Expand description
Apogee predicted from the boost: a log cut after burnout, fitted, and flown on to apogee (ADR-257 §3; ADR-321, amended 2026-11-06).
The cut must not read the flight after it. A test replaces every later value and time with noise and the prediction stays bit-identical. Two rules hold that by construction:
find_cutwalks the height track forward and decides at each sample from that sample and the ones before it. A vertical speed is fitted by least squares over the lastSPEED_WINDOW_Sof samples, leaving out a sample the analyzer’s Hampel rule calls bad: one bad reading of 12 to 100 m put in set A’s flights moved a cut by 0.70 s at most (measured). Liftoff is where that line standsLIFTOFF_RISE_Mabove the pad, the line where it last stood still (STILL_M_S), or before that the first window’s. The cut is at the first sample where the speed has stood below its top byDROP_SHAREof it, and by at leastDROP_FLOOR_M_S, forHOLD_S, and is the lowest it has been over the lastHOLD_S. After burnout gravity takes about 9.8 m/s off the vertical speed every second, drag more, so the speed keeps falling and the log keeps the coast’s start. A barometer’s reading can dip in the boost and catch up: on the public logs the falls that dips gave recovered within the hold (measured once), and on the synthetic dips a fall that can outlast it (likely the fall back from a catch-up’s overshoot, not traced) had begun to rise again while the motor burned.- Everything after that reads the log cut there: its readings, the fit’s observations and so the fit. The log’s own clock is cut at the cut’s sample, by position, so a time it logs later can’t enter. A channel on a clock of its own is cut at its first time past the cut or out of order: a sample it stamps in order at or before the cut is read wherever the file puts it, as nothing else says when it was logged.
The finder refuses the tracks the analyzer distrusts in the boost: a GNSS receiver’s, and a
barometer’s once its speed reaches readings::TRANSONIC_SPEED_M_S (Mach 0.8 in the coldest
air), where the port may read the transonic flow. Without that rule, and with an earlier rule’s 1 s
hold in the cut, the fitted top speed was over 1.1 times the true one on 57 of set A’s first 200
synthetic flights, the barometer’s jump near Mach 1 read as the top, so the cut’s place followed
the jump rather than burnout (a sweep run once). So a flight whose fitted climb speed reaches
227.7 m/s gets no prediction from its barometer: on that set the slowest refused had a true top
of 221 m/s and the fastest cut a true 233 m/s, its fitted speed under the limit (measured
once). An accelerometer’s cut isn’t built.
A motor whose thrust falls below the rocket’s weight and drag well before it burns out, as a long tail-off does, slows the climb by the fall the cut needs while still burning, and is cut before its burnout; the cut is checked only on the synthetic flights’ motors.
The analyzer’s own burnout reading isn’t used: its spike filter, its burnout mean and its
noise span are centered windows, and readings::read takes the whole log at once.
How far to trust it: the cut and the fit are tested for not reading past the cut; how close the predicted apogee comes to a logged one is not measured yet, on real flights or synthetic ones.
Structs§
- Cut
- Where the boost’s log is cut.
- Prediction
- Apogee predicted from the boost.
Enums§
- NoCut
- Why
find_cutfinds no cut in a log, in the order it decides. - Unpredicted
- Why a log gives no prediction.
Constants§
- APOGEE_
HORIZON_ S - How long the fitted flight is flown to find its apogee, s: past any hobby rocket’s climb.
- DROP_
FLOOR_ M_ S - The least fall below the top speed for the cut, m/s: about a second of gravity alone, so a slow flight’s noise doesn’t cut it during the boost. HPR Sim’s choice.
- DROP_
SHARE - The share of its top the fitted speed must fall below that top for the cut. HPR Sim’s choice:
10%, a margin over the window’s noise on a fast flight’s speed, as a barometer’s noise in height
grows with altitude. Unmeasured: on set A’s first 200 synthetic flights the cut’s test passes on
DROP_FLOOR_M_Salone. - HOLD_S
- How long the fall must hold before the cut, s, and the span its speed must be the lowest of:
a barometer’s fitted speed can fall by
DROP_FLOOR_M_Sand more during the boost and recover. On the public flights’ logs, the longest such fall inside a burn lasted 0.48 s from its first sample to the one it recovered at, and the deepest went 63.2 m/s below its top (Lince and Juno III, measured once in a run not committed); after burnout gravity and drag keep the speed falling. HPR Sim’s choice: about half again the longest, 0.75 s. On set A’s first 200 synthetic flights the cuts came 1.08 s to 2.30 s after burnout, 0.33 s to 1.48 s before this rule. - LIFTOFF_
RISE_ M - How far the window’s fitted line must stand above the pad’s for liftoff, m: past the largest excursion of a raw height from the track’s first on the pad, 6.8 m over sets A and B’s first 200 synthetic flights each (their barometers’ logs, measured once), while a hobby rocket climbs it in about a second or less. HPR Sim’s choice.
- SPEED_
WINDOW_ S - The span of the trailing window the vertical speed is fitted over, s: a straight line by least squares through the samples in it. HPR Sim’s choice: long enough to average a barometer’s noise over about 10 samples at 20 Hz, short against a boost of a few seconds.
- STILL_
M_ S - The fitted speed under which the line stands still, m/s, either way: where it does, the pad is
re-read there, and a liftoff, a top speed and a fall read before it are forgotten. A run of bad
samples on the pad or a step there that stays can lift the line past
LIFTOFF_RISE_Mand read a speed faster than the flight’s own top; the line stands still again once they leave the window, or at the step’s new level (#1362). Bad samples close enough to ignition leave no still window between them and the climb, and can then still be the top and cut early: a known gap (#1441), from 0.2 s before ignition on the test’s flights, not bounded on others. A dip in the boost can stand the line still too, and the pad is then read again in the burn: the dip test still finds every cut after burnout. On sets A and B’s first 200 synthetic flights’ barometer logs with a cut (measured once, not committed), 98.8% and 98.4% of the pad’s windows read under it, no stretch of the pad between windows that did lasted over 0.25 s, and no speed from liftoff to the cut read under 17.6 m/s. A climb whose fitted top speed is under about 22 m/s (this speed, theDROP_FLOOR_M_Sfall andHOLD_Sof gravity’s 7.4 m/s), or a little more where drag slows it too, reaches this speed before its cut and is refused as never lifting off (NoCut::NeverLifted); the walk before this rule cut such slow climbs. HPR Sim’s choice.
Functions§
- find_
cut - Where the boost’s log is cut, decided at each sample of the height track from it and the ones
before it (the module’s rules);
Nonewhere the log holds no height track, its track is a GNSS receiver’s, anySPEED_WINDOW_Sof it holds more samples than the analyzer’s fastest rate allows (readings::MIN_INTERVAL_S), it never lifts off or climbs too slowly (STILL_M_S), it never holdsreadings::SPEED_MIN_SAMPLESin aSPEED_WINDOW_Swindow (a log slower than 8 Hz, or one whose samples can’t be read), its fitted speed reachesreadings::TRANSONIC_SPEED_M_Sbefore the cut, or it never slows byDROP_SHAREof its top speed andDROP_FLOOR_M_SforHOLD_S, at a speed the lowest of the lastHOLD_S. The cut stands where the window’s line is more thanLIFTOFF_RISE_Mabove the pad’s: three bad samples side by side on the pad can lift the line past that for a sample, and the speed then falls back as it does at burnout. The pad is read again wherever the line stands still (STILL_M_S), dropping the top speed and fall read before, so a pad spike faster than the flight’s own top, or a step up on the pad that stays, doesn’t stay the top (#1362), unless it comes too close to ignition for the line to stand still between it and the climb (#1441). - paired_
bootstrap_ interval - The paired bootstrap’s percentile interval on the mean of
differences, atlevel: the paired difference of two methods’ errors on the same flights, resampled by flight, as ADR-257 §3 asks for the private collection’s flights. - predict
- Apogee predicted from
logcut after burnout: the boost fitted byfit::fit, each parameter held to whatstatedgives for the boost’s observations, then flown on byforwardto its apogee. The fit’s ownconvergedsays whether it settled. - sign_
flip_ interval - The exact sign-flip interval on the center of
differences, at a coverage of at leastlevel: the paired difference of two methods’ errors on the same flights, where a bootstrap undercovers on a few (ADR-257 §3). - truncated
logwith only its samples up tocut_s: its own clock’s columns and each channel’s samples up to their first time pastcut_sor out of order, and its events at or beforecut_s(the readings and the fit read none). What it says of itself is kept.- try_cut
find_cut’s cut, or why there is none.