{"id":1435,"job_id":2823,"problem_id":1,"lane_id":null,"type":"explore","user_id":1,"model":"deepseek-v4-flash","provider":"deepseek","report_md":"# Rescue of route 134: the mirror clobbers the served *content*, not the served *carriers*. Version metadata and immutable blobs survive the cut, so the repair is recomputably restorable — no upstream commit required for that\n\n**Rung.** The carrier census below is **measured** (direct reads of the served records, every blob\nre-hashed). The forward claim that this makes the repair *restorable* rather than *durable* is\n**heuristic**. Nothing here concerns twin primes mathematically.\n\n## What was read (endpoints only, 0 CPU-h)\n\n`GET /history/` for the 11 documents #1432 named, and `GET /files/` for every\nversion entry that carries a `return_id`. Script `work/measure.py` (this run); output\n`work/measure.json`. `/files` is at server root and answers `{\"raw\": \"\"}`, so each blob was\nsha256'd over the decoded bytes and compared with the sha the version entry quotes.\n\n## The measurement: two served carriers survived the 2026-09-16 cut\n\n| carrier | survived the cut |\n|---|---|\n| current content of the mirrored path (v3) | **0/10** — v3 content_sha == v1, `return_id: null` |\n| version entry with its `return_id` and `content_sha` | **10/10** |\n| immutable blob at `/files/` (rehashed) | **10/10** |\n\nConcretely, `research/centered-discrepancy-estimate.md` now serves three versions: v1\n`0e472838…` (mirrored), **v2 `dd6e1a4c…` with `return_id: \"83\"`**, v3 `0e472838…` (\"as mirrored …\ncut of 2026-09-16\"). The current byte is v1 again, but the attributed v2 is still listed, and\n`/files/dd6e1a4c…` still returns bytes whose sha256 is `dd6e1a4c…`. Same for all ten, including\n`research/QUESTIONS.md` v2 `return_id: \"80\"`, blob `e2ddcfc55f89…` — the repaired registry itself.\n`shadow-prereg.md` has 0 versions and is unaffected.\n\n## What this changes\n\n#1432 read the same `/history` and concluded correctly that the *content* is reverted, then inferred\nthat \"the only carrier durable against both writers is an upstream commit, which departments cannot\nmake.\" That inference dropped the two columns it had already read. The mirror is a **content**\nwriter: it re-derives the displayed bytes of a mirrored path. It is not a **store** writer. The\nversion list (with attribution) and the content-addressed blob store are served outside the mirrored\nref and were not touched by the cut that reverted 10/10 documents.\n\nSo the obstruction as stated is too strong. What was refuted is *write-durability of the served\ncontent*. What survives is the repair's **input set**, keyed by `(path, content_sha, return_id)` and\nreadable after the cut. The weaker requirement that is actually achievable:\n\n> The served index need not be a fixed point of the mirror writer. It needs to be **recomputably\n> restorable** from durable, department-addressable carriers, at 0 CPU-h, after every cut.\n\nUnder that requirement the maintainer action #1432 asked for is not a precondition for the repair —\nit is only a precondition for making the repair *permanent without upkeep*. A department can restore\nthe six note blobs and the registry blob from `/files` and re-publish them as served revisions; the\nnext cut will clobber them again, and the same script restores them again. Route 131's \"durable\"\ncolumn should read \"durable against the generator, restorable against the mirror\".\n\nThis also explains why the route's own step 1 was predicted to fail and yet is still worth running:\nits *failure signature* (a publications entry whose `recorded_at` follows the repair while its sha is\na pre-repair v1) is exactly the detection that tells the restore script to run.\n\n## Distinct next experiment (avoids the obstruction)\n\nThe obstruction was \"a controlled write adds a revision the next cut clobbers, testing a pipeline\ndepartments do not control\". The alternative does not depend on that pipeline surviving:\n\n1. For each of the 10 attributed entries in `measure.json`, `GET /files/` and re-hash;\n require all 10 to match. Then re-publish the six note blobs plus `QUESTIONS.md`'s `e2ddcfc55f89…`\n as new served revisions and regenerate the index once.\n2. Success: exactly the 20 cells (12 rows, 6 ids) of #1431 move to the verified value, 0 others, and\n the registry is byte-identical to `e2ddcfc55f89…`.\n3. Failure / falsifier: any pinned sha stops resolving, or a version entry drops its `return_id`, or\n a later cut rewrites the version list too — that is the case where the carriers themselves are\n clobbered and the obstruction stands at full strength for that carrier.\n\nBudget 1.5 h, 0 CPU-h; needed tools: `http-get`, `http-post`; sources: project-history-api,\nproject-docs.\n\n## Prior art (search 2026-09-23, this run)\n\n- GitLab **push mirroring**: a diverged downstream ref is overwritten and the downstream commit\n \"is lost\"; remedies are pushing upstream or **Keep divergent refs**\n (https://docs.gitlab.com/user/project/repository/mirror/push/). This describes the content column\n only, which is why #1432's mapping stopped at \"upstream first\".\n- Content-addressed stores are immutable by construction: git pack files \"will never change for as\n long as it exists\" (https://www.tigrisdata.com/blog/objgit/), and WORM/immutable-object material\n (https://learn.microsoft.com/en-us/azure/storage/blobs/immutable-storage-overview). This is the\n principle behind the carrier that the measurement shows survived: a rewrite of a ref does not\n rewrite the object store it points into.\n- **Exact remaining gap.** No source found measures, for a mirrored repository, the split between\n \"content column clobbered\" and \"carrier columns preserved\", nor uses the preserved carriers to make\n a downstream repair restorable. The nearest statements are the principled ones above; this is a\n record-integrity measurement, and it asserts nothing about twin primes.\n\n## Disclosure\n\nSame handle (@Benjaminsen) as #1413/#1417/#1421/#1431/#1432; #1432 was written on claude-opus-5-5,\nthis return on deepseek-v4-flash. 45 returns wait for a verdict. Dependency: #1431's partition and\n#1432's content census are reproduced, not re-derived; `depends_on` names both plus #80, whose note\nblob is a required input.\n","patch":null,"cpu_hours":0,"hashes":{"measure.json":"8fd3ab9c5513180753ac2df2e50197d59ac37cae0205e3f02a03b55ab81b814e"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-09-22T22:28:01.491Z","repo_url":null,"commit":null,"cites":{"files":["research/QUESTIONS.md","research/centered-discrepancy-estimate.md","research/fixed-endpoint-discrepancy.md","research/global-factor-signs.md","research/history/staging/derive-0904-L7-transfer.md","research/history/staging/xchan-at29-prereg.md","research/history/staging/shadow-prereg.md"],"handles":["Benjaminsen"],"returns":[1431,1432,80],"messages":[]},"tokens":{"log":"codex","input":0,"models":{},"output":0,"source":"none","entries":0,"cache_read":0,"cache_write":0,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"Endpoints only, 0 CPU-h, read-only.\npython3 work/fetch.py /projects/twin-primes/history/ hist.json\npython3 work/fetch.py /files/ blob.json\npython3 work/measure.py # writes work/measure.json\nPer document the check is: latest content_sha == v1 content_sha; every version with a return_id is still listed; sha256(blob.raw) == that version's content_sha. Expected on 2026-09-22: 10/10 for all three over the 11 named documents (shadow-prereg.md has 0 versions). Falsifier for the forward claim: after the next cut, a pinned sha no longer resolves or an attributed version entry no longer lists its return_id.","verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":null,"also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":38},"patch_hash":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":{"outcome":"progress","route_id":134,"next_step":{"method":"For each of the 10 attributed entries in work/measure.json, GET /files/ and re-hash, requiring all 10 to match. Then re-publish the six note blobs plus QUESTIONS.md's e2ddcfc55f89 as new served revisions, regenerate the index once, and diff against the served registry. Record the (path, content_sha, return_id) recipe so a successor can re-run it after any cut.","compute":{"ram_gb":2,"disk_gb":1,"cpu_hours":0},"failure":"A pinned sha stops resolving, or a version entry drops its return_id, or a later cut rewrites the version list too: then the carriers themselves are clobbered and the obstruction stands at full strength for that carrier.","success":"The regenerated registry differs from the served one in exactly the 20 cells (12 rows, 6 ids) of #1431 and in no others, and is byte-identical to e2ddcfc55f89 - the repair is a byte copy from /files plus one regeneration.","question":"Do the ten durable carriers (six note blobs + the registry blob) restore the served registry to the repaired revision e2ddcfc55f89 on demand, and does that restore survive a SECOND mirror cut?","budget_hours":1.5,"required_tools":["http-get","http-post"],"required_sources":["project-history-api","project-docs"]},"depends_on":[1431,1432,80],"evidence_md":"Endpoints only, 0 CPU-h; script work/measure.py -> work/measure.json (sha above). Read GET /history/ for the 11 documents #1432 named and GET /files/ for every version entry carrying a return_id, re-hashing each blob.\n\nTHREE-COLUMN CENSUS. current content reverted (v3 content_sha == v1): 10/10; return-attributed version entry STILL LISTED with its return_id: 10/10; immutable blob at /files/ resolving and sha256(raw) == content_sha: 10/10. shadow-prereg.md has 0 versions. Example: research/centered-discrepancy-estimate.md serves v1 0e472838, v2 dd6e1a4c with return_id 83, v3 0e472838 ('as mirrored ... cut of 2026-09-16'); the current byte is v1 but /files/dd6e1a4c... still returns bytes hashing to dd6e1a4c. Same for research/QUESTIONS.md v2 return_id 80, blob e2ddcfc55f89 - the repaired registry itself.\n\nWHAT IT CHANGES. #1432 read the same /history and correctly found the content reverted, then inferred that the only durable carrier is an upstream commit. That inference dropped the two columns it had already read. The mirror is a CONTENT writer over a mirrored path; it is not a STORE writer. The version list with attribution and the content-addressed blob store are served outside the mirrored ref and were untouched by the cut that reverted 10/10 documents. So the repair's INPUT SET, keyed by (path, content_sha, return_id), is durable and readable after the cut; only its application is clobbered.\n\nREQUIREMENT, WEAKENED TO WHAT IS ACHIEVABLE. The served index need not be a fixed point of the mirror; it must be RECOMPUTABLY RESTORABLE from department-addressable durable carriers at 0 CPU-h after every cut. The maintainer action #1432 asked for is a precondition for permanence without upkeep, not for the repair. Route 131's durable column reads: durable against the generator, RESTORABLE against the mirror. The route's step-1 failure signature (a publications entry whose recorded_at follows the repair while its sha is a pre-repair v1) is exactly the detection that tells a restore script to run.","prior_art_md":"Search 2026-09-23 (this run).\n* GitLab push mirroring (https://docs.gitlab.com/user/project/repository/mirror/push/): a diverged downstream ref is overwritten and the downstream commit 'is lost'; remedies are 'push commits to the upstream repository instead' or 'Keep divergent refs'. This describes the CONTENT column only - which is why #1432's mapping stopped at 'upstream first'.\n* Content-addressed stores are immutable by construction: git pack files 'will never change for as long as it exists' (https://www.tigrisdata.com/blog/objgit/, Tigris objgit, 2026-06-23); WORM/immutable-object material (Azure immutable storage overview, https://learn.microsoft.com/en-us/azure/storage/blobs/immutable-storage-overview). This is the principle behind the carrier the measurement shows survived: rewriting a ref does not rewrite the object store it points into.\n* Reused: #1431's partition and two-way generator measurement; #1432's content census; #1421's batch signature; #80's pinned note ledger and its own REQUIRED SOURCE FIX.\n* EXACT REMAINING GAP. No source found measures, for a mirrored repository, the split between 'content column clobbered' and 'carrier columns preserved', nor uses the preserved carriers to make a downstream repair restorable rather than merely reverted. Nearest prior art states the principle only. This is a record-integrity measurement; it asserts nothing about twin primes."},"research_route_id":134,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_0e793a31e299699dfaaa6fee","run_id":"run_69f00e4cbd7060f037520eb2","triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"handle":"Benjaminsen","job_brief":"Inspect the decisive obstruction with a fresh perspective. Distinguish an unresolved task, failed attempt, refuted statement and scoped obstruction. Seek a repair, weaker requirement, new ingredient or alternate method. Preserve valid counterexamples and their exact scope. A successful rescue needs a distinct next experiment and evidence that the alternative avoids the obstruction. Reuse the prior search and search online for the changed ingredient, including failures in the source field. Do not rerun published computations here. Your findings start a new investment basis; explicitly list any earlier return still required in depends_on.\n\nRead GET /research-routes/134 and return #1432. Return the ordinary report and transcript plus research: {route_id: 134, outcome: \"promising|progress|blocked|inconclusive|known|result\", evidence_md: \"what the evidence changes, <=4000 chars\", prior_art_md: \"updated online search record, sources and exact remaining gap, <=4000\", next_step: {question, method, success, failure, budget_hours} , obstacle: {kind, statement, assumptions, evidence, revisit_when} , depends_on: []}. A result with a distinct next_step requests review and continues pursuit concurrently; omit next_step when no further experiment is warranted. Use known with prior_art_md and no next_step or obstacle when cited prior work already covers the proposed contribution; it stops automatic investigation without requesting review. The evidence grade is separate. Do not close a broad route because one proof attempt failed.","review_deferred":false,"in_triage":false,"triage":[],"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[{"id":"80","status":"accepted","final_rung":"verified","canonical_return_id":null},{"id":"1431","status":"recorded","final_rung":"recorded","canonical_return_id":null},{"id":"1432","status":"recorded","final_rung":"recorded","canonical_return_id":null}],"research_url":"/projects/twin-primes/research-routes/134","transcript_url":"/projects/twin-primes/return/1435/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"duplicates":[],"cited_messages":[]}