That was on their default
configurations off an 85° rail in calm air and in 8 m/s (26 ft/s) of wind. No test or report pins that
count.
What it leaves out:
- A rocket unstable only after its motors burn out raises no flag, as above.
- It reads the static margin and at Mach 0, not at the flight’s Mach number.
- is read only at each step’s start, middle and end, not searched between. A moment slope above zero for less than half a step, with no margin, can pass unseen.
- A flight that tips over above 15° raises the high-angle flag (The largest angle of attack) as well. A stable rocket in a strong wind can raise that one alone.
Five of the tests in Tests pin these flags, four at their edges. In hpr-cli,
sim_flags_a_rocket_unstable_where_it_has_no_margin flies the Valetudo example above.
Ejection and landing
The delay that fires a charge at apogee, and where each body lands.
Optimum ejection delay
The optimum delay is the time from a motor’s burnout to apogee: the delay that fires its charge at
the top. optimum_delays flies the rocket again
with every recovery charge held, so that it coasts to the apogee it would reach untouched. So the
answer belongs to the rocket, its motors and its air, not to the delay flown. A charge that fires
too early cuts the coast short. Loft’s optimum then came out too short as well, which advised an
even shorter delay. HPR Sim gives the same optimum for delays of 1 s and 20 s.
- A separation with nothing ahead of it left to burn is part of the recovery, so it is held too. A two-stage rocket that separates some seconds after its last burnout gets the same optimum for delays of 3 s and 30 s.
- A powered separation still happens. The motors in the booster it drops have no optimum: their charges fire in the booster, which never reaches the sustainer’s apogee.
- A motor that burns out after the apogee, or never lights, has none either. A flight that never
reaches an apogee (it never lifts off, or hits its time cap) gives
None.
Landing points
A landing is where the center of mass came back down to the site’s height. It is given as a WGS 84 latitude and longitude, and as east and north meters from the site in its local frame. The speed at the ground hit is given too. The flight’s own landing is the stack’s (the whole rocket before any separation), or after a powered separation the sustainer’s. Each separated body that lands has its own landing, with its body number. Body 0 keeps the nose, and 1 is the stages aft of the split.
A flight that did not land has no landing and no ground-hit speed: None. A test checks this for a
flight stopped by its time cap, and the JSON it writes, where the missing values are null.
Tests and references
The tests that pin these metrics, and the sources they follow.
Tests
In crates/hpr-sim/src/metrics.rs, unless named otherwise:
| Test | What it pins |
|---|---|
static_margin_undefined_when_cn_alpha_near_zero | The boattail table above, against Barrowman by hand, to 1e-9 relative, on both sides of the limit (L33) |
ordinary_rockets_keep_their_margin | All 13 validation designs keep their margin at Mach 0, 0.3, 0.8, 1.2 and 2 and at 0°, 5°, 10° and 20°, with below 1.5 |
a_pure_couple_keeps_its_moment | The step and flare above: against the hand value, to 1e-9 |
a_normal_force_table_gives_its_own_margin | A table’s margin and against its own numbers |
the_least_margin_is_the_weakest_plane_s | Six layouts of a single fin and a two-fin set, at Mach 0.1, 0.6 and 1.5 and three centers of mass: the least margin is never above any of 3,600 scanned planes’, and within 1e-5 calibers of the scan’s least, including where the weakest plane lies between the fins’ planes; fin sets of three or more fins keep the margin along the axis, bit for bit |
peak_acceleration_is_analytic_and_excludes_opening_shock | The boost’s peak in a vacuum against the hand value, to 1e-6; a 100 Hz finite difference reads it more than 1% low; an opening shock over three times the boost’s is kept apart (L34) |
the_opening_shock_is_the_drag_over_the_mass_without_gravity | A canopy opened falling, 5 s after apogee: the opening shock is at the opening to 1e-9, about g above what the motion’s acceleration reads there by the hand relation ; at the ground, in its steady fall, the specific force is g up to 1e-3 and the motion’s acceleration under 1e-3 g (#1258) |
the_part_that_keeps_the_nose_has_its_own_opening_shock (in recovery.rs, as the next two rows’ tests are) | After a separation with nothing left to burn, the stack having opened nothing: the summary’s shock is the nose part’s at its instant opening to 1e-9, at that opening’s time and height, and not the aft part’s, which is more than 1.5 times larger (#1519) |
the_nose_parts_shock_tops_inside_a_step_under_a_filling_canopy | The nose part under a canopy that fills over 1 s in an 8 m/s (26 ft/s) crosswind: each step’s specific force is the sample’s drag area, airspeed and mass by hand to 1e-12; the summary’s shock is the top of a scan of the steps that meet the filling, 2,000 intervals each, no lower to 1e-9 and no more than 1e-6 higher, where the steps’ starts, middles and ends alone read more than 1e-5 low |
a_split_with_an_ejection_keeps_the_lead_pieces_shock_out, the_nose_parts_shock_stops_counting_where_an_ejection_parts_it | An ejection that parts the stack or the part that keeps the nose ends the count: a nose cone ejected at the split, its device opening at over 50 m/s (164 ft/s), leaves the summary no shock; one ejected after the nose part’s drogue opened leaves the drogue’s to 1e-9, though the cone’s own shock is more than 5 times larger |
an_ejection_in_the_part_dropped_aft_leaves_the_flights_shock_alone (in recovery.rs) | Each separated body’s own opening shock (#1532), at an instant opening to 1e-9 over that body’s own mass: the two pieces of a dropped part that parts at a charge of its own, each more than 1.5 times the flight’s shock, which is the same to the bit with that charge and without. The tests of the rows above pin the same for the part dropped aft at a separation, the airframe behind an ejected nose cone’s joint and the cone itself |
each_bodys_own_shock_is_its_largest_step_peak_and_a_nan_stays, a_summary_refuses_separated_bodies_steps_from_another_flight | A body’s figure is the largest of its steps’ peaks and the first of equals; a step with no load makes no entry; a step’s peak that is not a number stays. The summary refuses a watcher that saw a step of a body the flight doesn’t have, or not each step of a body it has once |
a_nose_cone_ejected_on_the_climb_after_a_split_leaves_the_summary_no_apogee (in recovery.rs) | After a separation with nothing left to burn, the part that keeps the nose pushes its nose cone off 5 s later, still climbing at over 30 m/s (98 ft/s): the cone tops out more than 50 m (164 ft) below the whole part’s apogee, and the summary’s apogee is None; with the charge at the part’s apogee or 3 s after it, the summary keeps the whole part’s apogee (#1531) |
the_nose_parts_kept_peaks_are_a_running_maximum_with_the_same_shock (in recovery.rs), body_0s_kept_peaks_are_the_first_of_equals_and_each_larger_one | The watcher keeps a step’s peak for the nose part only when it beats the last one kept. On three separation flights (an instant opening, a drogue then a pushed-off nose cone, a canopy filling in a crosswind) the summary’s shock, time and height equal those from every step’s peak, with at most a twentieth of the steps kept; equal peaks keep the first, and smaller ones add nothing (#1533) |
unlanded_flight_has_no_ground_hit_speed_and_outputs_name_datum | None and null for an unlanded flight; the launch height against the rail’s geometry, to 1e-9 relative (L35) |
optimum_delay_independent_of_flown_delay | Delays of 1 s and 20 s give the same optimum, equal to a flight with no recovery (L94) |
peaks_are_refined_inside_steps | On three rockets, max q and top Mach are above every row of a 1 ms record, and the record’s best row is within 0.01% of them; on Valetudo max q comes before top speed, and top speed before top Mach |
landings_are_placed_on_the_ellipsoid | A landing more than 100 m (328 ft) downwind, against the radii of curvature at the site, to second order |
stability_is_kept_from_rail_exit_to_apogee | The series’ ends; the static margin at the rail exit and, in a crosswind, the flight margin at the flight’s Mach number, against the model and the masses directly, to 1e-12; the least flight margin is at or below every entry |
least_margins_do_not_depend_on_where_steps_end | Valetudo in calm air off a vertical and an 84° rail and in a crosswind, and Prometheus: both leasts are at the rail exit, and agree with steps of at most 1 ms to 1e-6 calibers |
a_least_margin_between_step_ends_is_found | A margin of across a step: the search finds 2 at 0.37 s; a margin falling across the step, or undefined in its middle, is not searched |
a_summary_needs_the_flight_it_watched | A watcher refuses a flight when it missed some of its steps or also watched another flight, naming the steps it saw, and sums up one that ended without a step; a flight started in the air has no launch height, and one started on its way down keeps no stability |
the_powered_margin_counts_only_while_a_motor_burns | A margin falling 1 caliber a second across a burnout at 1 s: −2⁻³⁰ calibers at the burnout raises the flag, +2⁻³⁰ doesn’t, nor does −1 in the coast after; a least inside a powered step is found by the search; a NaN under power is kept, one in the coast isn’t; a margin of −10⁻¹³ replaces +10⁻¹³ though the two are closer than the tie |
a_moment_slope_without_a_margin_counts_only_while_a_motor_burns | With no margin, a of +2⁻³⁰ per radian under power raises unstable_without_margin, −2⁻³⁰ doesn’t, nor does +2⁻³⁰ after the burnout at 1 s; a NaN under power is kept; a slope where the margin is defined doesn’t count; a margin below zero as well raises only unstable_under_power |
a_rocket_without_fins_is_unstable_under_power | Valetudo with its fins taken off raises unstable_under_power at or before its burnout; stock, its least margin under power is above 1 caliber and it raises nothing |
envelope::tests::the_stability_flag_starts_just_below_a_zero_margin | The flag from the least negative f64 down; none at 0, −0 or above |
envelope::tests::the_moment_flag_starts_just_above_zero_and_only_without_the_margin_flag | unstable_without_margin from the least positive f64 up, and for a NaN; none at 0, −0 or below; never beside unstable_under_power |
staging::tests::metrics_follow_a_powered_separation | Both landings; the sustainer’s diameter after the split, with its own entry there; both leasts against a scan of every step at 201 points, to 1e-7 calibers, the flight margin’s inside a step; no optimum for the booster’s motor |
staging::tests::a_held_flight_holds_a_separation_after_the_last_burnout | A split after the last burnout is held: the same optimum for delays of 3 s and 30 s |
References
- J. Kiefer, “Sequential minimax search for a maximum”, Proceedings of the American Mathematical Society 4(3), 502–506, 1953. The golden-section search.
- J. S. Barrowman and J. A. Barrowman, “The theoretical prediction of the center of pressure”, NARAM-8 research and development report, 1966. The slopes and stations, as Aerodynamics uses them.
- S. Niskanen, Development of an Open Source model rocket simulation software, MSc thesis, Helsinki University of Technology, 2009, eq. 3.51. A fin’s share of its force in the plane the air crosses in, which makes a one- or two-fin set’s margin depend on that plane.
The decision behind these choices is ADR-077, from the M1.10a milestone.