The analysis engines, and what they refuse to answer
SeyAero's occurrence-analysis engines take the facts a case file holds and test each hypothesis against a stated measurement. They cover the phases an investigation actually meets — take-off, climb, cruise, approach, landing, ground roll and the impact site — and the occurrence classes that go with them: controlled-flight upset, high-energy impact, thrust loss, flight-control and trim malfunction, structural failure, landing overrun, obstacle and building strike, midair conflict, rotorcraft events, ditching, in-flight fire and icing.
Everything below is decision support for an accredited investigation. The engines discipline evidence, rank hypotheses, expose gaps and reject invalid counterfactuals. They do not conduct the investigation, and they do not stand in for the investigator-in-charge under ICAO Annex 13.
What each engine does
01
Physics matrix
Every hypothesis meets a measurement.
— Energy balance, specific excess power, V-n envelope, stall margin, Mach and VMO exceedance.
— Phase coverage from take-off roll through initial climb, cruise, approach, landing and ground roll — a landing overrun is analysed as a landing, not as an aborted cruise.
— Thrust availability distinguishes total power loss from an asymmetric failure, and withholds a verdict when an oscillating flight path makes the energy apex unreadable.
— Aircraft envelopes for common transport classes; an unlisted type is solved on a parametric synthetic envelope that is labelled as such on the result, never silently substituted.
02
Counterfactual simulator
Would the intervention have worked?
— Thrust restored, drag cleaned, bank limited, speed traded and immediate escape, each flown from the recorded state.
— Results are bounded by two physical ceilings: the class climb rate, and the energy the aeroplane actually held plus the energy the intervention added.
— A survivable-window scan reports the last instant at which the intervention still avoided ground contact.
— Interventions that do not apply to the recorded regime — an energy counterfactual against a pitch-driven dive, a low-energy trial on a high-energy event — are refused with the reason stated.
03
Sequence, sufficiency and gap radar
What is missing, and what it is worth.
— Causal sequencing with antecedent vocabulary for flight control, trim, air data, ATC and TCAS events.
— Evidence sufficiency scored per conclusion against ICAO Annex 13 reporting needs, with the specific holding that would move it most.
— Analytical results are reported as analysis, never as evidence: a reconstructed chain never lifts a sufficiency score in place of the recorder it stands in for.
— Gap radar ranks unexamined themes by the class of evidence that would close them.
04
Wreckage, structure and survivability
The site as a measurement.
— High-energy impact solving: TNT equivalence, crater scale, main-field and ejecta radii, fragmentation index and in-flight separation.
— Ground-roll and structure-strike frames for overruns, embankment and building impacts, including ground casualties.
— Survivability curves from deceleration, cabin and section integrity, exits, fire onset, rescue time and environment — reported as an uncertainty band, re-solved rather than widened.
— No survivor is invented where no structural pocket exists, and none is rounded away where one does.
05
Midair conflict geometry
Two aircraft, one solution.
— Closest point of approach, closure rate, separation-minima breach and warning time available.
— Advisory compliance judged from the advisory instant where the time is known, and stated as a trailing-window inference where it is not.
— Vertical and lateral avoidance counterfactuals, so a heading change can be tested and not only a climb or descent.
06
Hazard frames
Beyond fixed-wing energy.
— Rotorcraft: autorotation energy, height–velocity exposure and vortex-ring conditions.
— Ditching: water-entry loads, hull integrity and flotation time.
— In-flight fire: progression rate, containment and diversion time available.
— Icing: accretion rate, performance penalty and certification-envelope exceedance.
How the coverage was tested
Each engine change is exercised against a blind regression set built from public observables only — broadcast track, radiotelephony, aerodrome geometry, site and recovery description, weather, occupancy and survivor counts. Published findings, probable causes and system narratives are withheld from the run and used only to score it afterwards. The set spans distinct occurrence classes, because an engine tuned on one accident is not an investigation tool.
Dual thrust loss ranked highest, single-engine loss contradicted, one survivor of the occupants aboard.
Midair collision in controlled airspace
CPA and closure geometry, advisory compliance, dual-field wreckage
Zero separation at CPA resolved, opposite advisory movement identified, warning time quantified.
Unlisted airframe type
Envelope provenance
Synthetic envelope used and declared on the face of the result.
A subset of the same regression set runs inside the product, so a licensed investigator can re-run it from a case file and read the checks rather than take this page's word for it. Where a blind run has been compared against a published official report, the comparison is kept with the run, including the points on which the engine was wrong.
What the engines will not do
Probable cause. The engines rank and exclude; they do not state cause, and every run so far has returned that conclusion as unsupportable on public observables alone.
Measurements the data cannot carry. A test that would need recorder parameters the case does not hold returns insufficient, not a weak verdict.
Interventions outside the recorded regime, which are refused with the reason attached rather than answered optimistically.
Interviews, wreckage metallurgy, recorder read-out, human-factors interviewing and the judgement of the investigator-in-charge. None of these are modelled, and none are replaced.
Stated limits
— Aircraft envelopes are a finite library. An unlisted type is solved on a synthetic envelope derived from its class and mass, and the result says so.
— Site facts — fuel state, occupancy distribution, obstacle geometry, exit usability — remain investigator inputs. Where a track can supply them they are derived and labelled; where it cannot, they are listed as outstanding.
— Survivability is a band, not a count, and the band is the honest answer at this stage of an investigation.
— Rotorcraft, ditching, fire and icing are modelled as hazard frames with explicit uncertainty; they are not a substitute for type-specific certification analysis.
