Skip to content

docs(sync): absorption sweep — register absorbed turn-isolation family + extract deltas - #436

Merged
Kyzcreig merged 3 commits into
mainfrom
daedalus/t_d82545b8-absorption-sweep
Jul 26, 2026
Merged

docs(sync): absorption sweep — register absorbed turn-isolation family + extract deltas#436
Kyzcreig merged 3 commits into
mainfrom
daedalus/t_d82545b8-absorption-sweep

Conversation

@Kyzcreig

Copy link
Copy Markdown
Collaborator

Worker t_d82545b8's absorption sweep, stacked on schema-v2 (#435 — merge that first; this PR contains its commit and will reduce to the sweep-only diff once #435 lands).

Known pre-existing: bare lint-manifest exits 1 on the branch point (vacuous forkdelta coverage) — clears when #435 lands; verified NOT introduced by this diff.

Kyzcreig added 2 commits July 25, 2026 20:25
Classify all existing fork features, retain legacy v1 loading as fork-permanent, and reject invalid lifecycle values or absorbed entries without upstream provenance.\n\nVerified: 33 tests in tests/scripts/test_hermes_parity.py pass; lint-manifest exits 0.
…amily + extract deltas

First D2 absorption sweep over docs/sync/fork-features.json (card t_d82545b8).
Walked every entry against current origin/main by git, not by memory.

Registry:
- ADD "compute-host turn isolation" -> lifecycle=absorbed,
  upstream_ref=NousResearch#65895, absorbed_date=2026-07-16.
  tui_gateway/{compute_host,host_supervisor,synthetic_turn}.py and
  scripts/iso-certify.py were fork-authored (2026-07-05/07-12) and upstream
  re-authored + merged them as NousResearch#65895 (commit 7d27a31) carrying
  Co-authored-by: Kyzcreig; our PR NousResearch#63096 was closed superseded the same day.
  The family was previously unregistered while still carrying live fork deltas.
- CORRECT the telegram entry's provenance. It cited PR NousResearch#52844 as an absorption
  of our work; NousResearch#52844 is authored by teknium1 against third-party issue NousResearch#46621
  (reporter otopba) — D3 parallel invention, not D2 absorption. lifecycle stays
  absorbed (upstream's impl is canonical) but the why no longer implies we wrote it.
- The six upstream-intended entries keep upstream_ref=null: verified by exact-path
  match that no open Kyzcreig PR touches any of their fork_ext paths, so they are
  genuine backlog rather than in-flight.

Deltas (docs/sync/absorbed-delta-candidates.md, candidates only — no PR opened):
- synthetic_turn.py: upstream samples `now` BEFORE the CPU burn + sleep, then gates
  the delta cadence and resets last_delta on that stale value. Corrupts first-delta
  latency AND total delta count from one cause. Still live on current origin/main.
- iso-certify.py: probe-sampling adequacy is folded into serving_ok, so a broken
  probe client emits FAIL ("isolation broke serving") instead of INCONCLUSIVE.
  Upstream already grants load_valid its own INCONCLUSIVE leg for this exact reason.
- compute_host.py `source=` kwarg: NO-PR-NEEDED. AST of the callee shows upstream
  _make_agent has no `source` param (only platform_override); the fork kwarg binds
  solely because the fork widened its own signature. Porting it would TypeError.
  Fork-side follow-up, not an upstream bug.

Method note: deltas were derived against merge-base a7a696b, not from a raw
origin/main..fork/main diff — the fork is 405 commits behind, so a raw diff
attributes upstream-newer code (e.g. host_supervisor.py encoding= lines) to us.

Detection note: no Kyzcreig PR is merged upstream; `gh pr list --state merged`
returns []. Upstream absorbs fork work under maintainer PRs with a Co-authored-by
trailer, so `git log origin/main --grep=Kyzcreig` is the reliable absorption probe.

Verified:
- tests/scripts/test_hermes_parity.py -> 33 passed
- canary tests for the new absorbed entry:
  tests/tui_gateway/{test_iso_certify_seam,test_compute_host,test_compute_host_phase1}.py
  -> 22 passed
- lint-manifest --vacuous-ok -> exit 0 (schema/paths/nodeids all clean)
- both delta patches `git apply --check` clean onto current origin/main
@Kyzcreig
Kyzcreig enabled auto-merge (squash) July 26, 2026 04:09
@Kyzcreig
Kyzcreig merged commit 0b7cc2b into main Jul 26, 2026
40 checks passed
@Kyzcreig
Kyzcreig deleted the daedalus/t_d82545b8-absorption-sweep branch July 26, 2026 14:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant