737 down in the Arabian Sea.

[croyde]

Saw on-line yesterday when it first disappeared off Flight Radar. Lost height pretty rapidly.

Cargo flight from the UAE to Pakistan, 5 onboard

Wreckage found today.

Surprised that there wasn't a thread already as this is the first place I look when there's been an incident.

https://www.theguardian.com/world/2026/jul/08/boei...

[gotoPzero]

Its a strange one, the aircraft was in the cruise.

Lots of possibilities but I did note it was on a ferry flight back from maintenance...

[BrettMRC]

Some of the reports state "erratic changes in altitude" prior to it going down.

[Nuclearsquash2]

I used ai to analyse the granular data posted here. https://www.flightradar24.com/blog/flight-tracking...

to build out a possible conclusion. I think it seems pretty plausible.

Claude Fable 5 interpretation of events...

1. KTA1732 / AP-BOI — Investigator's Initial Assessment

• Basis:** granular Mode-S/ADS-B data (8,714 records, 13:23–16:22 UTC, 7 Jul 2026), correlated with public reporting. This is a preliminary read pending recovery of the FDR/CVR — everything below is working hypothesis, not finding.

1. What is now confirmed externally

The occurrence flight was a scheduled overnight cargo service from Sharjah to Karachi operated by K2 Airways' sole aircraft, a 27-year-old Boeing 737-4M0 freighter (AP-BOI) powered by CFM56-3C1 engines — an airframe that began life with Aeroflot in 1999, passed through Garuda, TNT, ASL and FedEx, was converted to a freighter in 2012, and was leased to K2 in October 2024. Five crew were aboard. Critically, the crew reported a navigational system malfunction at approximately 21:18 local (16:18 UTC); controllers provided navigational guidance, and three minutes later radar showed a rapid descent and significant heading change before contact was lost. Wreckage has been recovered approximately 53 NM south of Ormara. Reporting also notes the aircraft experienced GNSS interference affecting ADS-B reception earlier in the flight, a problem also affecting other aircraft in the region.

Both externally reported facts are independently verifiable in your granular file, which gives me confidence in the dataset — and the data adds considerable detail the public record doesn't have.

1. The evidential timeline in the data

• GNSS degradation is real and prolonged.** The navigation integrity category (NIC) collapses to 0 for extended periods (roughly 13:50–14:10 and again 15:00–15:50 UTC, i.e., through departure and initial cruise near the Gulf of Oman corridor), with position availability dropping to 14–25% of messages. This corridor is a well-documented GNSS jamming/spoofing environment. By 16:00 the aircraft was clear — NIC back to 8, position availability ~80–87% — meaning the *final sequence itself is well-observed data, not a jamming artifact*.

• 16:17:03–16:17:24 — the heading source fails before the aircraft visibly deviates.** This is the single most important finding. The transmitted magnetic heading freezes at 082° and reported roll freezes at exactly 0.0° while the ground track begins creeping right (082→088°). Simultaneously, the aircraft-computed wind — which is derived from the heading/TAS vs track/GS triangle — inflates from a steady 082°/6 kt to a phantom 004°/32 kt in ten seconds. That wind spike is precisely the mathematical signature of a *frozen or erroneous heading input feeding a live wind computation on a genuinely turning aircraft*. At 16:17:24 the heading and roll registers cease transmitting altogether, while air-data-sourced parameters (IAS, TAS, Mach, from the ADC) continue for another three minutes. The selectivity matters: the transponder and air data were healthy; the *inertial/heading source* was not.

• Concurrently, altitude-hold quality collapses.** Altitude standard deviation goes from 12 ft (16:00–16:15, textbook autopilot altitude hold) to 146 ft with ±1,400 fpm phugoid-like oscillation after 16:17:20. On the 737 Classic the autopilot depends on a valid IRS; this is consistent with autopilot disconnection or degradation at the moment the inertial data failed. The MCP heading bug remained at 080° — the autopilot was never *commanding* the turn — and the MCP altitude stayed at 35,000 ft throughout.

• 16:18:27 — the heading bug moves 080→085.** The crew was alive, powered, and interacting with the mode control panel, plausibly attempting to reselect heading or re-engage automation, while the aircraft was already ~40° right of the selected heading. This coincides almost exactly with the crew's reported radio call about a navigation malfunction.

• 16:18:50–16:19:20 — graveyard spiral.** The right turn tightens continuously (track 128→165→227→266°), descent rate reaches −15,000 fpm, and speed builds to 506 kt GS / Mach 0.80 — essentially at the Classic's MMO of 0.82. The *gradual then accelerating* divergence over ~90 seconds is the classic signature of an unnoticed, slowly increasing bank angle — not of an abrupt control hardover.

• 16:19:20–16:20:00 — recovery attempt, then stall.** A sustained pull-up of roughly 2.1g (smoothed estimate from the VS reversal, −15,000 to +17,800 fpm) converts the dive into a zoom climb to 36,675 ft, during which IAS bleeds from ~330 kt to **117 kt** — far below any flyable speed at that altitude. GPS-minus-baro altitude at the apogee (+2,406 ft) matches the cruise baseline (+2,445 ft ± 48), so the barometric data is trustworthy: the aircraft genuinely stalled at the top.

• 16:20:00 onward — loss of control to impact.** Final descent reaching −22,400 fpm, last return at 1,100 ft. The squawk remained 3442 for the entire flight; no emergency code was ever set.

1. Comparable accidents and initial reasoning

The profile maps with striking fidelity onto **Adam Air 574** (737-4Q8, 1 January 2007, Makassar Strait): same type and generation, night over water, crew preoccupied with troubleshooting a malfunctioning Inertial Reference System, autopilot disconnected during the troubleshooting, an unnoticed slow right roll developed, and the aircraft entered a spiral dive approaching Mach limits; an aggressive pull-up (~3.5g there) followed before impact/breakup at sea. The KTA1732 data reproduces nearly every element of that chain, including the detail that the inertial data stream failed *before* the deviation became apparent — in Adam Air 574 the crew's own switching of the IRS mode selector dropped attitude information and the autopilot simultaneously.

Secondary analogues reinforce individual links in the chain: **Flash Airlines 604** and **Kenya Airways 507** (both 737s, both slowly developing right banks into spiral dives at night with no visible horizon — spatial disorientation); and **Indonesia AirAsia 8501**, where fault troubleshooting led to autopilot loss, an upset, a zoom climb, and an unrecovered stall — the same end-game seen here at the apogee.

Hypotheses I would currently rank *lower*: a 737 rudder PCU hardover (UA585/USAir 427 pattern) produces an abrupt upset, not a 90-second gentle divergence; an intentional act (SilkAir 185 / MU5735 pattern) is inconsistent with the crew's malfunction report to ATC, the MCP interaction mid-event, and the clear, vigorous recovery attempt; weather appears benign in the data (OAT steady at −34 to −36°C, light winds until the artifact spike; no convective temperature signature), though July is monsoon season and satellite imagery for the area should still be pulled; and structural failure as an *initiating* event is unsupported, though overstress *during* the 2g+ pull at near-MMO must be examined on recovered wreckage — Adam Air is again the precedent.

1. Initial causal reasoning (subject to FDR/CVR)

The most probable sequence: **failure or corruption of an inertial reference/heading source in cruise** — possibly conditioned by the prolonged GNSS jamming exposure earlier in the flight degrading the hybrid nav solution on this elderly, thrice-converted airframe, or a discrete IRS fault the crew then troubleshot — leading to **autopilot disconnection or degradation around 16:17**, followed by **an unnoticed, slowly increasing right bank at night over a featureless sea** while the crew's attention was on the navigation problem and the radio, developing into **a graveyard spiral to near-MMO**, a **late but genuine recovery attempt** whose 2g pull was converted into a zoom climb that **bled the energy to a full aerodynamic stall at 36,675 ft**, from which recovery was not achieved. In ICAO taxonomy: **LOC-I, with contributing factors in system malfunction (navigation/inertial), spatial disorientation, and startle/upset recovery**.

Priority actions I'd recommend: recover FDR/CVR (the data suggests they'll be within a tight radius of 24.28°N 64.43°E); obtain the IRS/AHRS maintenance history for AP-BOI, especially since the October 2024 lease transfer and any deferred-defect entries; quantify the GNSS interference exposure against other traffic on the airway that night; examine wreckage for pre-impact structural overstress from the pull-up; and review K2's UPRT (upset recovery) training records — a single-aircraft operator's training pipeline will be an area of interest.

One honest caveat to carry throughout: this is all surveillance-derived data. The frozen heading/roll could in principle be a downlink decoding issue rather than an aircraft-side failure — but the internal consistency of three independent signatures (frozen heading, phantom wind growth, and simultaneous altitude-hold degradation) inside the same 20-second window makes a genuine inertial-source failure the far stronger reading. The FDR will settle it.

[BrettMRC]

If it did a zoom climb to 36,000ft after recovering from the initial dive, why not level out while airspeed was still adequate etc?

[Mabbs9]

Jamming is very normal now sadly. We had similar last night. It feels like a leap to have it as the critical factor to me.

[croyde]

There were two pilots and two engineers on the flight, plus another chap.

Just seems crazy to not have one person concentrating on flying whilst the others troubleshoot. But I'm not a pilot.

[48k]

croyde said:

There were two pilots and two engineers on the flight, plus another chap.

The "another chap" was a Load Master.

[croyde]

Yes, I couldn't remember his exact title