WIP: iterative buildCells rewrite

WIP.md

@@ -0,0 +1,33 @@
+# WIP: Iterative buildCells rewrite
+
+## What's done
+- Block has `weight uint64`, `triggers []*Trigger`, `topRow int` fields
+- TriggerSource and TriggerTarget have `block *Block` pointer fields
+- `linkTriggers()` resolves string IDs to block pointers, populates `block.triggers`
+- `computeWeights()` sets weight = `(cue index) << 32` + 0 if cue-ended, 1 otherwise
+  - Higher weight = more likely to be moved
+  - Later cues = higher weight
+- `buildCells()` is rewritten to be iterative:
+  1. Init: cues get sequential topRow, non-cues start at 0
+  2. Loop (trigger pass): move target blocks so trigger source row >= target hook row
+  3. Loop (overlap pass): sort same-track blocks by (topRow, weight), push overlaps to prev end + 1
+  4. After convergence: move each cue to max row of its triggered cells
+  5. `emitCells()` creates actual TimelineCell objects from computed topRows
+- Test uses `BuildTimelineDebug` with testWriter for debug output
+
+## What's broken
+- Cue track comes out unordered after the cue-move step (cues placed at their highest triggered cell row, which scrambles scene order)
+- Gap padding in emitCells skips cue track (`if block.Type != "cue"`) so cue cell rows don't match topRow
+- The +1 gap between blocks may be wrong (was +2, changed to +1)
+- assignRows still can't converge after buildCells — off-by-one misalignments between cue rows and target rows
+
+## Key design decisions
+- topRow coordinate space should include gap cells (emitCells just places cells at topRow)
+- blockHeight: cue=0 (own track, can overlap), other=4
+- Gap between same-track blocks = 1 row (chain or gap cell)
+- Cross-track alignment still handled by assignRows solver after buildCells
+
+## Next steps
+- Fix cue track: need gap padding on cue track too so cell rows match topRow
+- May need to rethink cue-move step — moving cues to highest triggered row scrambles cue order
+- Verify topRow math matches actual emitted cell rows end-to-end