[patrickrb]

Why

Reported: "my Delta time indicator will say it's 2.5 seconds off or 1.5 or whatever even though it seems like it's working stations perfectly. It makes the app look unstable and bad."

It was. The clock-sync pill showed the raw arithmetic mean of a slot's decode DTs, re-posted on every decode pass:

• A mean has no defence against one bad decode. DT is per-station — a station transmitting late, or a marginal decode the searcher placed a couple of seconds off, carries its own big DT. Three good decodes at +0.1 s plus one at +7 s average to +1.8 s: a red "clock off" verdict produced entirely by somebody else's clock.
• Every pass re-posted. A slot is decoded several times (fast pass, then deep/late passes that dig out the weakest signals — exactly the decodes with the sloppiest DT), so the number jumped several times within one 15 s slot.
• No slot-to-slot smoothing, so even clean data bounced between a 2-decode slot and a 9-decode one.

The kicker: ClockSelfSync, the estimator that actually corrects the clock, has always been robust (median + MAD rejection + multi-slot confirmation). So the display and the correction openly disagreed — the pill flashed "2.5 s off" while auto-sync, looking at the same decodes, correctly concluded there was nothing to do.

What changed

New DecodeDtDisplay (pure Java, next to ClockSelfSync, reusing its median/MAD helpers so both paths measure the same way):

	before	after
per slot	arithmetic mean	MAD-rejected median
multiple passes on one slot	each replaced the reading	accumulate into that slot, refining its median
across slots	none	median of the last 3 slots
tiny changes	re-posted every time	held below a 0.05 s deadband
after a clock correction	averaged across it	window dropped, reading re-acquires

A real, sustained error is still followed within ~2 slots (tested). The manual "suggested correction" in Settings → Time Sync reads the same value, so it gets steadier too.

Also logs DT: raw=+2.31 shown=+0.12 n=5 once per slot — a debug.log now shows directly whether a wandering raw mean is being smoothed or a clock is genuinely drifting.

Verification

• Full testDebugUnitTest suite — pass. DecodeDtDisplayTest adds 12 cases, including the exact reported shape (three in-time decodes + one station 7 s late must not move the reading) and the converse (a genuine +1 s drift must still be followed).
• Built and installed on an emulator; app runs clean.

Not verified against live signals — the emulator decodes nothing and no rig/phone was attached to this machine. Worth a session on the air before merging: the DT: raw=… shown=… lines make it easy to confirm from debug.log.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Lbd9brnZDMCzipgsu9n8Xg

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 41.50%. Comparing base (c603d1c) to head (d75b212).

Additional details and impacted files

@@            Coverage Diff            @@
##                dev     #915   +/-   ##
=========================================
  Coverage     41.50%   41.50%           
  Complexity      252      252           
=========================================
  Files           280      280           
  Lines         33973    33973           
  Branches       4033     4033           
=========================================
  Hits          14100    14100           
  Misses        19584    19584           
  Partials        289      289           

Flag	Coverage Δ
android	17.59% <ø> (ø)
native	9.93% <ø> (ø)

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

Files with missing lines	Coverage Δ
...app/src/main/kotlin/radio/ks3ckc/ft8af/FT8AFApp.kt	0.37% <ø> (ø)
...tlin/radio/ks3ckc/ft8af/ui/components/ClockSync.kt	25.31% <ø> (ø)
...n/radio/ks3ckc/ft8af/ui/settings/TimeCorrection.kt	100.00% <ø> (ø)
...radio/ks3ckc/ft8af/ui/settings/TimeSyncSettings.kt	0.00% <ø> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.