Pure planLaneMinting decides which tracks hold >1 mode's content and which managed
lane each item lands on; view.cpp applies it (I_FREEMODE/I_NUMFIXEDLANES/P_LANENAME/
I_FIXEDLANE) under one undo block, driven off the auto-tag detection tick. Manual lanes
and their items are never touched. Load-time reconcile rebuilds ownership from durable
lane names before reapplying active-mode visibility.
Cut the null-test verify action and pre-FX-dry mechanism (Daniel's
call); keep provenance + re-capture-from-source. Sample.provenance
already exists/round-trips (M1), so M10 populates and consumes it. Add
docs/product/provenance.md with dual-canvas forks P1-P4 and the leaned
bank-only path; rewrite PLAN.md M10 as a proposal pending fork picks.
The detector re-armed its GuidBaseline on a pointer compare (proj != lastProject),
a weaker signal than persist's GUID-primary identity. On a load onto a recycled
ReaProject* the baseline never reset, so the just-loaded project's pre-existing
tracks diffed against the previous project and were mass-tagged into the active
mode — opening a Design-saved project mis-tagged its Arrange tracks. Re-arm now
rides persist's authoritative load signal via bankPanelNotifyProjectLoaded().
markTailDirty() called MarkProjectDirty() but never called
saveToActiveProject(), so the new tail value was never written
to SetProjExtState. On unsaved projects saveToActiveProject()
no-ops cleanly, resolving the companion Major at the same time.
Relocate TailSetting into ReaSamplerSession; persist serializes it to
a new forever-stable "tail_setting" ext-state key (load-per-project,
save-on-save). Footer scroll-wheel adjusts Manual length in 250ms
steps; label now shows "Tail: Manual X.Xs". Pure step/label/JSON
round-trip covered by tests.
Auto/Manual/Off tail handling for realtime track-tap capture. Auto scans
recorded PCM backward for last frame above -72 dB and truncates the wav
header-aware (8 s cap); Manual pads a fixed tail; Off is byte-identical.
wav_trim rejects WAVE_FORMAT_EXTENSIBLE with non-float SubFormat GUID.
Drives fixed-lane C_LANEPLAYS for managed lanes only (name-keyed, ordinal-renumber
safe) in the view shell; diffs live track/item GUIDs on the panel timer to auto-tag
new content to the active mode, first-poll-guarded. Pure guid_diff + lane_keys
modules unit-tested; folds in Wave-1 polish.
Replace the ...-with-tail variant actions with a pure tail_control module +
a docked-panel footer toggle (Off/Auto/Manual) read by CAPTURE_ITEM/TRACK.
Default None (byte-identical); Manual fixed 2s, in-memory session lifetime.
Tail mechanism unchanged; retired the variant ids.
Add a Phase D2 (two-canvas) section to PLAN.md with the open D2-W2
(shell) and D2-W3 (actions, persist, UI) waves in house style. Archive
the landed D2-W1 pure view_mode_model lane extension in COMPLETED.md.
Add TailMode{None,Auto,Manual} + derived -72dB TRIMEND + 8s cap in render_settings;
wire into OfflineRenderBackend with snapshot/restore of the tail settings; add
CAPTURE_ITEM_TAIL/CAPTURE_TRACK_TAIL variant actions (plain actions stay exact-bounds).
Extends the REAPER-free view_mode_model for per-item mode membership via fixed
lanes: lane->mode C_LANEPLAYS mapping, managed-vs-manual lane-ownership index,
managed-only item-lane ops in the toggle planner, "which lanes may this toggle
touch" query, auto-tag decision (manual-lane items exempt), and JSON round-trip
of the lane index. D1 behavior and tests unchanged.
Update CONTEXT.md capture sections to two FX scopes (item/track,
chain-independent, no master, always wet): build-order step 7, offline/realtime
render API surface, the capture module sketch, and mark the realtime
wet-master-routing open question resolved (track-tap).
Update the CLAUDE.md FX-scope invariant to item+track only; note master
removal in COMPLETED.md; drop master references from the tail spec and PLAN
Milestone T; realtime now taps the selected track.
Capture is now item + track only (master removed as a scope; still bypassed
as out-of-scope chain). Realtime records the selected track's own output via
per-track post-fader sends into a hidden temp track, fixing the silent file.
Item-level mode separation via REAPER fixed lanes; new content auto-tags to
the active mode; inactive lanes hidden and silenced; a managed-vs-manual lane
index keeps toggles off user comp lanes; capture placement mode-aware.
Precision invariants unaffected; REAPER 7 floor.
Timer-driven realtime capture behind ICaptureBackend (begin/tick/abort via
OnTimer, non-blocking). Records master into a hidden temp track, moved to the
bank non-destructively with idempotent restore across every terminal path.
Pure phase machine unit-tested. Track/item deferred; realtime is non-deterministic.
Mirrors capture: when membership is non-empty and the project has no .rpp,
Main_SaveProject(proj, true) prompts Save-As before persisting; cancel leaves
the in-session tag state intact and warns via console.
Design View actions mutated the model but never pushed it to ext-state, so
tags/active-mode/snapshots were lost on save/reopen. Add persistViewState()
(saveToActiveProject) after each handler's apply/reapply.
Note the M7 four-mode -> three-FX-scope rework in COMPLETED.md; add the capture
FX-scope precision invariant to CLAUDE.md; record the send/routing isolation TODO
(coloring sends bleed into item captures) in PLAN.md open questions.
Reconcile-on-delete shipped; narrow the remaining deferrals to the FX-GUID
keying hazard and dormant-membership compaction. Drop the resolved
snapshot-durability question.
FxBypassGuard now snapshots/restores D_VOL, D_PAN, D_WIDTH, D_PANLAW, I_PANMODE
alongside I_FXEN for every bypassed track (forcing stereo-pan mode for an
unambiguous render), so Track/Item captures are uncolored by parent/folder/master
fader and routing. All restored on every path.
Add ViewModeModel::reconcile(liveGuids) to prune snapshots of deleted
tracks; wire it into applyMode before planning. Keep membership so
undo-delete (same GUID) preserves the Design tag.
5a: named-banks tab strip is LICE-drawn (matching M5 grid + Design View
segmented switch), with overflow/scroll in scope from B4 start.
5b: move ships as both a bindable "move to bank" menu and drag-between-regions;
copy stays the deliberate secondary act. Drag mis-drop mitigated by drop-target
highlighting and recoverable by design (index-only, reversible).
Phase B now fully settled (all five forks resolved), ready to scope into
implementation waves; only the active-bank indicator placement polish remains.
Replace the four capture modes with item/track/master scope actions. Each infers
its range (razor-else-time) and enforces FX scope via non-destructive
FX-bypass-around-render (RAII I_FXEN snapshot/restore over ancestors + master).
Corrects the defect of items captured through parent FX.
Fold Daniel's decisions into CONTEXT.md, PLAN.md, and the product notes:
- Fork 1 (persistence): pool folds into the `banks` blob; legacy
`bank_index` key retired after a one-way lossless migration.
- Fork 2 (delete): delete drops member index entries; add an `evacuate`
verb (B1 pure op + B3 action) returning members to the pool. Design the
orphaned-until-prune window and confirm-on-non-empty-delete guardrail.
- Fork 3 (move): move is the default gesture, copy the deliberate secondary.
- Fork 4 (active bank): active-bank and shown-tab stay distinct; the
active-bank indicator must be visually unmistakable.
Fork 5 (tab rendering + move affordance) remains open with an options
analysis and pending recommendations. Additive; M0-M11 and Phase D untouched.
Add the multi-bank pillar as Phase B (lettered, parallel to M-line and
Phase D): CONTEXT.md §Multi-bank spec, PLAN.md Phase B block (B1 bank_book
pure / B2 persist / B3 actions / B4 panel split), and docs/product/multi-bank.md
framing. Additive only — BankIndex, the M0–M11 roadmap, and Phase D untouched.
daniel