fix(#221): restore #210's boundary-B mechanism, silently reverted by #213's rebase

[thorwhalen]

What this restores

PR #213's rebase onto main silently reverted PR #210's entire "boundary B" mechanism (real-time output nodes go silent rather than pitch-shift when Clock.timeScale ≠ 1). Full diagnosis in #221.

Verified before restoring: git diff 074b88d^ HEAD -- <file> was empty for all six source files #210 touched — i.e. HEAD's state for each was byte-identical to before #210 ever landed. This PR restores exactly those six files to their post-#210 state (git checkout 074b88d -- <file>) and re-creates the two files that were deleted outright (src/dag/timescale.ts, test/timescale_boundary.test.ts). No other change — this is a mechanical restoration of previously-reviewed, previously-merged code, not a reimplementation.

docs/design/stream-applier.md needed no change: it was the one file the rebase left untouched, so it already (correctly) describes the mechanism as shipped.

Gate

• npm run typecheck — clean
• npm test — 139/139 files, 1696/1696 tests green, including the restored test/timescale_boundary.test.ts (the structural test that fails if a synth-role node stops consulting realtimeOutputAllowed)
• npm run build — clean
• npm run catalog — no diff (no node/param surface changed)

Closes #221

🤖 Generated with Claude Code

https://claude.ai/code/session_011HSBVhDjRU4apSLcRkavv9

---

Self-reviewed only — no independent sub-agent spawn tool was available in this worker's session; see the review note comment on this PR for the self-adversarial checklist performed instead. Flagging for a post-merge review to be queued.

[thorwhalen]

Review note: self-reviewed only

No independent sub-agent spawn tool was available in this worker's toolset (no Agent/Task
launcher — only SendMessage to already-running peers), so per the crowsnest REVIEW RULE's
explicit fallback, I merged after a rigorous self-adversarial pass rather than an independent
review. Flagging for a post-merge review to be queued.

What I checked, playing skeptic against my own diff:

1. Is "byte-identical to pre-#101 M-G: boundary B becomes a mechanism — accelerated playback mutes, never pitch-shifts #210" actually true, not assumed? git diff 074b88d^ HEAD -- <file>
   is empty for all six source files this PR restores — a verifiable fact, not an inference from
   the PR titles/messages.
2. Did anything else touch these files between #101 M-G: boundary B becomes a mechanism — accelerated playback mutes, never pitch-shifts #210 and now, such that restoring the pre-revert
   blob could silently clobber unrelated work? git log --oneline 074b88d..HEAD -- <file> for each
   of the six shows exactly one commit: 9150a02 (the revert itself). Nothing else to lose.
3. Does a new synth-role node exist that #101 M-G: boundary B becomes a mechanism — accelerated playback mutes, never pitch-shifts #210 never saw, which the restored structural test
   (test/timescale_boundary.test.ts) might have needed to know about but doesn't? The test derives
   its node list from the registry at run time, not a hand-written list — so a synth-role node
   added after #101 M-G: boundary B becomes a mechanism — accelerated playback mutes, never pitch-shifts #210 (e.g. anything from Epic: conductor mode wired — conducting-gesture understanding (ictus TS core), a score pipeline, and a virtual orchestra #187/Design: generative AI inside instruments, and lazy loading as a catalogued pattern #188) is automatically in scope. The full suite is green
   with it restored, so nothing new is failing the guard.
4. Does the restored code integrate cleanly with the current file bodies, or did the old blob
   paste in stale context? Diffed 73d8b20..HEAD for the three node files — each change is a small,
   additive, single-purpose insertion (an import + one guard clause / condition), no leftover or
   duplicated logic, consistent with each file's current surrounding code.
5. Full gate: npm run typecheck clean, npm test 139/139 files · 1696/1696 tests, npm run build
   clean, npm run catalog no diff.

Nothing above surfaced a defect. Proceeding to the hosted CI gate and squash-merge.

[thorwhalen]

Note found post-merge: this duplicated a parallel session's fix. Issue #221 (opened by this session) was closed as a duplicate of #219 (filed ~3 min earlier by a different session investigating the same regression independently) before this PR was even opened; #219's fix landed as #220 (02660c8), merged at 16:35:16Z — before this PR (#222) was created at 16:36:46Z.

Verified harmless: git diff 02660c8d 0362ba7 (the commit before this PR's squash-merge vs. after) is empty — this PR's net effect on main was a no-op, since #220 had already applied byte-identical content (both restored the same 074b88d blobs for the same six files). No functional change, no conflict, no lost work — just a redundant CI run, merge commit and deploy trigger. Leaving this on the record rather than deleting the trail.