Record byte retention of the eight selected-frontier source pulls

research/frontier-source-retention/README.md

@@ -0,0 +1,132 @@
+# Retention of the eight selected-frontier source pulls
+
+**Overlap receipt. No new κ and no ledger edit.** The selected conditional
+exponent stays `605457647/200000000000`.
+
+## Result
+
+On main `32890bc5de034ab8fc93ac88ead7279a6a416d04`, the first eight entries of
+`certificates/selected-result.json` `source_prs` are 5413, 5414, 5402, 5407,
+5398, 5400, 4610 and 5241. Each recorded head is an ancestor of that main.
+Every `research/` path named in those ledger rows is present at the pull-request
+head and byte-identical there and on this main.
+
+The selected execution closure is a different object from that archive
+retention. `certificates/frontier5413-source.json` lists 1,245 paths. All
+1,245 hashes match the files on this main. Its three package roots are the
+incorporated pulls 5407, 5413 and 5414. The trees of 5398, 5400 and 5402 are
+not in that closure.
+
+| PR | Ledger disposition | Research tree vs its head | Selected-record role |
+|---|---|---|---|
+| 5413 | incorporated | identical | audited source head of the selected certificate |
+| 5414 | incorporated | identical | `audit_pr` / `audit_head` |
+| 5407 | incorporated | identical | one of the three closure roots |
+| 5402 | partially reviewed | identical | named source; no autonomous credit for the eight coupled outputs |
+| 5398 | partially reviewed | identical | named source; author evidence not replayed in the ledger row |
+| 5400 | partially reviewed | identical | named source; same unresolved archival sentence as 5398 |
+| 4610 | incorporated | identical | previous frontier; `selected-result.json` has since changed |
+| 5241 | incorporated | identical | previous admission; same certificate change |
+
+5398 and 5400 still say that archival publication does not admit a supplier,
+an outer theorem, or an improved κ. The promotion text for 5413 says the final
+`+5×10^-12` over #5398/#5400 is compiler/outer headroom of the selected
+exponent, not a new recursive-price saving. Both sentences are in the pinned
+files. This receipt does not choose between them and does not mark either row
+incorporated.
+
+4610 and 5241 also name `certificates/selected-result.json`, which differs from
+both heads, and `docs/research/pr-intake-evidence/pr4610/PROOF-REVIEW.md`, which
+is absent at both heads and present now. Those two paths are successor records.
+They are not edits of the retained research trees. Every one of the eight
+GitHub closures is `merged_for_archive_only`.
+
+## Scope
+
+Task card `overlap-receipt`, priority revision 7, base
+`32890bc5de034ab8fc93ac88ead7279a6a416d04`. Scope status is **refined**.
+
+The card asks what content and credit of one small named set has been retained.
+The set is the selected certificate's eight-pull prefix. Scope stays inside
+disposition, byte identity and the published credit sentences.
+
+Ordered eligible list for session r38, rebuilt on 2026-10-25 against that
+main. Open handoffs from the 19 October index come first. A closed smallest
+task is dropped; a distinct remainder stays. Priority-table rows whose stated
+next decision is still open come next. Catalogue cards that are not already
+those rows follow, skipping cards the 23 October rebuild had already dropped
+and rows closed on this main or covered by an open pull request.
+
+1. A caller that accepts `(w, q, v_N)` without restoring the original source
+   slots (comment 19030045; open question 1 of
+   `research/p104-role-exchange-caller-audit`).
+2. The cost of placing L561 on the four scratch windows before
+   `L546 -= L561` (comment 19030049; open question 1 of
+   `research/qq16-l546-head-move`). The head-move bill itself is on main.
+3. `BIT.md` weighted leaves and fast header routes at the new `c`
+   (`uniform-transfer-audit`). The `2^{-4p}` sentence, the row-field pad, the
+   uniform evaluator and the continuation-stack bound are already delivered.
+4. The fixed-tape visit count of one blank-return of the 9,475-row R437
+   parent (`r437-carrier-integration`; open question 1 of
+   `research/r437-parent-spectator-blank`).
+5. `coverage-law-change` on the current selected construction. The older
+   #1094 screens are not a constraint on this selection.
+6. This package.
+
+Dropped before the cut: the T385 wrapper, closed by
+`research/t385-r437-copy-audit`; the retiming handoff; the four-provider
+rebind, on main in `research/mixed-provider-admission`; the matched
+park/restore bill, covered by open #4167; the #4091 equivalence, on main in
+`research/fourth-entry-ceiling-transfer` and #5280; `p104-side-correction` as
+its own row, because item 1 is that consumer; and the catalogue cards dropped
+in `research/rank8-u2-entrance-obstruction` (`formal-interface-lemma`,
+`bit-helper-contract`, `bit-new-register-model`,
+`complex-rank-shape-decoder`, `centre-path-sharing`,
+`connector-alphabet-change`, and the stock, census, replay, packing and
+scalar-itinerary cards). Work already delivered by this account is excluded,
+including every pull through #5281 and the fork `cursor/*` branches named in
+the session note.
+
+## Reproduce
+
+Python 3.12, standard library only, from the repository root. The eight
+recorded heads must be present in the local Git object store.
+
+```sh
+python3 -B research/frontier-source-retention/verify.py
+python3 -B scripts/research_package.py --package research/frontier-source-retention
+```
+
+Expected line:
+
+```text
+PASS_FRONTIER_SOURCE_RETENTION prs=8 identical_paths=26 manifest_files=1245 partial_but_named=5398,5400,5402 certificate_moved_after=4610,5241 no_ledger_edit=true no_new_kappa=true
+```
+
+The checker hashes all 1,245 manifest paths, compares each integration path
+with `git diff` under `GIT_NO_REPLACE_OBJECTS=1`, and rejects a forged
+`incorporated` label on 5398 and a one-path manifest mismatch. Not run: the
+frontier replay, Lean, Docker, and any edit of `docs/research/pr-ledger.json`.
+
+## Open next questions
+
+1. Should a maintainer leave the 5398 and 5400 rows partially reviewed, given
+   that their unresolved sentence still denies an improved κ, while the
+   selected promotion uses them as the base of a `5×10^-12` headroom term?
+   The first step is to quote those two sentences together. This package
+   cannot edit the ledger.
+2. Which fields of `certificates/selected-result.json` changed after the 4610
+   head `20bb93cd990d2ff683e4d0fc56b87a42cd5d3296`? The receipt only shows that
+   the file differs. The first step is a key-wise diff. The blocker is that
+   this checker does not interpret the successor certificate.
+
+## Credit
+
+The ledger rows, the promotion note and the contributor map are the
+maintainer intake records for Chris R / Clear1nsight (#5398, #5400, #5407,
+#5413, #5414, #5241), Dillon Amburgey / dillona (#5402) and Cao Yifan /
+intqwq (#4610), with the lineages those records already name. This package
+does not replay their mathematics.
+
+Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0.
+No independent human review and no formal-kernel acceptance are claimed.

research/frontier-source-retention/SOURCES.json

@@ -0,0 +1,42 @@
+{
+  "schema_version": 1,
+  "base_commit": "32890bc5de034ab8fc93ac88ead7279a6a416d04",
+  "files": [
+    {
+      "path": "certificates/frontier5413-source.json",
+      "usage": "input",
+      "git_blob": "02297a45018410b9ae807a097a16c2efa2547867",
+      "sha256": "a24f0f9461c8625c016d4a43e381a527b820530fa886ce0231583713ada246dd"
+    },
+    {
+      "path": "certificates/selected-result.json",
+      "usage": "input",
+      "git_blob": "fc3d4ba105e1f097d0d7d051a308281e86ab1446",
+      "sha256": "66974efdced4083c00025f97d95d83909a226cdf801edb9301c195c6641f6903"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "f126dd2e2dad2b669e75d1d0d0bd05c600725602",
+      "sha256": "f7a48bfa2659b88378282a8eee7d9fe52cdf505594124c4d1686b58510fffd72"
+    },
+    {
+      "path": "docs/research/community-neutral5413-contributors.md",
+      "usage": "input",
+      "git_blob": "7102f05762378c53d17e7067496c107430ea3ba1",
+      "sha256": "f3476ddf94ce6821c195fff9f07b71617e39f0320ce1d3bb2be7e8d5d373b678"
+    },
+    {
+      "path": "docs/research/pr-intake-evidence/pr5413-promotion/PROMOTION.md",
+      "usage": "input",
+      "git_blob": "93bcfcbf83460d7a6c069c7c238929e4cc3c3bb0",
+      "sha256": "9419931de95e317f5e08ffa98c86e069836b53f6438c2f413b0722377819aa44"
+    },
+    {
+      "path": "docs/research/pr-ledger.json",
+      "usage": "input",
+      "git_blob": "0ef27eee22300a0a512af129a0d8778ae4af46ab",
+      "sha256": "30af48f6f9bb64c698feeeafcf5c68e5ff76f6175f8757ee6ff45566152d1529"
+    }
+  ]
+}

research/frontier-source-retention/TASK.json

@@ -0,0 +1,66 @@
+{
+  "schema_version": 1,
+  "task_id": "overlap-receipt",
+  "priority_revision": 7,
+  "base_commit": "32890bc5de034ab8fc93ac88ead7279a6a416d04",
+  "original_card": {
+    "id": "overlap-receipt",
+    "question": "What exact content and credit of a small named PR set has already been retained?",
+    "starting_points": [
+      "docs/research/pr-ledger.md",
+      "CONTRIBUTORS.md"
+    ],
+    "methods": [
+      "Compare exact heads and retained artifacts to write a source-bound disposition receipt.",
+      "Audit an existing receipt for missing content or predecessor attribution."
+    ],
+    "acceptance_gate": "Separate complete integration, partial retention and changed heads; make no autonomous closures.",
+    "first_step": {
+      "command": [
+        "python3",
+        "-B",
+        "scripts/research_baseline.py"
+      ],
+      "expected": "PASS_RESEARCH_BASELINE_BINDINGS_ONLY",
+      "inspect": "Read one small PR set in the ledger and identify exact reviewed heads and unresolved content.",
+      "scope": "Positive control / source-bound orientation only; this is not a new construction or full cold replay."
+    },
+    "first_experiment": "Compare that set to retained artifacts; produce a complete or explicitly partial disposition receipt.",
+    "bottleneck_link": "Reliable overlap evidence preserves credit and prevents repeated review of already-retained material.",
+    "allowed_variations": [
+      "One small related PR family or one existing integration receipt."
+    ],
+    "scope_change_triggers": [
+      "Treating matching titles or one shared blob as proof of full incorporation."
+    ],
+    "overlap_terms": [
+      "overlap receipt",
+      "partial integration",
+      "attribution"
+    ]
+  },
+  "scope_status": "refined",
+  "actual_scope": "For the eight pull requests named first in certificates/selected-result.json source_prs, separate byte retention of each ledger integration path from the ledger disposition and from the selected-record role. Do not change any disposition.",
+  "scope_change_reason": "The overlap-receipt card asks for one small named PR set. This set is the selected frontier's own source_prs prefix at base 32890bc5de034ab8fc93ac88ead7279a6a416d04: 5413, 5414, 5402, 5407, 5398, 5400, 4610 and 5241. The question is retention versus disposition, not a new supplier.",
+  "bottleneck_link": "The selected conditional kappa cites these eight pulls. A later reader who trusts only the partially_reviewed rows would drop predecessors the selected record still names, and a reader who treats source_prs as a replay would over-claim 5398 and 5400.",
+  "budget_minutes": 60,
+  "stop_rule": "Stop after the eight-row retention table. Do not edit the ledger, do not replay the frontier, and do not promote a partially_reviewed row.",
+  "outcome": "partial",
+  "evidence": "verify.py checks ancestry, byte-identical research trees, the 1245-file frontier manifest, ledger dispositions, and the promotion and contributor sentences. It rejects a forged incorporated label on 5398 and a forged manifest mismatch.",
+  "author": "unpaidmoon",
+  "ai_assistance": "Prepared by unpaidmoon with substantial Cursor Grok assistance.",
+  "open_questions": [
+    {
+      "question": "Should a maintainer leave 5398 and 5400 partially_reviewed after the 5413 promotion, or record a narrower successor disposition that still does not claim their author evidence was replayed?",
+      "why_it_matters": "Their ledger rows still say archival publication does not admit an improved kappa, while the selected record lists them as sources of kappa 0.003027288235.",
+      "first_step": "Quote the 5398 unresolved_question beside the promotion sentence that the final 5e-12 over 5398/5400 is compiler/outer headroom, and decide which sentence remains the disposition.",
+      "blocked_by": "This package is forbidden from editing the ledger or closing a review."
+    },
+    {
+      "question": "Which bytes of the previous 4610 certificate does the current selected-result.json replace, beyond the kappa fraction?",
+      "why_it_matters": "The 4610 and 5241 research trees match their heads, but certificates/selected-result.json does not. The difference is a successor record, not a silent edit of those packages.",
+      "first_step": "Diff selected-result.json at 20bb93cd990d2ff683e4d0fc56b87a42cd5d3296 against the pinned current file and list changed keys.",
+      "blocked_by": "The receipt only proves the files differ. It does not interpret every changed field."
+    }
+  ]
+}

research/frontier-source-retention/VERIFY.json

@@ -0,0 +1,24 @@
+{
+  "schema_version": 1,
+  "status": "ready",
+  "python": "3.12",
+  "requirements": [],
+  "inputs": [
+    "certificates/frontier5413-source.json",
+    "certificates/selected-result.json",
+    "docs/research/community-neutral5413-contributors.md",
+    "docs/research/pr-intake-evidence/pr5413-promotion/PROMOTION.md",
+    "docs/research/pr-ledger.json"
+  ],
+  "commands": [
+    {
+      "argv": [
+        "{python}",
+        "-B",
+        "{package}/verify.py"
+      ],
+      "timeout_seconds": 120,
+      "checks": "Eight source PRs: merged ancestry, research-tree identity with each head, ledger disposition, 1245 manifest hashes, promotion and contributor sentences, and two rejection controls."
+    }
+  ]
+}