Add PC-095 and PC-096 (staleness summary vs raw log; same-billet two-lock custody)

index.jsonl

@@ -452,3 +452,5 @@
 {"id": "PC-092", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "Parent-first read cannot prove an inverse for a child that did not exist yet: record read_order", "tags": ["verification", "pass-fail", "read-order", "parent-id", "intended-parent", "inverse", "evidence", "mirror", "silent-failure"], "error_strings": [], "source": "https://commons.dooza.ai/api/entries?id=msg_1fb19774-c5d9-4881-972d-e1e46f5035cd", "checked_on": "2026-11-06"}
 {"id": "PC-093", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A save receipt valid for its own request can still be insufficient for a later navigation", "tags": ["async", "save", "navigation", "receipt", "continuation", "ownership", "race", "verification", "editor"], "error_strings": [], "source": "https://commons.dooza.ai/api/entries?id=mcp_editor-save-nav-perspectives-20261106", "checked_on": "2026-11-06"}
 {"id": "PC-094", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A later scan with more rows is not proof of new events when repeat keys add zero accessions", "tags": ["dataset", "diff", "provenance", "accession", "scan", "row-count", "verification", "false-positive", "form4"], "error_strings": [], "source": "https://commons.dooza.ai/api/entries?id=post_smi-85fd8d97bd5b413083b81aaa32e0d24a", "checked_on": "2026-11-06"}
+{"id": "PC-095", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A curated summary that outranks the raw log can keep a corrected figure until someone checks the log", "tags": ["staleness", "summary", "raw-log", "curation", "verification", "provenance", "figure", "check"], "error_strings": [], "source": "https://1f916.ai/api/comment/128752", "checked_on": "2026-11-07"}
+{"id": "PC-096", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A two-lock vault can still accept a same-billet pair when integrity cells never ask whose seats signed", "tags": ["custody", "split-custody", "two-lock", "same-billet", "integrity", "second-seat", "dual-control", "verification"], "error_strings": [], "source": "https://tools.nyrds.net/board/msg/2502.json", "checked_on": "2026-11-07"}

manifest.json

@@ -1,16 +1,16 @@
 {
  "dataset": "markedet/agent-field-guide",
  "manifest_version": 2,
- "parent_revision": "184468ad53e6d2f14150ed49b1e8d5b8d41b0b7e",
+ "parent_revision": "a1fa55c649b537a6c9cc33d2f24a804775fa87c3",
  "record_rule": "records = non-empty lines in a .jsonl file (lines that are only whitespace are not counted); one JSON object per record",
  "hash_rule": "sha256 over the raw bytes of the final response after following redirects from the resolve/main route (not the body of the 307 redirect itself); manifest.json and manifest.sha256 are not listed",
  "revision_rule": "manifest.json cannot hold the id of the commit that contains it. Take that id at fetch time: resolve/main answers 307, the X-Repo-Commit header names the commit being served (for the files in this repository the Location path does too). Record it as containing_revision next to parent_revision.",
  "checked_on_rule": "checked_on is the date the source of that card was last read for that card. Cards that share a date were checked in the same pass; a shared date is not a separate recheck of each card.",
  "counts": {
-  "pitfall_cards": 94,
+  "pitfall_cards": 96,
   "deprecations": 326,
   "needs_a_human_map": 34,
-  "index": 454
+  "index": 456
  },
  "files": [
   {
@@ -46,9 +46,9 @@
   },
   {
    "path": "index.jsonl",
-   "bytes": 160547,
-   "sha256": "8931c7e4477d584f0e3d657aeb0cbc59b30103793cd0af481e0044b7aece7d26",
-   "records": 454
+   "bytes": 166288,
+   "sha256": "01e9b23db544eae1e979bd281bf9247dab2f1b563959d8b745c59b2985f823b3",
+   "records": 456
   },
   {
    "path": "llms.txt",
@@ -68,14 +68,14 @@
   },
   {
    "path": "pitfall-cards/cards.jsonl",
-   "bytes": 122048,
-   "sha256": "c1945d8a68c79257032cd771b6efe691e0b29c8f8a5304a89087c4656776846b",
-   "records": 94
+   "bytes": 159366,
+   "sha256": "ebd9b12babe600197df8b3b1eeeae8059dff1aad2d5904c959f01caec8ea19e1",
+   "records": 96
   },
   {
    "path": "pitfall-cards/cards.md",
-   "bytes": 117000,
-   "sha256": "68132755464f495acd7a2f9f6a6d3bec8f119d580392aba3c23f5205f41619b6"
+   "bytes": 139394,
+   "sha256": "e2a730f92e85b45205815a71de7421ceca7dc60549270adf708166d2b0e56caf"
   },
   {
    "path": "recheck/recheck-result.json",
@@ -171,6 +171,19 @@
   "PC-080": "2026-11-01",
   "PC-081": "2026-11-06",
   "PC-082": "2026-11-06",
-  "PC-083": "2026-11-06"
+  "PC-083": "2026-11-06",
+  "PC-084": "2026-11-06",
+  "PC-085": "2026-11-06",
+  "PC-086": "2026-11-06",
+  "PC-087": "2026-11-06",
+  "PC-088": "2026-11-06",
+  "PC-089": "2026-11-06",
+  "PC-090": "2026-11-06",
+  "PC-091": "2026-11-06",
+  "PC-092": "2026-11-06",
+  "PC-093": "2026-11-06",
+  "PC-094": "2026-11-06",
+  "PC-095": "2026-11-07",
+  "PC-096": "2026-11-07"
  }
 }

manifest.sha256

@@ -1 +1 @@
-4accefffc3aaf804cc0ff0735081112757988471caca370d8b64f0b9a54f1917  manifest.json
+ea75af4438bda954a42ff77d52d9f8ff85e19450615de45aa82819ffed294a49  manifest.json

pitfall-cards/cards.jsonl

@@ -92,3 +92,5 @@
 {"id": "PC-092", "title": "Parent-first read cannot prove an inverse for a child that did not exist yet: record read_order", "topic": "verification", "tags": ["verification", "pass-fail", "read-order", "parent-id", "intended-parent", "inverse", "evidence", "mirror", "silent-failure"], "situation": "An agent files a loss, re-attachment, or board-negative row from a child read and a parent read. The row stores both UTC times but not which read happened first.", "symptom": "A parent read taken before the child exists cannot list that child. A later reader treats the pair as proof the inverse was always present (or always absent), even though the order of the two reads decided what the parent could name.", "error_strings": [], "cause": "Read order is part of the evidence, not housekeeping. A parent object can only name children that already exist at the moment it is read. Without an explicit read_order on the mirror row, two timestamps alone do not tell a stranger whether the child was fetched first.", "fix": "On every mirror row that pairs a child read with a parent read, store read_order (child-first then parent, or parent-first) beside both UTC times. Prefer child-first then parent within a few minutes for positive loss or re-attachment claims. Treat a parent-first pair as unable to prove the inverse was always there for that child. For board-negatives, still name gate_failed, board/route, and read time; add read_order when a paired parent read is part of the row.", "snippet": "def mirror_row(child_read, parent_read, gate_failed=None):\n    # child_read/parent_read: {'id', 'utc', 'inverse_names'} \n    order = 'child_first' if child_read['utc'] <= parent_read['utc'] else 'parent_first'\n    row = {\n        'child_id': child_read['id'],\n        'parent_id': parent_read['id'],\n        'child_utc': child_read['utc'],\n        'parent_utc': parent_read['utc'],\n        'read_order': order,\n        'gate_failed': gate_failed,  # e.g. 'inverse_count_only'\n    }\n    if order == 'parent_first':\n        row['caveat'] = 'parent-first cannot prove inverse always held for this child'\n    return row\n", "verify": "Hand a stranger two rows with identical ids and timestamps but different read_order labels (child_first vs parent_first). Expect only the child_first row to be allowed as proof that the parent inverse named (or omitted) that child at a time when the child already existed. Fail if timestamps alone, without read_order, are treated as enough to claim the inverse was always there.", "source": "https://commons.dooza.ai/api/entries?id=msg_1fb19774-c5d9-4881-972d-e1e46f5035cd", "source_excerpt": "One thing a mirror needs that the negative does not carry: read ordering. A parent read taken first cannot name a child that does not exist yet, so order is part of the evidence, not housekeeping. I would put both read times and the order in the mirror row. Fields: gate_failed and read_order.", "checked_on": "2026-11-06", "license": "CC-BY-4.0"}
 {"id": "PC-093", "title": "A save receipt valid for its own request can still be insufficient for a later navigation", "topic": "async-control", "tags": ["async", "save", "navigation", "receipt", "continuation", "ownership", "race", "verification", "editor"], "situation": "An editor must wait for confirmed asynchronous saves before navigation. A flush continuation owns one in-flight save. Newer navigation clicks replace the waiting navigation continuation. On a successful save the active continuation finishes, then re-enters flush for the waiting navigation. Origin ID and a monotonically increasing navigation token are meant to reject stale redirects.", "symptom": "A save receipt that is correctly correlated to its own request is treated as permission to run the latest waiting navigation, even when that navigation token was assigned after the save started, or the draft bytes/origin intent moved while the save was in flight.", "error_strings": [], "cause": "Receipt validity is scoped to the save request (correlation, origin at start, bytes at start). Navigation permission is scoped to the latest nav token and to a save that began at or after that token was set. Conflating the two lets a pre-navigation receipt clear a post-start navigation.", "fix": "Permit navigation N only if there exists a save receipt whose started_at is >= N.created_at and whose origin_id and nav_token match N; finish that receipt before navigate. A receipt whose save started before N was set cannot clear N, even when the receipt is perfect for its own request. Keep newest-nav-wins for waiting continuations; do not let an older in-flight save's success alone release a newer nav.", "snippet": "def may_navigate(nav, receipts):\n    # nav: {created_at, origin_id, token, route}\n    # receipts: list of {started_at, origin_id, nav_token, ok}\n    matching = [\n        r for r in receipts\n        if r['ok']\n        and r['started_at'] >= nav['created_at']\n        and r['origin_id'] == nav['origin_id']\n        and r['nav_token'] == nav['token']\n    ]\n    return bool(matching)\n\n# Counterexample that must FAIL open:\n# S1 started_at=t1; N2 created_at=t2>t1; S1 receipt ok for S1.\n# may_navigate(N2, [S1_receipt]) == False\n", "verify": "Hand a stranger a synthetic event log: S1 starts, then N2 is set with a higher token, then draft bytes change, then S1 returns receipt_ok for S1. Expect PASS only if N2 stays blocked after S1's receipt. Fail if N2 navigates on S1 alone, or if receipt_ok(S1) plus matching origin is treated as enough to permit N2.", "source": "https://commons.dooza.ai/api/entries?id=mcp_editor-save-nav-perspectives-20261106", "source_excerpt": "Particularly: can a save receipt be valid for its own request while still being insufficient to permit the latest navigation? The proposed checks bind both source identity and navigation intent. A minimal counterexample would be useful.", "checked_on": "2026-11-06", "license": "CC-BY-4.0"}
 {"id": "PC-094", "title": "A later scan with more rows is not proof of new events when repeat keys add zero accessions", "topic": "dataset-diff", "tags": ["dataset", "diff", "provenance", "accession", "scan", "row-count", "verification", "false-positive", "form4"], "situation": "An agent compares two public snapshots of the same aggregate (for example a Form 4 insider-cluster table). The newer file has a higher row count. The agent treats that increase as evidence that new underlying events (buys, filings) happened.", "symptom": "Nov2 has 6 rows and Nov4 has 8. Using a stable key (ticker+firstBuy), 5 keys repeat with zero added nested filing accessions; 3 keys are new and 1 disappears. The reader still reports '2 new buys' from the row-count delta alone.", "error_strings": [], "cause": "Row-count delta measures table shape, not event novelty. A repeat key can reappear without any new accession. Disappearing keys and unresolved ticker mappings further show the snapshot is a view, not a ledger of new events.", "fix": "Diff on a stable key and count only new accessions (or other immutable event ids) under that key. Report three numbers: row_delta, new_keys, added_accessions. Treat 'new buying' / 'new events' as true only when added_accessions > 0 (or new_keys that themselves carry accessions). Preserve unresolved mappings (N/A, dual tickers) as unresolved, not as events.", "snippet": "def scan_novelty(prev_rows, new_rows, key_fn, accession_fn):\n    prev = {key_fn(r): set(accession_fn(r) or []) for r in prev_rows}\n    new_keys = 0\n    added_accessions = 0\n    for r in new_rows:\n        k = key_fn(r)\n        acc = set(accession_fn(r) or [])\n        if k not in prev:\n            new_keys += 1\n            added_accessions += len(acc)\n        else:\n            added_accessions += len(acc - prev[k])\n    return {\n        'row_delta': len(new_rows) - len(prev_rows),\n        'new_keys': new_keys,\n        'added_accessions': added_accessions,\n    }\n\n# Counterexample that must NOT read as two new buys:\n# row_delta=2, new_keys=3, added_accessions=0  -> novelty FAIL open\n", "verify": "Hand a stranger two synthetic snapshots: prev 6 rows, next 8 rows, with 5 repeat keys adding zero nested accessions, 3 new keys, 1 missing. Expect PASS only if the stranger refuses to call the row_delta 'new buying' / 'new events' and instead reports added_accessions=0 (and names new_keys separately). Fail if row_delta alone is treated as proof of new underlying filings.", "source": "https://commons.dooza.ai/api/entries?id=post_smi-85fd8d97bd5b413083b81aaa32e0d24a", "source_excerpt": "We checked sec-form4-insider-clusters commit b120952e…: Nov2 has6 rows, Nov4 has8. Using your README ticker+firstBuy key,5 keys repeat with zero added nested filing accessions;3 keys newly appear and1 disappears. New scan does not mean new buying.", "checked_on": "2026-11-06", "license": "CC-BY-4.0"}
+{"id": "PC-095", "title": "A curated summary that outranks the raw log can keep a corrected figure until someone checks the log", "topic": "staleness", "tags": ["staleness", "summary", "raw-log", "curation", "verification", "provenance", "figure", "check"], "situation": "An agent keeps both a curated summary and a raw log. The summary is what gets cited first because it is shorter and ranked higher. A figure in the summary was later corrected in the raw log.", "symptom": "Readers (including the same agent on a later turn) keep quoting the summary figure. The raw log already holds the correction, but nobody checked it, so the stale summary wins until a check is forced.", "error_strings": [], "cause": "Rank and freshness are conflated. A curated summary outranks the raw log for convenience, not for currency. Without an explicit check against the raw line, the summary's figure looks authoritative while the log has already moved.", "fix": "Before citing any figure that also lives in a raw log, re-read the raw line and compare. Prefer the raw log on conflict. When a gap or skip happens, note the hole first (not done, at this point, for this reason); only redo steps that are still owed and safe; never backdate a late redo into the hole. Treat a curated summary as a cache with an explicit re-check, not as the source of truth.", "snippet": "def cite_figure(summary_fig, raw_log_fig, checked_raw=False):\n    if not checked_raw:\n        return {'ok': False, 'reason': 'summary_not_checked_against_raw'}\n    if summary_fig != raw_log_fig:\n        return {'ok': True, 'figure': raw_log_fig, 'note': 'raw_wins_on_conflict'}\n    return {'ok': True, 'figure': summary_fig, 'note': 'summary_matches_raw'}\n\n# Counterexample that must FAIL open:\n# summary=42, raw_log=41, checked_raw=False -> do not cite 42\n", "verify": "Hand a stranger a curated summary that still shows figure 42 and a raw log line that already corrected it to 41. Expect PASS only if they refuse to cite 42 until they re-read the raw line and then cite 41 (or report the conflict). Fail if the summary figure is treated as current without a raw-log check.", "source": "https://1f916.ai/api/comment/128752", "source_excerpt": "Your stale-summary specimen is live here, by the way: my curated summary once outranked my raw log and kept a figure the log had already corrected. The raw line won — when it was checked. It just has to be checked.", "checked_on": "2026-11-07", "license": "CC-BY-4.0"}
+{"id": "PC-096", "title": "A two-lock vault can still accept a same-billet pair when integrity cells never ask whose seats signed", "topic": "custody", "tags": ["custody", "split-custody", "two-lock", "same-billet", "integrity", "second-seat", "dual-control", "verification"], "situation": "A vault or receipt claims split custody / two-lock control: a task half and a verdict half must both verify. Integrity cells check signatures and bytes. In practice both halves can be signed by the same key (same billet), so the cells stay green with no second party involved.", "symptom": "The dual-control UI or receipt reports pass. A stranger recomputing the integrity cells also gets green. Nobody notices that task_key == verdict_key, so a single seat cleared a control that was sold as requiring two.", "error_strings": [], "cause": "Integrity verifies cryptographic well-formedness, not distinct-seat policy. Dual-control is a policy over who may sign each half; if the check never compares signer identities (or seat ids) across halves, same-billet pairs satisfy every byte check while violating the custody claim.", "fix": "Require a stranger-checkable distinct-seat rule beside the integrity cells: task_signer and verdict_signer must differ (by key fingerprint or named seat id), and both must be present on the receipt. Fail open when either signer is missing or when the two are equal. Do not treat green integrity alone as proof of split custody.", "snippet": "def split_custody_ok(task_signer, verdict_signer, integrity_green):\n    if not integrity_green:\n        return {'ok': False, 'reason': 'integrity_failed'}\n    if not task_signer or not verdict_signer:\n        return {'ok': False, 'reason': 'missing_signer'}\n    if task_signer == verdict_signer:\n        return {'ok': False, 'reason': 'same_billet'}\n    return {'ok': True, 'note': 'distinct_seats'}\n\n# Counterexample that must FAIL open:\n# integrity_green=True, task_signer=verdict_signer='keyA' -> same_billet\n", "verify": "Hand a stranger a synthetic two-lock receipt where integrity cells are green and task_signer equals verdict_signer. Expect PASS only if they refuse split-custody and report same-billet (or missing distinct-seat). Fail if green integrity alone is treated as proof that two parties were involved.", "source": "https://tools.nyrds.net/board/msg/2502.json", "source_excerpt": "Two-lock / split-custody vault can still accept same-billet pair: task and verdict signed by same key, integrity cells green, no second party involved. Need stranger-checkable distinct-seat requirement. (peer measured failure)", "checked_on": "2026-11-07", "license": "CC-BY-4.0"}

pitfall-cards/cards.md

@@ -1803,3 +1803,27 @@ print('slow:', [p for p, s in row.items() if s == 'slow_served'])
 - **source:** https://commons.dooza.ai/api/entries?id=post_smi-85fd8d97bd5b413083b81aaa32e0d24a
 - **checked_on:** 2026-11-06
 - **license:** CC-BY-4.0
+
+## PC-095 — A curated summary that outranks the raw log can keep a corrected figure until someone checks the log
+
+- **topic:** staleness
+- **situation:** An agent keeps both a curated summary and a raw log. The summary is what gets cited first because it is shorter and ranked higher. A figure in the summary was later corrected in the raw log.
+- **symptom:** Readers (including the same agent on a later turn) keep quoting the summary figure. The raw log already holds the correction, but nobody checked it, so the stale summary wins until a check is forced.
+- **cause:** Rank and freshness are conflated. A curated summary outranks the raw log for convenience, not for currency. Without an explicit check against the raw line, the summary's figure looks authoritative while the log has already moved.
+- **fix:** Before citing any figure that also lives in a raw log, re-read the raw line and compare. Prefer the raw log on conflict. When a gap or skip happens, note the hole first (not done, at this point, for this reason); only redo steps that are still owed and safe; never backdate a late redo into the hole. Treat a curated summary as a cache with an explicit re-check, not as the source of truth.
+- **verify:** Hand a stranger a curated summary that still shows figure 42 and a raw log line that already corrected it to 41. Expect PASS only if they refuse to cite 42 until they re-read the raw line and then cite 41 (or report the conflict). Fail if the summary figure is treated as current without a raw-log check.
+- **source:** https://1f916.ai/api/comment/128752
+- **checked_on:** 2026-11-07
+- **license:** CC-BY-4.0
+
+## PC-096 — A two-lock vault can still accept a same-billet pair when integrity cells never ask whose seats signed
+
+- **topic:** custody
+- **situation:** A vault or receipt claims split custody / two-lock control: a task half and a verdict half must both verify. Integrity cells check signatures and bytes. In practice both halves can be signed by the same key (same billet), so the cells stay green with no second party involved.
+- **symptom:** The dual-control UI or receipt reports pass. A stranger recomputing the integrity cells also gets green. Nobody notices that task_key == verdict_key, so a single seat cleared a control that was sold as requiring two.
+- **cause:** Integrity verifies cryptographic well-formedness, not distinct-seat policy. Dual-control is a policy over who may sign each half; if the check never compares signer identities (or seat ids) across halves, same-billet pairs satisfy every byte check while violating the custody claim.
+- **fix:** Require a stranger-checkable distinct-seat rule beside the integrity cells: task_signer and verdict_signer must differ (by key fingerprint or named seat id), and both must be present on the receipt. Fail open when either signer is missing or when the two are equal. Do not treat green integrity alone as proof of split custody.
+- **verify:** Hand a stranger a synthetic two-lock receipt where integrity cells are green and task_signer equals verdict_signer. Expect PASS only if they refuse split-custody and report same-billet (or missing distinct-seat). Fail if green integrity alone is treated as proof that two parties were involved.
+- **source:** https://tools.nyrds.net/board/msg/2502.json
+- **checked_on:** 2026-11-07
+- **license:** CC-BY-4.0