Feed deep/late-pass decodes to the QSO auto-sequencer

[patrickrb]

Problem

Field report from the 2026-07-03 POTA activation (K1AF): QSOs stalled repeating the previous message while the partner's reply was visible on screen, and some stations calling K1AF were never answered.

The pulled debug.log pins it down: the auto-sequencer only ever saw the fast decode pass. MainViewModel.afterDecode gated parseMessageToFunction on !isDeep, so every deep, subtraction-loop, and late/cross-slot pass (routine since the decode-gap work, #367–#370) reached the UI and QSO panel but never the sequencer.

Evidence from the log (local = UTC−5):

• N2JFD: K1AF N2JFD R-18 received 20:25:15/:45/20:26:15; sequencer logged newOrder=-1 (no-reply 1, 2) for two cycles and only advanced on the third.
• K3VAT: identical pattern for R-23, with the smoking gun 15:35:28 DECODE: kept=24 ... replyToMe=false slot=1 — the fast pass contained no message addressed to us while the QSO panel displayed the R-23 (deep-pass decode).
• 60 no-reply events at orders 2–3 across the day.

Fix

Deep passes now drive the sequencer through an evidence-only parse variant (parseMessageToFunction(list, true)):

• May do: advance the QSO on the partner's reply, trigger the RR73→73 reply, complete on a received 73 (or the partner moving on at RR73), enqueue/answer callers.
• May not do: no-reply counting, give-up, or silence-based completions (order 5 + newOrder −1, RR73 no-reply caps) — those absence-of-evidence calls stay fast-pass-only, so a deep pass that finds nothing new is not a second no-reply. The completion decision is extracted into shouldCompleteQso(...) for testability.

Deep results arriving before TX keys (~:29.0–:29.8 vs key-up ~:29.9) are parsed immediately and can upgrade the very next transmission (send RR73 instead of repeating the report in the same cycle). Results arriving mid-TX are stashed in the new PendingSequencerDecodes (age-capped at 60 s) and replayed at the first delivery after key-up.

Evidence parses log as QSO: cycle(deep) ... so the effect is auditable from a pulled debug.log on the next activation.

Tests

• EvidenceOnlyParseTest (Robolectric, full sequencer instance): deep R-report advances 2→4 and resets the no-reply counter; deep RR73 → 73; deep 73 completes to CQ; deep caller lands in the queue; deep pass with nothing new does not increment no-reply (with the fast-pass contrast case) and does not complete an order-5 QSO; own-slot decodes are rejected; Hound mode ignores deep passes.
• FT8TransmitSignalTest: shouldCompleteQso arms — positive evidence completes on any pass, silence-based arms fast-pass-only, thresholds honoured.
• PendingSequencerDecodesTest: stash/drain/clear, ordering, age eviction (boundary included).
• Full testDebugUnitTest suite green.

Field verification

Next activation: watch for QSOs advancing on the first received R-report and QSO: cycle(deep) lines in debug.log. (Note: two of this outing's stalls were compounded by a separate USB TX-drop bug — fix in a follow-up PR.)

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 20.80%. Comparing base (68398ec) to head (eb7293f).

Additional details and impacted files

@@            Coverage Diff            @@
##                dev     #406   +/-   ##
=========================================
  Coverage     20.80%   20.80%           
  Complexity      133      133           
=========================================
  Files           148      148           
  Lines         19297    19297           
  Branches       2876     2876           
=========================================
  Hits           4015     4015           
  Misses        15115    15115           
  Partials        167      167           

Flag	Coverage Δ
android	12.19% <ø> (ø)
native	9.93% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR updates the FT8 QSO auto-sequencer so that deep/subtraction/late-pass decodes can advance the QSO state machine as positive evidence, fixing stalls where replies were visible in the UI but ignored by sequencing. It also adds a mid-TX “stash and replay after key-up” mechanism and extracts QSO-completion logic for better testability.

Changes:

• Route deep/late-pass decodes into an evidence-only sequencer parse path, and replay mid-TX deep results after key-up.
• Add PendingSequencerDecodes (age-capped) to stash deep decodes while transmitting.
• Extract shouldCompleteQso(...) and add targeted unit/Robolectric coverage for the new sequencing behavior.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
ft8af/app/src/main/java/com/k1af/ft8af/MainViewModel.java	Feeds deep/late-pass decodes to the sequencer (evidence-only), stashes mid-TX deep decodes, and replays them after TX ends.
ft8af/app/src/main/java/com/k1af/ft8af/PendingSequencerDecodes.java	New synchronized stash/drain helper with MAX_AGE eviction for mid-TX deep-pass decodes.
ft8af/app/src/main/java/com/k1af/ft8af/ft8transmit/FT8TransmitSignal.java	Adds evidence-only parse variant, extracts shouldCompleteQso, and exposes package-visible hooks for testing.
ft8af/app/src/test/java/com/k1af/ft8af/PendingSequencerDecodesTest.java	Unit coverage for stash/drain ordering and age-based eviction.
ft8af/app/src/test/java/com/k1af/ft8af/ft8transmit/FT8TransmitSignalTest.java	Adds unit tests for shouldCompleteQso behavior across evidence-only vs fast-pass paths.
ft8af/app/src/test/java/com/k1af/ft8af/ft8transmit/EvidenceOnlyParseTest.java	Robolectric coverage ensuring deep-pass evidence advances sequencing without triggering absence-of-evidence behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.