Skip to content

Consolidation: overlap receipt for the open selected-replay tooling PRs; #1142 fails on main and collides with #1108 (no new kappa) #1654

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Mister-Meeseeks merged 1 commit into from
Oct 14, 2026
Merged

Consolidation: overlap receipt for the open selected-replay tooling PRs; #1142 fails on main and collides with #1108 (no new kappa) #1654

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
137 changes: 137 additions & 0 deletions research/selected-replay-tooling-receipt/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,137 @@
# Overlap receipt: open selected-replay tooling PRs

**Consolidation receipt. No new κ, no disposition, nothing closed.** The selected
frontier and every pinned file are unchanged.

This receipt covers eight open PRs at exact heads, against main `1f7a9349`. Seven
have the selected #1094 cold replay path as their premise; that path runs
`verify_frontier1094.py` → `1042` → `799` → `741`. The eighth, #1108, is included
because #1142 re-adds its package. Four of them (#1101, #1108, #1142, #1151) are
in the maintainer queue's next batch
(`research/community-efficient200-audit/QUEUE-CURSOR.json`).

| PR | Head | Author | Subject | Applies to main | Own checks at trial merge |
|---|---|---|---|---|---|
| #1101 | `4d45bef9` | rcfc-jreed | concurrent bit589/complex741 driver for the #741 ancestry | clean | PASS (8 tests) |
| #1108 | `b76bb285` | martinxxx9 | compact-boundary cancellation obstruction | clean | PASS |
| #1142 | `0c2873e6` | martinxxx9 | #1042 replay-capsule closure audit | clean | **FAIL** (see below) |
| #1151 | `99d4f42f` | thakrarsagar | parallel layer-end controls, frontier1042 chain | clean | PASS (`verify.py`, 155 s) |
| #1277 | `02d27951` | thakrarsagar | #1094 reuse-mode ancestry binding audit | clean | PASS |
| #1361 | `40965b22` | thakrarsagar | CI drift/trigger guard for the 35 hash-asserted files | clean | PASS (35 pins, 0 drift, 21 tests) |
| #1509 | `362d2f91` | thakrarsagar | checked `cpp_int` stand-in for bit589 step 4 | clean | PASS |
| #1646 | `3f21415b` | JosieElliston | sparse phase recipes, frontier1042 `phases` steps | clean (head on `858771bc`) | not rerun |

**Retention.** None of the eight heads has content already on main: no changed
path has an identical blob at the base. The only path that exists on main is
`.github/workflows/frontier1094.yml`, which #1361 edits. So there is nothing to
receipt as "already integrated". Each PR needs its own review.

**#1142 does not pass on current main.**

1. `research/frontier1042-capsule-closure-seed75/verify.py` prints `FAIL: repository pin
differs: certificates/selected-result.json`. `PINS.json` pins the #1042 record
(sha256 `207d1f49…`) at the path that now holds #1094's record.
2. That record is retained byte for byte at `certificates/checkpoints/frontier1042.json`.
Changing only `selected_record.path` to that file gives
`PASS_FRONTIER1042_CAPSULE_CLOSURE`. This was tested in scratch and is not part of this
package. The contributor or reviewer should decide whether to repin.
3. The body's reproduce command, `make frontier1042-capsule-closure-verify`, exists
neither in the PR nor on main.
4. #1142 also adds all seven files of #1108's `research/compact-boundary-cancellation-seed51/`.
Five of them still differ after CRLF→LF normalization. #1142's copy is the
earlier one: it pins `certificates/selected-result.json` to the #942 blob `3d7529d7`
and fails with `Source drift`. #1108's copy pins `certificates/checkpoints/frontier942.json`
and passes. Merging either PR and then the other gives an add/add conflict.

**Premises that still hold at the base.**
- #1277: `verify_frontier1094.py` still has exactly one bare `check_selected(...)`
statement, so the `--baseline1042` reuse path discards its result.
- #1361: the guard still finds 35 live pins.
- #1101: `verify_frontier741.py` is unchanged since #1101's base.
- #1151, #1509: their declared inputs are unchanged.

**Pairwise overlap.** Apart from the #1108/#1142 collision, no two PRs in the family change the same file. The pairs that
look similar address different steps or questions:
- #1151 and #1646 cover the lec and phases steps of the frontier1042 chain.
- #1101 and #1509 both concern bit589 in the #741 ancestry. #1101 runs it concurrently
with complex741; #1509 replaces Boost for its step 4. #1101's driver does not scrub
`BOOST_INCLUDE`, but I did not test the two together.
- #1277, #1361 and #1142 cover three different parts of the selected evidence:
#1277 checks the reuse-mode binding, #1361 checks CI drift and triggers, and #1142
checks the archive bytes.

All the speed-up PRs (#1101, #1151, #1646) are opt-in and unwired. The pinned drivers
are unchanged, so none of them shortens `make frontier1094-verify` by itself. Their
equality claims differ, as their own PR bodies state:
- #1151: its receipts are byte-identical to the archived ones.
- #1646: it reproduces the pinned recipe digests.
- #1101: its output trees match the sequential driver's only after timing fields and
timing-dependent digests are masked.

## Reproduce

```sh
python3 -B research/selected-replay-tooling-receipt/verify.py # base-side facts + 5 controls
for n in 1101 1108 1142 1151 1277 1361 1509 1646; do
git fetch origin pull/$n/head:refs/receipt/pr$n; done # exact heads, if still reachable
python3 -B research/selected-replay-tooling-receipt/verify.py --heads # recompute per-PR facts
```

The default run, which is the registered one, needs Git history containing the base.
It checks the following at `1f7a9349`:
- the selected and #1042-checkpoint record hashes;
- the cold-chain calls;
- #1277's premise;
- the absent make target.

It also checks that five mutated receipts are rejected.

`--heads` recomputes, for every recorded head:
- the changed paths;
- the identical-on-main paths;
- clean application to the base;
- the #1108/#1142 byte differences, pin and add/add conflict;
- #1142's capsule pin.

It writes one temporary merge commit object, which Git garbage-collects as unreachable.
If a contributor force-pushes, `--heads` fails rather than silently re-reading.

The trial-merge check results in `RECEIPT.json` (`trial_merge_checks_recorded`) are
recorded evidence from this session, not recomputed by `verify.py`. They come from a
shared macOS arm64 host with Python 3.14, and Python 3.12 for the registered packages.

**Not checked:** correctness of any PR's mathematics or speed claims beyond its own
verifier, the cold `make frontier1094-verify`, Linux, and the PRs outside this family.

## Open next questions

1. **Should #1142 be repinned or narrowed before review?** Its audit holds once it points
at the retained #1042 checkpoint. The bundled seed51 copy duplicates #1108. First step:
the contributor drops the seed51 files and repins one path. Blocked on: the contributor
or a maintainer decision.
2. **Do other next-batch PRs collide with each other?** #1142 re-adding #1108's directory
was found only because both were in this family.

I ran an unregistered path-collision scan of the 100 `next_batch` heads (recorded only
here, not checked by `verify.py`). Each pair was merged in order with true merge parents.
- #1129→#1157 and #1148/#1178/#1195 merge cleanly in any order tried.
- Besides #1108/#1142, these pairs conflict textually:
- #1110/#1190 (`docs/research/pr-ledger.md`);
- #1112/#1137 (`Makefile`);
- `.gitattributes` in the gauge44-ends7 family. Merging any of #1104, #1107, #1117 or
#1121 and then #1124 or #1130 conflicts, even though those heads are ancestors of
#1124 and #1130; the reason was not traced. Merging any of the seven and then #1193
also conflicts. #1104→#1107→#1117→#1121 merge cleanly.
- #1172 conflicts with each of #1107, #1117, #1121, #1124 and #1130 inside
`research/gauge44-ends7-t16/` (`MANIFEST.json`, `README.md`, `RESULT.json`,
`data/`…). These look like competing versions of one package, which needs a reviewer's
choice rather than a mechanical merge. I did not compare their contents.
3. **Which opt-in speedups should be wired into the pinned cold chain?** #1101, #1151 and
#1646 each state their own equality evidence (see above). Wiring them needs pinned-driver
changes that only a maintainer can approve.

## Credit

Covered PRs and their authors are listed above. The maintainer queue cursor and
efficient-review audit are by Douglas Colkitt. Prepared by Josie Elliston's donated
session with Claude Code (Claude Opus); see [TASK.json](TASK.json). Apache-2.0.
Loading
Loading