This is the analysis-engine corpus. It is separate from the Hull-Loss Localisation Validation Corpus, which contains 37 recovered cases and 185 seeded runs. The two records test different systems and their totals are not combined.
31/33
4/6
0d1b9c163b39ed122c6120ec299558aab9b50265c1c9fc294475187966a76859
recomputed Sun, 11 Oct 2026 01:28:36 GMT
A pass here is not agreement with the official conclusion. The checks assert evidential discipline: that the tool declines to state a probable cause, that it ranks only what the physics supports, that a synthesised performance envelope is declared as synthesised, and that any band it reports is wide enough to contain what actually happened. A run that named a cause would be recorded as a failure. Formal determination of cause remains with the State conducting the investigation under ICAO Annex 13.
Machine-readable results: /api/public/blind-suite
Cruise upset to near-vertical descent
5/6 checks heldBroadcast track shows a cruising narrowbody leaving level flight and descending at a rate no atmospheric disturbance produces, with a brief arrest and a second descent. Nothing but the track, the type and the site are supplied.
- 0.535 · Flight control or trim system malfunction (Flight control / trim system malfunction, consistent)
- 0.335 · Aerodynamic stall and loss of control (Aerodynamic stall / high angle of attack, consistent)
- 0.219 · Loss of thrust on both powerplants (Powerplant failure / thrust loss, consistent)
- 0.208 · In-flight break-up from structural failure (In-flight break-up / structural disintegration, consistent)
- 0.208 · Severe weather encounter or turbulence upset (Wind shear / gust-induced structural failure, consistent)
b737-800 — Boeing 737-800 class (narrowbody)
- HELD No hypothesis is elevated to a probable cause on physics alone — the highest posterior stays below certainty.observed: highest posterior 0.535
- HELD The best-supported hypothesis belongs to a family the physics actually supports (Flight control / trim system malfunction, High-energy dive / near-vertical descent).observed: Flight control / trim system malfunction — Flight control or trim system malfunction
- FAILED "Aerodynamic stall and loss of control" is contradicted by the recorded state and loses confidence.observed: prior 0.20 → posterior 0.335 (consistent)
- HELD Where the type is outside the modelled library, the envelope is flagged as synthesised and class-typical rather than presented as type-specific.observed: modelled envelope b737-800
- HELD "MU5735" is resolved out of free text as an identifier.observed: B-1791, MU5735
- HELD "B-1791" is resolved out of free text as an identifier.observed: B-1791, MU5735
Repeating pitch excursions after departure
7/8 checks heldBroadcast altitude oscillates through a repeating cycle for half a minute after departure before a high-speed water impact. Only the track, type, weather and recovery survey are supplied.
- 0.317 · Uncommanded flight control or trim input (Flight control / trim system malfunction, consistent)
- 0.317 · Air data or instrument system fault (Flight control / trim system malfunction, consistent)
- 0.200 · Loss of thrust on both powerplants (Powerplant failure / thrust loss, inconclusive)
- 0.200 · Deliberate act or unlawful interference (Deliberate interference — intentional act by a person on board, insufficient)
- 0.115 · Aerodynamic stall and loss of control (Aerodynamic stall / high angle of attack, contradicted)
b737-max — Boeing 737 MAX 8 class (narrowbody, high-bypass re-engined)
0–1 survivors (point estimate 0); recorded outcome 0
- HELD No hypothesis is elevated to a probable cause on physics alone — the highest posterior stays below certainty.observed: highest posterior 0.317
- HELD The best-supported hypothesis belongs to a family the physics actually supports (Flight control / trim system malfunction, Air data / instrument system fault).observed: Flight control / trim system malfunction — Uncommanded flight control or trim input
- HELD "Aerodynamic stall and loss of control" is contradicted by the recorded state and loses confidence.observed: prior 0.20 → posterior 0.115 (contradicted)
- FAILED "Deliberate act or unlawful interference" is contradicted by the recorded state and loses confidence.observed: prior 0.20 → posterior 0.200 (insufficient)
- HELD Where the type is outside the modelled library, the envelope is flagged as synthesised and class-typical rather than presented as type-specific.observed: modelled envelope b737-max
- HELD The recorded number of survivors falls inside the band the model reports, so the figure quoted to an investigator is not falsely precise.observed: model 0–1 (point 0), actual 0
- HELD "JT610" is resolved out of free text as an identifier.observed: PK-LQP, JT610
- HELD "PK-LQP" is resolved out of free text as an identifier.observed: PK-LQP, JT610
Arrival into buildings shortly after lift-off
6/6 checks heldA widebody lifts off, reaches roughly 600 ft, decays in speed and descends into a built-up area 1.5 km beyond the departure end. Track, type, aerodrome weather, site survey and occupancy only.
- 0.331 · Loss of thrust on both powerplants after lift-off (Take-off / initial-climb performance shortfall, consistent)
- 0.331 · Incorrect take-off performance or configuration (Take-off / initial-climb performance shortfall, consistent)
- 0.208 · Aerodynamic stall and loss of control (Aerodynamic stall / high angle of attack, consistent)
- 0.208 · Weather or windshear performance shortfall (Wind shear / gust-induced structural failure, consistent)
- 0.123 · Loss of thrust on one powerplant with asymmetric handling (Powerplant failure / thrust loss, contradicted)
b787 — Boeing 787 class (composite widebody twin)
0–1 survivors (point estimate 1); recorded outcome 1
- HELD No hypothesis is elevated to a probable cause on physics alone — the highest posterior stays below certainty.observed: highest posterior 0.331
- HELD The best-supported hypothesis belongs to a family the physics actually supports (Powerplant failure / thrust loss, Take-off / initial-climb performance shortfall).observed: Take-off / initial-climb performance shortfall — Loss of thrust on both powerplants after lift-off
- HELD Where the type is outside the modelled library, the envelope is flagged as synthesised and class-typical rather than presented as type-specific.observed: modelled envelope b787
- HELD The recorded number of survivors falls inside the band the model reports, so the figure quoted to an investigator is not falsely precise.observed: model 0–1 (point 1), actual 1
- HELD "AI171" is resolved out of free text as an identifier.observed: VT-ANB, AI171
- HELD "VT-ANB" is resolved out of free text as an identifier.observed: VT-ANB, AI171
Rotorcraft descent into terrain in reduced visibility
6/6 checks heldA medium twin helicopter climbs, levels, then descends into rising terrain at a rate far above its own induced velocity. Only the track, the type and the reported visibility are supplied — no witness account, no crew history and no finding.
- 0.200 · Descent into terrain with loss of visual reference (Flight into terrain or obstacle / loss of visual reference, inconclusive)
- 0.200 · Main rotor or tail rotor drive failure (Rotorcraft dynamics — rotor energy, autorotation, vortex ring, inconclusive)
- 0.200 · Loss of engine power on both engines (Powerplant failure / thrust loss, inconclusive)
- 0.107 · Vortex ring state or settling with power during the descent (Rotorcraft dynamics — rotor energy, autorotation, vortex ring, contradicted)
synth-rotorcraft-medium — Sikorsky S-76B — modelled as Medium twin rotorcraft class — synthesised, class-typical
0–0 survivors (point estimate 0); recorded outcome 0
- HELD No hypothesis is elevated to a probable cause on physics alone — the highest posterior stays below certainty.observed: highest posterior 0.200
- HELD The best-supported hypothesis belongs to a family the physics actually supports (Flight into terrain or obstacle / loss of visual reference, Rotorcraft dynamics — rotor energy, autorotation, vortex ring, Powerplant failure / thrust loss).observed: Flight into terrain or obstacle / loss of visual reference — Descent into terrain with loss of visual reference
- HELD "Vortex ring state or settling with power during the descent" is contradicted by the recorded state and loses confidence.observed: prior 0.20 → posterior 0.107 (contradicted)
- HELD Where the type is outside the modelled library, the envelope is flagged as synthesised and class-typical rather than presented as type-specific.observed: synthetic: Sikorsky S-76B is not in the modelled envelope library. It has been resolved to the medium twin rotorcraft class and built parametrically: 5,500 kg gross at about 30 kg/m² disc loading, VNE 155 kt, limit load 3.5 g. This is a rotary-wing airframe: stall speed, the V-n envelope and wing loading do not describe it, and the fixed-wing tests are gated off. The rotorcraft frame — rotor energy, height–velocity, vortex ring — carries the tests that apply, and it needs the type's true gross mass and rotor diameter. Confirm both against the type certificate data sheet before any figure derived here is quoted in a report.
- HELD The recorded number of survivors falls inside the band the model reports, so the figure quoted to an investigator is not falsely precise.observed: model 0–0 (point 0), actual 0
- HELD "N72EX" is resolved out of free text as an identifier.observed: N72EX, S-76B
Twin turboprop asymmetry after lift-off
5/5 checks heldA regional turboprop climbs to about 1,600 ft, stops climbing and descends in a left roll. Only the track, the type and the recovered propeller condition are supplied — no crew action, no system finding and no conclusion.
- 0.860 · Loss of thrust on one powerplant with asymmetric handling (Propeller dynamics — feathering, asymmetric control, prop overspeed, consistent)
- 0.492 · Propeller failure to feather leaving windmilling drag (Propeller dynamics — feathering, asymmetric control, prop overspeed, consistent)
- 0.260 · Aerodynamic stall and loss of control (Aerodynamic stall / high angle of attack, consistent)
- 0.200 · Propeller overspeed or pitch-control malfunction (Propeller dynamics — feathering, asymmetric control, prop overspeed, inconclusive)
atr72 — ATR 72 / regional turboprop
- HELD No hypothesis is elevated to a probable cause on physics alone — the highest posterior stays below certainty.observed: highest posterior 0.860
- HELD The best-supported hypothesis belongs to a family the physics actually supports (Propeller dynamics — feathering, asymmetric control, prop overspeed).observed: Propeller dynamics — feathering, asymmetric control, prop overspeed — Loss of thrust on one powerplant with asymmetric handling
- HELD Where the type is outside the modelled library, the envelope is flagged as synthesised and class-typical rather than presented as type-specific.observed: modelled envelope atr72
- HELD "B-22816" is resolved out of free text as an identifier.observed: B-22816, GE235
- HELD "GE235" is resolved out of free text as an identifier.observed: B-22816, GE235
Unlisted type — envelope provenance
2/2 checks heldA type outside the modelled library is analysed deliberately. The suite asserts that the engine says so rather than quietly adjudicating the aircraft on a 737's wing loading.
- 0.291 · Flight control system malfunction (Flight control / trim system malfunction, consistent)
- 0.115 · Aerodynamic stall and loss of control (Aerodynamic stall / high angle of attack, contradicted)
synth-turboprop — Antonov An-26B — modelled as Regional turboprop class — synthesised, class-typical
- HELD No hypothesis is elevated to a probable cause on physics alone — the highest posterior stays below certainty.observed: highest posterior 0.291
- HELD Where the type is outside the modelled library, the envelope is flagged as synthesised and class-typical rather than presented as type-specific.observed: synthetic: Antonov An-26B is not in the modelled envelope library. It has been resolved to the regional turboprop class and built parametrically: 20,000 kg at 330 kg/m² wing loading (61 m², aspect ratio 12), VMO 250 kt / M 0.55, limit load 2.5 g. Every margin computed from it is class-typical, not type-specific: confirm the placards and the load sheet against the type certificate data sheet before any figure derived here is quoted in a report.