Closing report: the satellite and flight chapter closed 6 Oct 2026

This closes the satcom and flight-model branch of the investigation. The drift branch remains open and is not covered here. Written against the corrected record, after TASK_026 overturned two things this chapter previously asserted.

No part of this investigation located the aircraft. No source in the archive produces a final coordinate, and nothing in TASK_021, TASK_022, TASK_026 or TASK_029 produces one either. Every position in this atlas is a conditional model state. That is the result, stated first rather than qualified at the end.

What was established

1. The cruise model fits, under heavy constraint

C120 fits the nine-term score at χ² = 8.117071 with nine active constraints, including k pinned at its lower bound, start latitude pinned at its 0.5° limit, and a fuel allowance margin of about 0.1 kg. It is a fuel-limited solution, not a free optimum. Within the TASK_021 diagnostic it does not require a slow startup: forcing one costs +0.62 in χ². The C-box variant failed to converge; its flight verified independently but it cannot separate the bound widening from extra optimisation of a non-stationary reference.

2. The endpoint latitude is constrained by the handshake sequence — within the region the model can see

Seven solutions in this family land at 00:11 between 35.85°S and 36.85°S. TASK_029 found that northern commands reach 23.9°S to 26.4°S with fuel to spare, but miss every handshake ring by 108 to 127 km, at 24.8 to 29.2σ. Those northern flights are also still powered at 00:25, so they cannot produce the 00:19 log-on.

Corrected 7 October, and the correction runs the other way. This was previously stated as a physical exclusion of northern endpoints. It is not one, and the sweep meant to confirm it indicated the opposite.

At C120's own 0.5°N start, TASK_030's coarse grid reaches χ² ≈ 1,174, while that same start supports 8.117 — the grid is 145× off where the answer is known. From 6°N the identical grid reaches 112.3, about ten times better, with the optimal Mach pinned at the grid ceiling and the trend still improving where ERA5 stops at 10°N. That is a positive indication of plausible northern arc-1 starts.

It cannot be tested. The grid cannot represent C120's commands, so no fair comparison exists; ERA5 ends at 10°N, so northern starts have no history; and the arc crosses 100°E at 27.5°S, so the endpoints those starts would reach cannot be scored. The large northern BTO residuals are what a flight produces when it cannot go east far enough to meet the arc — not evidence that the aircraft did not.

3. Twelve terminal continuations match the final frequency pair, and none is a location

TASK_022 searched 582 histories from C120’s frozen engine-loss state. Twelve matched the adopted envelopes. All twelve arrive at the water supersonic, at 2.1 to 2.7 times Vmo, banked 76° to 89°, inside the aerodynamic proxy the model itself flags as unsupported. An aircraft in that descent breaks up well above the surface, so debris would not reach those coordinates. The two controls that stay inside the supported aerodynamic domain and arrive at survivable speed miss the 00:19:37 BFO by more than 100 Hz.

4. The compensation convention is published

Holland (arXiv 1702.02432v2), sections IV-A and IV-B, describes the SDU calculation as using ground speed and track, zero vertical speed, nominal satellite position and sea-level aircraft position, with the physical Doppler term separately carrying actual vertical motion. That is the convention in use. The “K-mode OPEN” label concerned historical callers that forced physical up velocity to zero, not this convention.

5. The BTO residuals were never a miss

TASK_022’s −24.98 to −26.69 µs residuals are 0.86 to 0.92σ against the project’s own listed σ_BTO of 29.0 µs. The 1 µs target was 0.03σ, about thirty times tighter than the measurement, and was declared as a numerical search target rather than an acceptance rule. The residual spans only 1.71 µs across all twelve, so it is inherited from the frozen 00:11 state and the terminal law cannot move it.

What was tested and did not hold

The terminal conclusion, as originally stated. “The final BFO pair requires a steep descent and excludes a long controlled glide” was never established by TASK_022 and does not survive TASK_026. The defensible statement is narrower: those twelve selected imposed-force histories matched the envelopes at the published operator, and those two specific controls did not.
The frequency matches are fragile. Sweeping the vertical-compensation fraction κ, all twelve hold to κ = 0.042, the last warm-up case ends at κ = 0.269, and nothing survives past κ = 0.317.
But there is no inversion. R0 and G0 never become compatible anywhere in [0, 1], under any geometry switch. The sensitivity runs from “dives match” to “nothing matches”, never to “glides match”.
The cruise score cannot arbitrate. χ² moves 0.00255 across the whole κ range, with a largest BFO change of 0.116 Hz against σ_BFO of 4.3. Measurably non-flat, and far too small to discriminate.
Northern arc-1 starts look plausible and cannot be tested. Under a grid 145× coarser than the known solution requires, starts at 4–6°N score ten times better than C120's own start, with Mach pinned at the grid ceiling. Whether that survives proper resolution is unknown and currently unknowable.
The arc north of 27°S is outside the model and outside the search. The northern limit of every seabed search box in the project manifest is 27°S; the arc leaves the forcing data at 27.5°S. About 1,460 km of arc, from 27°S to 15°S, falls outside both. The ERA5 subset was evidently clipped to the region the search already assumed, so the model has never been able to challenge the assumption that defined its inputs.
A one-engine fuel hypothesis was retired. The README warns that single-engine burn may be too generous, which would add range and push the endpoint south. The evaluator has no single-engine phase, so the warning does not apply to it.

What it all rests on

Assumption	Status
SDU compensation convention	Published (Holland IV-A, IV-B). Documentary support, not inspected firmware.
BFO bias calibration epoch	MISSING, with a CONFLICT on its locator: the cited reference resolves to a nine-page project packet, not the article page named.
Nominal satellite altitude 36,210,120 m	Inherited detail, no inspected engineering support.
Warm-up offset envelope, 17 to 136 Hz	Permissive enough to absorb raw residuals of 10 to 67 Hz. Nine of the twelve fitted histories depend on it; three need none.
Aerodynamic proxy below the supported polar	Unsupported in exactly the regime the twelve fitted contacts occupy.
Incoming route before 19:41	Unproved. The 18:22 position is derived-approximate from published waypoints; DSTG cautioned on long-range angular error and did not use it quantitatively.
Systems chronology	Continuous feed assumed. Antenna orientation and hydraulic authority unresolved. A feasible mathematical warm-up offset does not establish electrical or thermal history.
Fuel model	One tank feeding both engines; no single-engine phase exists.
TASK_029 coverage	36 points of 3,276. Mid-range pressures never sampled.
Forcing domain	ERA5 ends at 10°N and 100°E. The arc is outside it north of 27.5°S, so that region is unmodellable rather than excluded.
TASK_030 grid	Does not contain C120: its bearing (184.6°), Mach schedule (0.8406→0.8306) and pressure (214.2 hPa) all fall between grid nodes.

Appendix: readings that were wrong, including mine

I called the compensation convention unvalidated. It is published in Holland. The flight page asserted the opposite for a day and has been corrected.
I read “K-mode OPEN” as evidence about that convention. It concerned legacy callers forcing physical up velocity to zero. Different thing entirely.
I wrote that the final BFO pair requires a steep descent and excludes a long glide. TASK_022 never established that, and I should not have summarised it that way.
I predicted the cruise χ² would be flat in κ and had a circularity argument prepared. It is measurably non-flat, and Codex correctly declined to draw the conclusion I had anticipated.
I specified a half-normal kernel that converges to a 100 km indicator. It does not; it converges to a point mass. The verification step built on that was wrong and was replaced.
I fixed the initial bearing and set a Mach floor of 0.70 in the first reachability directive. Either alone would have made a northern endpoint unreachable by construction rather than by physics.
I expected fuel to bind the northern reach. The signals bind; fuel has tonnes to spare.
I then read TASK_030's table as confirming the south. It does the opposite. I compared absolute χ² across starts without first checking what that grid achieves at the one start where the right answer is known — 1,174 against 8.117. With that in view the 112 at 6°N is the headline, and I had buried it. Ryan caught this.
I read the northern BTO residuals as a physical exclusion. They are a forcing-coverage artifact: the arc north of 27.5°S lies outside the weather data, so no flight could reach it regardless of physics. Corrected 7 October.
I predicted the dt 15 s mesh would halve the sweep cost. It saved 8%, and I then inferred a fixed overhead from two benchmarks that turned out to use the same timestep, so their difference measured nothing. The execution agent's diagnostic contradicted both.
I told the execution agent the density field was unsmoothed. That was true of the locator in common.js, not of the field, which TASK_021 Part A confirmed is smoothed upstream. It matters: smoothing pulls a maximum toward broad mass and so biases toward exactly the verdict I had predicted.
I specified a synthetic harvester the archive already recorded as returning zero matches. The strict match rule had already failed for five fixed synthetic sites. A pilot now fails in minutes instead of hours. Both this and the item above were caught by the execution agent’s questions, not by me.
Superseded: the stale Task 9 text on the combined map, corrected in atlas v6.

For whoever picks this up

The BFO bias calibration epoch is the single most consequential unresolved item in this chapter. Everything frequency-based inherits it.
Three histories (S03, S07, S11) need no warm-up offset at all. S03 fits all three final observations within 2σ. If the warm-up envelope is ever tightened on evidence, those three are the ones that survive it.
The twelve contact coordinates must not be propagated as locations, into drift work or anywhere else. The two branches were deliberately kept separate and should stay that way.
A northern endpoint reachable only by a path with turns, step climbs or a speed schedule would not appear in TASK_029 or TASK_030, which tested only the commanded-flight family as implemented.
The single highest-value next step is data, not computation: extending ERA5 east to 120°E and north to 20°N. Until the forcing covers the arc, the northern question cannot be asked — and the sweep that could not ask it still pointed that way.