Research: published P104 channel leaves every endpoint frame unrestored (no new kappa)

research/p104-channel-touch-restoration-audit/CERTIFICATE.json

@@ -0,0 +1,49 @@
+{
+  "schema_version": 1,
+  "claim_id": "p104-channel-touch-restoration-audit",
+  "pinned_main": "6f100d4bb5083f8c35d96fd95a539811c62f245c",
+  "flags": 208,
+  "unique_source_target_pairs": 208,
+  "program_events": 4058,
+  "joint_source_target_events": 0,
+  "matched_pairs_target_before_source": 208,
+  "matched_pairs_cooccurring": 0,
+  "matched_same_first_event": 0,
+  "latest_target_first_event": 410,
+  "latest_target_firsts": ["Y103"],
+  "earliest_source_event": 488,
+  "earliest_sources": ["X74"],
+  "population_gap": 78,
+  "minimum_matched_first_touch_gap": 84,
+  "minimum_matched_pair": {"source": 58, "target": 102, "target_first": 405, "source_first": 489},
+  "source_initial_ranks": {"1": 104},
+  "source_final_equals_coordinate_basis": 104,
+  "target_initial_empty": 104,
+  "target_final_ranks": {"8": 104},
+  "frame_restored_sources": 0,
+  "frame_restored_targets": 0,
+  "matched_pairs_with_either_endpoint_frame_restored": 0,
+  "sources_starting_at_basis_head": 208,
+  "targets_whose_final_contains_basis_head": 4,
+  "basis_head_in_target_final": [
+    {"flag_index": 0, "source": 0, "target": 34, "basis_head": 232},
+    {"flag_index": 1, "source": 0, "target": 7, "basis_head": 232},
+    {"flag_index": 9, "source": 4, "target": 23, "basis_head": 196},
+    {"flag_index": 43, "source": 21, "target": 59, "basis_head": 15}
+  ],
+  "flag0": {
+    "source": 0,
+    "target": 34,
+    "basis_head": 232,
+    "source_events": [591, 3954],
+    "target_events": [129, 3500],
+    "source_coefficients": ["-2", "2"],
+    "target_coefficients": ["-1/4", "1/4"],
+    "source_initial": [232],
+    "source_frame_restored": false,
+    "target_frame_restored": false,
+    "target_final_contains_basis_head": true
+  },
+  "author_3242_fields_match": true,
+  "no_new_kappa": true
+}

research/p104-channel-touch-restoration-audit/MASTER-HANDOFF.md

@@ -0,0 +1,29 @@
+# Handoff: published-word chronology and frame restoration
+
+## Result
+
+Independent replay of the pinned #2981 matching on the #1991 channel witness confirms the #3242 first-touch separation (targets by event 410, sources from event 488, no joint X/Y event) and shows that none of the 104 source frames or 104 target frames is restored.
+
+Pinned main: `6f100d4bb5083f8c35d96fd95a539811c62f245c`. Selected κ is unchanged at 0.001672660197076918.
+
+## Switch
+
+The requested #3259 audit was not repeated. `research/p104-pooled-cut-independent` (Nick Petalas, head `b572b0c25569898579041200e95731c5a43c9189`) already registers a passing Python 3.12 verifier for the same missing pair `C[71,102] = -1/2`, forward cone 1,725, and backward cone 37 under the immutable-X, unread-Y, independent-dirt, monotone equal-frame contract. That cut originates in dcharb78's #3182. The live gap after it, in `docs/research/p104-live-support-map.md`, is chronological realization and dirty restoration. This package takes those for the published word only.
+
+## What was checked
+
+- Two scanners, neither imported from #3242 or #3259, read the hash-pinned flag certificate, channel witness, and geometry naming lines.
+- They agree with each other and with `CERTIFICATE.json`.
+- The overlapping #3242 fields match hash-pinned `research/p104-flag-channel-schedule/RESULT.json`.
+- Frame restoration is set equality of initial and final frames. All 104 sources end at the coordinate basis; all 104 targets start empty and end at rank 8.
+- The basis head lies in the target final for flags 0, 1, 9, and 43 only.
+
+## What remains open
+
+Coefficient execution, logical restoration of arbitrary dirt, a chronological correction program, native lowering, precision, physical hosting, and a complete invoice. This is not a universal P104 barrier and not a supplier admission.
+
+## Credit
+
+Nick Petalas (#2981 flags, #3259 support certification), dcharb78 (#3182 cut), jwr404 (#3242 chronology receipt), and the reviewed #1991 channel. Their notices are untouched.
+
+Prepared by Wentao Li with substantial AI assistance.

research/p104-channel-touch-restoration-audit/NOTICE.md

@@ -0,0 +1,18 @@
+# Notice
+
+Copyright 2026 Wentao Li
+
+This directory is a scoped audit of archived research. It is licensed under
+the Apache License, Version 2.0, the same license as the surrounding project.
+
+The 208-flag matching is Nick Petalas's work from PR #2981, retained at
+`research/p104-joint-unit-flags`. The 4,058-event witness is the reviewed
+#1991 native channel. The chronology claim under audit is jwr404's PR #3242,
+retained at `research/p104-flag-channel-schedule`. The fixed-cover support
+cut that this package does not re-derive is Nick Petalas's PR #3259
+certification of dcharb78's PR #3182, retained at
+`research/p104-pooled-cut-independent`.
+
+Those notices and files are unchanged. This package adds an independent
+reading of the pinned bytes. Prepared by Wentao Li with substantial AI
+assistance. No independent human or formal-kernel review is claimed.

research/p104-channel-touch-restoration-audit/PROOF.md

@@ -0,0 +1,54 @@
+# Chronology and frame restoration of the published P104 word
+
+## Claim
+
+Under the pinned #2981 208-flag matching and the #1991 4,058-event channel witness, every target register is first touched at or before event 410, every source register is first touched at or after event 488, no event names both an X register and a Y register, and zero of the 104 source frames and zero of the 104 target frames return to their initial frames.
+
+No new κ is claimed. The selected conditional result remains κ = 0.001672660197076918.
+
+## Why this package exists
+
+Pinned canonical main is `6f100d4bb5083f8c35d96fd95a539811c62f245c`. On that tree, `research/p104-pooled-cut-independent` is Nick Petalas's PR #3259 at `b572b0c25569898579041200e95731c5a43c9189`. Its registered verifier already certifies the PR #3182 fixed-cover cut: exactly one missing required pair among 4,992, `C[71,102] = -1/2`, with a forward cone of 1,725 edges and a disjoint backward cone of 37 edges, under the immutable-X, unread-Y, independent-dirt, monotone equal-frame contract. That certification is not repeated here.
+
+The live gap named after that cut is chronological realization and dirty restoration of the published word. PR #3242 (`research/p104-flag-channel-schedule`, `f3c7e5229d1ce856fe3d6fa5467b10679c0fda14`, jwr404) states the first-touch separation and records that the worked pair X0/Y34 is not restored. Its scientific review was still open on the pinned map, and no second package replays the claim. This note is that replay, plus the restoration census for every endpoint.
+
+## Contract
+
+Sources and targets are the names fixed by the pinned geometry lines
+
+```
+sources = [paths[f'X{s}'] for s in range(104)]
+targets = [paths[f'Y{t}'] for t in range(104)]
+```
+
+A register is touched by an event when its name occurs as a top-level field of that event. Event indices are zero-based positions in `program`. The first touch is the least such index, and the last touch is the greatest. An event is joint when it names at least one X register and at least one Y register.
+
+The matching is the 208 helpers in Nick Petalas's #2981 certificate. Each helper names a source, a target, and a basis whose first vector is the basis head. There are 208 distinct source/target pairs. Each source index and each target index occurs twice.
+
+A register is frame-restored when its initial frame and final frame are equal as sets of integers. Frames in the witness have no repeated vectors, so sorted-tuple equality and set equality coincide. This is a statement about the recorded frames of the published word. It does not track logical coefficients, and it does not say that some other schedule is unable to restore them.
+
+## Chronology
+
+The witness has 4,058 events, and every `X0`–`X103` and `Y0`–`Y103` occurs at least once. The forward scan and the index-list scan agree on the following.
+
+The latest first touch of a target is event 410, attained only by `Y103`. The earliest first touch of a source is event 488, attained only by `X74`. The population gap is 78. Every target first touch is at most 410, and every source first touch is at least 488. Therefore every target begins before every source. In particular, all 208 matched targets are first touched before their matched sources, and no matched pair shares a first index.
+
+No event is joint. The two scanners also compare the full touch sets of each matched pair and find 0 meetings. The smallest matched gap is 84, on source 58 and target 102 (target first at 405, source first at 489). The population gap of 78 is the separation between `Y103` and `X74`, which are not a matched pair.
+
+Flag 0 is source 0, target 34, basis head 232. `X0` occupies events 591 through 3954. `Y34` occupies events 129 through 3500. Their add coefficients are `{-2, 2}` and `{-1/4, 1/4}`. These fields match the pinned #3242 receipt. The coefficient lists in this certificate are sorted lexicographically; #3242 stores the same sets in a different order.
+
+## Frame restoration
+
+Every source initial frame is a singleton, and every source final frame is the coordinate basis `(1, 2, 4, 8, 16, 32, 64, 128, 256)`. Every target initial frame is empty, and every target final frame has rank 8. None of the 104 sources and none of the 104 targets is frame-restored. All 208 matched pairs therefore have both endpoints unrestored.
+
+All 208 source initial frames are exactly the singleton of that flag's basis head. The target final contains that basis head for only four flags: `(0, Y34, 232)`, `(0, Y7, 232)`, `(4, Y23, 196)`, and `(21, Y59, 15)`. The worked #3242 pair is one of those four. It is not the shape of the other 204 flags.
+
+`X0` starts at `[232]` and does not finish there. `Y34` does not start empty-equal to its final frame, and 232 does lie in that final frame. That is the #3242 restored flag, read as frame inequality.
+
+## What this does not prove
+
+A first-touch order is not subspace support. The #3259 cut remains the support statement for `C[71,102] = -1/2` under its own contract; this package does not re-prove it. A changed frame is not a logical identity. Exact coefficient delivery, restoration of arbitrary dirty inputs, native lowering, precision, physical hosting, and a complete channel-plus-correction invoice stay open. Nothing here admits a supplier or changes κ.
+
+## Credit
+
+Nick Petalas supplied the 208-flag certificate in PR #2981 and the independent support-cut certification in PR #3259. dcharb78 supplied the cut itself in PR #3182. jwr404 supplied the chronology reading audited here in PR #3242. The channel witness is the reviewed #1991 word those packages pin. Their files and notices are not modified.

research/p104-channel-touch-restoration-audit/README.md

@@ -0,0 +1,45 @@
+# Channel first-touch order and endpoint frame restoration
+
+**Scoped audit. No new exponent.**
+
+The #3259 support cut is already certified on pinned main by
+`research/p104-pooled-cut-independent`. This package instead replays the
+published 208-flag matching against the 4,058-event channel witness.
+
+Under that matching and witness, every target is first touched at or before
+event 410, every source is first touched at or after event 488, no event
+names both an X register and a Y register, and zero of the 104 source frames
+and zero of the 104 target frames return to their initial frames.
+
+The population bound and the unrestored X0/Y34 example agree with jwr404's
+PR #3242 receipt. The new census is that the frame failure covers every
+endpoint, and that the basis head meets the target final on only 4 of 208
+flags. Details, the matched gap of 84, and the limits of the claim are in
+[PROOF.md](PROOF.md).
+
+## Reproduce
+
+From the repository root, CPython 3.12:
+
+```sh
+python3.12 -I -O -Werror -B research/p104-channel-touch-restoration-audit/verify.py
+python3.12 -I -O -Werror -B research/p104-channel-touch-restoration-audit/independent.py
+python3.12 -I -O -Werror -B research/p104-channel-touch-restoration-audit/test_package.py
+```
+
+`verify.py` is read-only. Admission goes through `need()` and still fires
+when Python is started with `-O`. The two scanners do not import each other.
+
+Pinned main: `6f100d4bb5083f8c35d96fd95a539811c62f245c`. Input hashes are in
+[SOURCES.json](SOURCES.json).
+
+## Credit
+
+Nick Petalas / #2981 for the flags, Nick Petalas / #3259 and dcharb78 / #3182
+for the support cut that this package leaves in place, and jwr404 / #3242 for
+the chronology receipt under audit. Prepared by Wentao Li with substantial
+AI assistance.
+
+## Open next questions
+
+Can a chronological correction deliver the published coefficients and restore arbitrary dirt without repaying the frame loss? This census shows every endpoint frame fails to return. It does not construct or price that repair.

research/p104-channel-touch-restoration-audit/SOURCES.json

@@ -0,0 +1,43 @@
+{
+  "schema_version": 1,
+  "base_commit": "6f100d4bb5083f8c35d96fd95a539811c62f245c",
+  "upstream": "CrocSwap/integer-mult-bounds",
+  "files": [
+    {
+      "path": "research/p104-joint-unit-flags/CERTIFICATE.json",
+      "usage": "input",
+      "git_blob": "0fa7d2d40566e92f3ea1539723abf438ba65adb8",
+      "sha256": "a98126428a8da6eae91c50afcc42e014a284c82b554ef77e5b3a34fc6d228128",
+      "credit": "Nick Petalas, PR #2981"
+    },
+    {
+      "path": "research/p104-native-channel/inputs/LOWRANK-WITNESS.json",
+      "usage": "input",
+      "git_blob": "86beaf424e9f0f800ae3a62797d94ec9386b7706",
+      "sha256": "0f541bef7f19fdeffd67157684e9f03d785f42c89973632f508fc1a0452b580e",
+      "credit": "Reviewed #1991 native channel witness"
+    },
+    {
+      "path": "research/p104-endpoint-geometric-budget/geometry.py",
+      "usage": "input",
+      "git_blob": "f49752bcfc321a6129186d4daa774b2c3fdb84f1",
+      "sha256": "1a2792189beaf15aa8180136ca007cd2301992a62e77ba72edea99b298e4148f",
+      "credit": "X/Y naming lines for the 104 sources and 104 targets"
+    },
+    {
+      "path": "research/p104-flag-channel-schedule/RESULT.json",
+      "usage": "input",
+      "git_blob": "c2d7443350e7a8fdb77910689df75516b789b55d",
+      "sha256": "4ba6fd7555d8907ce28d52d8de5087133e3649431f8ef0617a8961ebcbb73e16",
+      "credit": "jwr404, PR #3242 author receipt"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "67f085acd86acd7d4e03922d9cf4b26bca66fed5",
+      "sha256": "47b42433f5a85809a3aadd29c2c339da2c6ed728e4f7762df6b88bccb7874cf0",
+      "pin": "base_commit:docs/research/agent-task-catalog.json",
+      "credit": "Priority revision 6 task catalogue at the pinned base"
+    }
+  ]
+}