5.0 KiB
Field-flight operational audit — read-only, before first100m trial
Scope: staged production ERR_AircraftFieldFlight.c, ERR_AircraftFieldRules.c, existing native20m log, and field fixture evidence contract. No engine, display, gameplay, source or fixture changes were made. The following offline calculation executes actual pure rule bodies inside an explicitly idealized reference-following model; it is neither native measurement nor a performance benchmark.
Actionable mismatch: approach travel versus immediate stopping reserve
ERR_AircraftFieldFlight.CheckTranslation enters braking at25m or the speed-derived reserve. BeginFieldBrake freezes the normal landing target at the destination and caps the prior reference speed by0.9. FieldDemand then continues toward that destination using ApproachSpeed(remaining,min(current,cap)). The inverse-reserve schedule correctly includes the1s latency and3m margin, but it intentionally keeps advancing until roughly3m remains; it does not request an immediate stop within the reserve measured at the beginning of that phase.
The fixture's read-only BrakeBegan records that initial reserve and BrakeStopped rejects sampled path or displacement exceeding it. This is a stronger immediate-stopping contract. The two contracts disagree even with instantaneous, perfectly followed velocity references. /tmp/erri_field_idealized_stop.py extracts and compiles the actual Rules bodies, integrates at0.01s with actual vector slew and frozen0.9 cap, and stops after two seconds below0.25m/s. /tmp/erri-field-idealized-stop.txt records its explicitly non-native results:
| Initial actual/reference speed | Declared immediate reserve | Idealized phase travel | Idealized elapsed |
|---|---|---|---|
| 1.0m/s | 7.333m | 21.877m | 28.041s |
| 1.5m/s | 12.000m | 21.877m | 21.780s |
| 2.0m/s | 18.333m | 21.877m | 19.480s |
These values are analytical counterexamples to the joint policy contract, not predictions of actual Mi8 stopping. Native overshoot, lag, autohover and integral history can change them. A successful endpoint arrival does not cure the evidence mismatch.
Smallest next decision: retain the strict fixture and make production explicitly stop within its declared initial reserve (e.g. a genuine immediate zero-reference braking phase), accepting that slow positioning afterward may exceed the unchanged45s phase budget. Alternatively, deliberately design separate bounded approach and stopping phases with the stopping point/reserve declared before the actual stop command, then independently review its measurement contract. Do not merely relabel the current25m point, increase the reserve to the observed travel, or remove the failed stopping criterion after a trial.
Other reviewed behavior and limits
- Field-demand signs and airframe projection follow the existing independently authored route controller: world desired-minus-actual velocity, then projections onto actual forward/right basis and nonnegative opposing action pairs. The20m native run202706 moved along positiveX using this path. I found no source-backed sign inversion. This does not establish stability of the new0.08 cap, integral gain0.02 or higher velocity.
- Integral suppression when saturated and outward, vector0.2m/s² reference slew, native actual-speed2.5m/s refusal, and reset before slow-position/landing are present. The autohover transfer function and physical deceleration remain unidentified. Integral lag/oscillation is a native risk, not a proven defect from source alone.
- The immutable original deadline remains inherited. Owner Tick begins reserve cancellation by86.5s remaining, and braking/settling45s, landing25s and shutdown15s remain individually clamped by that original deadline. This covers the85s allocation only while the expected owner cadence holds; arbitrary scheduling stalls cannot guarantee the reserve. Native callbacks still refuse hard expiry.
- Early airborne POSITIVE/NEGATIVE cancellation now enables and reads back autohover before committing braking. Mid-settle cancellation retains the original point and deadline. No public base Start/StartTranslation shortcut admits a field flight.
- Preflight50 and abort20 native fuel units are explicit candidate thresholds. They are not a measured endurance model. The existing20m native cancellation202706 completed roughly74s after initial control admission and16.67s after serviceStop; this does not establish higher-speed100m fuel or stopping requirements.
- No additional source-proven operational defect was found in this pass. Native geometry/survey refusal, actual control consumption, cruise speed, braking, landing and finite fuel remain pending. The fixed airport corridor may refuse before allocation.
Exact source identities read
- FieldFlight:
e3322b6d85b1234bfbf218a119cf6f7eed0b9ddd160004d2d9fe02c64ed857ba - FieldRules:
395eadae1067f230c26dbf73f43e39180173e173fd04a346839c474e084feaae - NativeFieldProbe:
d2d2dd65e202672f285ed9afb7ee0247da31b8eb29e0fc3091ccb846f49236e8 - Installed/current20m evidence read:
/tmp/erri-workbench-profile/logs/logs_2026-09-19_20-27-06/console.log(read only).