Cap readRange's allocation and reject size_t overflow instead of truncating;
re-append .rsbank when the export picker omits it; add cafe coverage for
writeFileExclusive and writeLandedFile; loop write() on EINTR.
Both pickers now ride GetUserFileName (mode 0/1); the "no save picker" premise was false.
Landing uses O_EXCL so the create is the existence check, not a TOCTOU pair.
Replace the self-comparing render-window loop with a genuinely discriminating
floor-vs-exact check; correct two stale claims; mark the Auto/Manual floor-parity
premise as unverified; drop the STARTPOS/ENDPOS comment's circular justification.
The millisecond floor lives in the custom-bounds field, not the engine, so
RENDER_BOUNDSFLAG=2 is now the only bounds mode: the two-position type, the
console verdict and the STARTPOS/ENDPOS drift probe all go. capture.cpp 697 -> 622.
The SDK's own single-component sample drains inputParameterChanges in
process() and implements setParamNormalized; automation was reading the
GUI channel alone. The audio thread now patches a block it solely owns.
Grid-ness of an edge was a proxy for "no floor could explain this count," not
the test itself — equal remainders on both edges cancel under a full floor. Now
checks all three floored models directly and corrects the SHORT/LONG
floor-signature docs.
Verdict can no longer print a false EXACT on an on-grid end, no longer names a
bounds channel a content-derived render never consulted, and the grid-align doc
premise is corrected without implementing it.
Corrects a false comment example, fixes two tests that couldn't detect their
own regressions, adds two more read-back checkpoints around Main_OnCommand so
a drift report self-locates, guards a spurious zero-vs-zero coincidence match,
and softens two sentences that overclaimed cause or defect.
No truncation exists on our side of that boundary, so the read-back is the
only evidence available for whether REAPER kept the window — and it fires on
every tail mode, where only None was ever judged.
setInstrumentParams now arms a sticky pending restart that flushLatencyRestart drains from the sync tick; setState and the bake's adopt flush at their own tails.
Row membership is now the group's own property, not a wrap outcome. FILTER's Band|Notch moves to its caption slack, which is what makes the sound row fit. MASTER gains the limiter toggle, the output meter and the GR lamp.
hasLoop becomes user-owned with the gestures as shortcuts onto it; no format change. START uses overlay/trace, not accent/primary, which is the waveform's own fill.
Retire the effectiveGain blend so the meter's minimum tracks smoothGain's own
reduction against real input, unscaled by the transition mute — a toggle over
quiet material now reads no reduction instead of pinning to 0.
Fold setLimiterEnabled's restart request into setInstrumentParams so every writer
keeps the host's latency report in sync. Pin the window-sizing identity, drop the
per-sample modulo, tighten the ceiling tolerance, publish the blended gain.
New pure curve_tessellate joins the overlay's node vertices through curveMap,
one sample per pixel column; the knot no longer floats off its own trace.
The dedup path claimed it directly, so a key could be answered Ok pointing at an entry a failed persist never wrote. Write-back is now guarded and three-valued, and landing one request split out of the scan pass.
SetProjExtState's return covers the whole extname, so it never saw one key. The persist verdict now reaches the report, and a throw mid-write no longer claims the landing left nothing behind.
Splits Ignore into unreadable vs not-a-request and counts every verdict; the report prints only when the pass answered nobody, so its absence proves the action never ran.
Batch captures now name the retained-render folder once instead of nothing; Auto/Manual tail modes refuse a 0-byte render like None does; VERIFICATION.md steps 1-3 no longer invite a false conclusion; docs/comments no longer overclaim.
The one-frame bound is not provable for a per-edge renderer; the test now shows where it breaks. Refused renders move out of the bank instead of being deleted, so the DAW experiment has something to read.
Rewords Cleared/Unanswered/ForeignRequest to name only observed facts and every live hypothesis instead of picking a winner; adds a guarded answeredOutcome accessor against a future unhandled BakeAnswerKind; folds an unreachable BakeScanKey/Context test state; fixes generation-stamp timing.
A no-answer stays a failure; it now says whether the extension never ran the landing, answered a stale generation, spoke a wire this build cannot read, cleared the request, or refused it.
The tolerance is unchanged and now derived, not assumed: frameCountFor lands in {floor(L), ceil(L)}, so a non-frame-aligned window can never miss by more than a frame. Naming the source is what tells a self-bounding render from a short one.
Corrected the Render-dialog source name the refusal's evidence depends on, added a by-ear content check and a file-size channel proxy, and stopped two comments from overclaiming.
daniel