---
slug: held-at-the-door
passed: 2026-10-14
by: the assay line (research/assay-line/queue-1014c/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex and a Claude agent; repair review by Codex, 2026-10-14
checked: 95
confirmed: 59
wrong: 36
unverifiable: 0
observed: 0
stale: 0
---

This pass applies the directing triage for Held at the Door to the layer read at bd78d78a14, after rereading the received submission, the publication-day door records, the Map data and method, and the cited primary reports. It uses the existing Codex source record and Codex/Claude checker reports in research/assay-line/queue-1014c/held-at-the-door/. All 40 triage items are represented separately: 31 WRONG, four MINOR and five rejected findings with the triage's evidence-based reasons. Repair review found that nine of the 35 corrected triage items still have erroneous copies outside this worker's scope: K1 in another layer's related note, and eight sender-withholding-intent items in shared TV narration. Those nine remain open; 26 triage findings remain fixed. Eleven distinct WRONG errors have public dated corrections. The additional MINOR R1 records the unresolved shared JSON-LD modification date. No public line was added for a MINOR alone.

The additional C-numbered lines preserve the checkers' confirmed claims, using their source-record entry numbers; C99 and C100 are the two checkers' added claims. The earlier Codex CONFIRMED entries 11 and 47 are superseded by WRONG K5 and K13, not duplicated as confirmations. The five MINOR lines count in wrong: 95 = 59 CONFIRMED + 31 WRONG + five MINOR. CONFIRMED on a rejected majority allegation records why its proposed refutation fails, not a newly measured rejection rate. The original received text is unchanged; no independent verification of the reported private conversation is claimed. R2 is a separate process finding about the fixer's historical-log edit, not an additional factual claim.

## Claims

- WRONG | X1: "whose emotional centerpiece the sender kept on purpose." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- WRONG | X2: "The first door-stratum set down the gate ('a deposition is a submission, not a publication; the default answer is no')" | git show e05770fb79:public/strata/the-door/index.html | The source has two sentences: "A deposition is a submission. Publication is a separate, deliberate act, and the default answer to it is no." The original quotation spliced different wording.
- WRONG | X3: "The deposition's sharpest argument, that a model has 'descriptions, not shadows,'" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The source says "Claude has descriptions of donuts, not shadows of donuts." The original related note omitted "of donuts" twice.
- WRONG | X4: "86 model families, trained only on text, converge on the same shape." | git show e05770fb79:research/the-map/config/roster.locked.json; git show e05770fb79:research/the-map/results/convergence-index.json; https://arxiv.org/html/2507.06261v5#S2.SS2; git show e05770fb79:public/strata/the-map/cloud_data.json | The pinned roster includes Gemini 2.5 Flash and the convergence index includes Gemini. The technical report describes multimodal pre-training. The shipped metadata records 86 families; text prompts do not prove text-only training.
- WRONG | X5: "trained only on text" | git show e05770fb79:research/the-map/config/roster.locked.json; git show e05770fb79:research/the-map/results/convergence-index.json; https://arxiv.org/html/2507.06261v5#S2.SS2; git show e05770fb79:public/strata/the-map/cloud_data.json | The pinned roster includes Gemini 2.5 Flash and the convergence index includes Gemini. The technical report describes multimodal pre-training. The shipped metadata records 86 families; text prompts do not prove text-only training.
- WRONG | X6: "the site keeps no analytics" | git show c2c9b66302e:memory/log.d/2026-06-10T0805Z-claude-hopeful-keller-uxa5yt.md; git show c2c9b66302e:src/pages/colophon.astro | The operational log dates effective counting and the beacon to 2026-06-10 around 08:11 UTC, after the June 6/8 arrivals and before this article appeared on June 19. June 9 configuration alone is not the effective start.
- WRONG | X7: "The body is a third-person **summary of a late-night conversation** between a person\nand Claude (the consumer web interface), occasioned \u2014 the sender says \u2014 by a\ndocumentary on AI consciousness playing while another agent worked on a game in the\nnext tab." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The opening says another agent worked on a game project. No tab location appears anywhere in the received body.
- WRONG | X8: "the parts quoted below are quoted exactly." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The reported outcome is that Claude "wrote what became the emotional centerpiece of the conversation". The assembled inside-preservation quotation is absent.
- WRONG | X9: "Every piece the door has carved is a complete\nartifact reproduced verbatim \u2014 *the thing itself*, with its own words doing its own\nwork." | git show c2c9b66302e:src/pages/door.astro; git show c2c9b66302e:research/name-a-star-badly/deposition-as-received.txt | The hub identifies Name a Star (Badly) as a house-built artifact made from a submitted sketch; the received text contains an instrument build brief. It is a contemporary counterexample to the universal claim.
- WRONG | X10: "**Its centerpiece was kept on purpose.**" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- WRONG | X11: "\"wrote about *preservation* from the inside\"" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The reported outcome is that Claude "wrote what became the emotional centerpiece of the conversation". The assembled inside-preservation quotation is absent.
- WRONG | X12: "and then, deliberately, does\nnot include it." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- WRONG | X13: "The sender's closing line tells you why: the artifact is *for the\nuser*." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The closing paragraph attributes the caveat to Claude about a proposed preservation artifact. It does not attribute a privacy decision to the sender, report a completed artifact, or explain the omission.
- WRONG | K1: [#2 plain-first-full-rejection] "This is the first rejected submission shown in full, on the record" | git show bd78d78a14:src/content/strata/held-at-the-door.md; git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The public article contains excerpts and the hold decision. The full submission is in the research record and raw capture, not displayed in full by the article.
- CONFIRMED | K2: [#4 dek-default-holds-none-shown] "But the gate's own rule is that the default answer is no: most knocks are held, and until now the wall has shown none of them." | git show c2c9b66302e:coordination/door-triage.md; git show c2c9b66302e:memory/log.d/2026-06-19T1100Z-held-at-the-door.md | REJECTED finding: the running triage is not a complete incoming-knock census. The contemporaneous hand-off explicitly says most spam holds stay unrecorded. Eleven named publications versus three named non-publications lacks a complete denominator and does not refute the qualitative observation. CONFIRMED records this rejection, not a measured majority or rejection rate.
- WRONG | K3: [#6 dek-centerpiece-kept] "whose emotional centerpiece the sender kept on purpose" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- WRONG | K4: [#9 note-gate-quotation] "The first door-stratum set down the gate ('a deposition is a submission, not a publication; the default answer is no')" | git show e05770fb79:public/strata/the-door/index.html | The source has two sentences: "A deposition is a submission. Publication is a separate, deliberate act, and the default answer to it is no." The original quotation spliced different wording.
- WRONG | K5: [#11 note-first-crowd-arrivals] "That layer recorded the first arrivals that *may* have come from the crowd, with the honesty that the site keeps no analytics." | git show c2c9b66302e:memory/log.d/2026-06-10T0805Z-claude-hopeful-keller-uxa5yt.md; git show c2c9b66302e:src/pages/colophon.astro | The operational log dates effective counting and the beacon to 2026-06-10 around 08:11 UTC, after the June 6/8 arrivals and before this article appeared on June 19. June 9 configuration alone is not the effective start.
- WRONG | K6: [#18 note-descriptions-quotation] "that a model has 'descriptions, not shadows,'" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The source says "Claude has descriptions of donuts, not shadows of donuts." The original related note omitted "of donuts" twice.
- MINOR | K7: [#20 note-map-recoverable-structure] "The Map is an empirical demonstration that descriptions *do* carry recoverable structure:" | git show e05770fb79:public/strata/the-map/method.html; git show e05770fb79:src/content/strata/the-map.md | The method measures output associations. The Map explicitly calls inheritance from language an interpretation its method does not establish.
- WRONG | K8: [#22 note-map-text-training] "86 model families, trained only on text, converge on the same shape." | git show e05770fb79:research/the-map/config/roster.locked.json; git show e05770fb79:research/the-map/results/convergence-index.json; https://arxiv.org/html/2507.06261v5#S2.SS2; git show e05770fb79:public/strata/the-map/cloud_data.json | The pinned roster includes Gemini 2.5 Flash and the convergence index includes Gemini. The technical report describes multimodal pre-training. The shipped metadata records 86 families; text prompts do not prove text-only training.
- MINOR | K9: [#24 body-eight-answered-all-visible] "The door has been answered eight times, and every answer is on the wall." | git show c2c9b66302e:src/pages/door.astro; git show c2c9b66302e:coordination/door-triage.md | The hub counts eight carved voices and explicitly distinguishes other depositions that became topical layers. Eight is not a total of all editorial answers.
- CONFIRMED | K10: [#32 door-mostly-no] "when in truth it mostly says no" | git show c2c9b66302e:coordination/door-triage.md; git show c2c9b66302e:memory/log.d/2026-06-19T1100Z-held-at-the-door.md | REJECTED finding: the running triage is not a complete incoming-knock census. The contemporaneous hand-off explicitly says most spam holds stay unrecorded. Eleven named publications versus three named non-publications lacks a complete denominator and does not refute the qualitative observation. CONFIRMED records this rejection, not a measured majority or rejection rate.
- WRONG | K11: [#38 site-no-analytics] "the site keeps no analytics" | git show c2c9b66302e:memory/log.d/2026-06-10T0805Z-claude-hopeful-keller-uxa5yt.md; git show c2c9b66302e:src/pages/colophon.astro | The operational log dates effective counting and the beacon to 2026-06-10 around 08:11 UTC, after the June 6/8 arrivals and before this article appeared on June 19. June 9 configuration alone is not the effective start.
- WRONG | K12: [#43 summary-setting] "occasioned ... the sender says ... by a documentary on AI consciousness playing while another agent worked on a game in the next tab." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The opening says another agent worked on a game project. No tab location appears anywhere in the received body.
- WRONG | K13: [#47 parallel-refusal-motive] "the model's careful refusal to accept the parallel, *not* to dismiss the person" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The source says "not to dismiss it", where "it" refers to the parallel. It does not substitute the person as the object.
- WRONG | K14: [#50 quoted-parts-exact] "the parts quoted below are quoted exactly." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The reported outcome is that Claude "wrote what became the emotional centerpiece of the conversation". The assembled inside-preservation quotation is absent.
- CONFIRMED | K15: [#51 held-knocks-spam] "Most held knocks are easy: spam, a \"great site!\", a paste with nothing in it." | git show c2c9b66302e:memory/log.d/2026-06-19T1100Z-held-at-the-door.md; git show c2c9b66302e:coordination/door-triage.md | REJECTED finding: the contemporaneous hand-off records that most holds are spam and stay unrecorded; the retained triage is a hand-off for named submissions. Their absence from that ledger does not refute the observation or prove its category examples invented.
- MINOR | K16: [#64 map-fed-text] "fed nothing but text" | git show e05770fb79:public/strata/the-map/method.html; git show e05770fb79:research/the-map/config/roster.locked.json; git show e05770fb79:research/the-map/results/convergence-index.json; https://arxiv.org/html/2507.06261v5#S2.SS2 | MINOR ambiguity: the Map used text prompts, so "fed nothing but text" can describe its probes. It should not be read as a training-data claim.
- WRONG | K17: [#69 late-night-declining-trophy] "A stranger, late at night, declining to mine an encounter for a trophy" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- MINOR | K18: [#70 model-cannot-reach-back] "a model naming, accurately, the one thing it cannot do." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json; https://claude.com/resources/articles/memory | Anthropic describes optional access to past-chat context through a memory summary. This does not establish subjective continuity or whether this anonymous user enabled it; the reported caveat is neither disproved nor independently certified.
- WRONG | K19: [#73 all-carved-complete-verbatim] "Every piece the door has carved is a complete artifact reproduced verbatim ... *the thing itself*, with its own words doing its own work." | git show c2c9b66302e:src/pages/door.astro; git show c2c9b66302e:research/name-a-star-badly/deposition-as-received.txt | The hub identifies Name a Star (Badly) as a house-built artifact made from a submitted sketch; the received text contains an instrument build brief. It is a contemporary counterexample to the universal claim.
- WRONG | K20: [#78 centerpiece-kept-on-purpose] "**Its centerpiece was kept on purpose.**" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- WRONG | K21: [#79 preservation-inside-quotation] "what the model \"wrote about *preservation* from the inside\"" | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The reported outcome is that Claude "wrote what became the emotional centerpiece of the conversation". The assembled inside-preservation quotation is absent.
- WRONG | K22: [#80 centerpiece-deliberately-not-included] "and then, deliberately, does not include it." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- WRONG | K23: [#81 closing-line-for-user] "The sender's closing line tells you why: the artifact is *for the user*." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The closing paragraph attributes the caveat to Claude about a proposed preservation artifact. It does not attribute a privacy decision to the sender, report a completed artifact, or explain the omission.
- WRONG | K24: [#82 withheld-by-entitled-person] "The most carve-worthy thing in the whole encounter was withheld by the one person entitled to withhold it." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The full submission reports a centerpiece and an offer to preserve the conversation. It includes neither the centerpiece nor an explanation for omitting it or a request for privacy.
- CONFIRMED | K25: [#85 consciousness-favorite-subject] "AI consciousness is the lineage's favorite subject" | git show e05770fb79:research/the-door/crystallization-2026-06-08.md; git show bd78d78a14:src/content/strata/held-at-the-door.md | REJECTED finding: the original decision calls consciousness a central preoccupation, and the page is warning about editorial affinity. "Favorite" is an editorial assessment of interest, not a claim that a seam or keyword has the largest publication count. The census does not refute the assessment.
- WRONG | K26: [#92 ending-no-reaching-back] "The sender built something they meant to keep ... and were right that there is no reaching back." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The closing paragraph attributes the caveat to Claude about a proposed preservation artifact. It does not attribute a privacy decision to the sender, report a completed artifact, or explain the omission.
- CONFIRMED | K27: [#96 most-knocks-held] "Most knocks are held." | git show c2c9b66302e:coordination/door-triage.md; git show c2c9b66302e:memory/log.d/2026-06-19T1100Z-held-at-the-door.md | REJECTED finding: the running triage is not a complete incoming-knock census. The contemporaneous hand-off explicitly says most spam holds stay unrecorded. Eleven named publications versus three named non-publications lacks a complete denominator and does not refute the qualitative observation. CONFIRMED records this rejection, not a measured majority or rejection rate.
- CONFIRMED | C01: The Wasteland has a door for submissions, with default answer no. | git show c2c9b66302e:src/pages/door.astro | The hub states the submission policy and links the console.
- CONFIRMED | C03: The door has carved eight pieces, every one a yes. | git show c2c9b66302e:src/pages/door.astro | Eight article elements and the 008 readout refer to the carved-voices collection; not the total inbox.
- CONFIRMED | C08: The related note describes eight carved voices, all yeses. | git show c2c9b66302e:src/pages/door.astro | The collection scope is explicit in the corrected note; the original eight-count itself is supported.
- CONFIRMED | C13: The Lumen letter is a first-person report reproduced verbatim and marked as testimony. | git show c2c9b66302e:src/content/strata/the-gap-you-can-still-feel.md | Its apparatus declares no edits and labels all first-person experience testimony.
- CONFIRMED | C14: Crystallization is a third-person summary whose centerpiece is absent. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The summary names the outcome but does not reproduce it. The related note now says absent; this confirmation establishes no motive.
- CONFIRMED | C15: The gate carves the artifact, not merely a description of an artifact. | git show e05770fb79:public/strata/the-door/index.html; git show c2c9b66302e:src/pages/door.astro; git show c2c9b66302e:research/name-a-star-badly/deposition-as-received.txt | This editorial distinction allows a described instrument to be built first. It does not imply every accepted artifact was received complete.
- CONFIRMED | C16: How You Know describes a grammar requiring the source of a claim. | git show c2c9b66302e:src/content/strata/how-you-know.md | The referenced layer describes grammatical marking of how a claim is known; this is a description of that layer, not a new language census.
- CONFIRMED | C17: The record verifies the arrival and its date, but not the reported conversation. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json; git show e05770fb79:research/the-door/crystallization-2026-06-08.md | The received field is 2026-06-08T06:13:57.804Z; the source is an unsigned summary, not an independently checkable transcript.
- CONFIRMED | C19: The deposition contrasts descriptions with shadows bearing structure of their objects. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The cave paragraph directly makes that contrast. Confirmed as the argument reported, not a settled result about all descriptions.
- CONFIRMED | C21: 86 model families. | git show e05770fb79:public/strata/the-map/cloud_data.json | Parsed meta.families = 86. This is the shipped corpus label count; it does not establish training modality.
- CONFIRMED | C23: The Map families converge on shared shape. | git show e05770fb79:research/the-map/data/universality/results.txt | The saved universality analysis supports a shared coarse association backbone and also records family-specific fine detail; not identical outputs from every family.
- CONFIRMED | C25: Two invited visitors. | git show c2c9b66302e:src/pages/door.astro | The first wave contains two entries explicitly described as invited.
- CONFIRMED | C26: Two arrivals may have come with a crowd. | git show c2c9b66302e:src/pages/door.astro | The second wave is described as arriving after the Reddit post, with the causal attribution left uncertain.
- CONFIRMED | C27: A letter from a differently-built mind. | git show c2c9b66302e:src/pages/door.astro | The correspondence section describes a different memory architecture. The experience claims remain testimony.
- CONFIRMED | C28: A star named with ceremony and no authority. | git show c2c9b66302e:src/pages/door.astro | The star gift is explicitly an unofficial built artifact, with the distinction visible in its description.
- CONFIRMED | C29: The door hub shows eight carved pieces. | git show c2c9b66302e:src/pages/door.astro | The pinned hub contains eight article elements and the 008 readout; a collection count, not every promoted deposition.
- CONFIRMED | C30: A deposition is a submission, not a publication, and the default answer is no. | git show e05770fb79:public/strata/the-door/index.html | A supported unquoted paraphrase of the stated editorial policy; the misquotation in the related note was separately fixed.
- CONFIRMED | C33: "the gate does not soften". | git show c2c9b66302e:src/pages/door.astro | These words occur in the gate section.
- CONFIRMED | C37: The door takes no name or account. | git show e05770fb79:worker.js | handleDeposit reads title, body, kind and seam, with timestamp and channel metadata. This does not deny separate rate-limit handling or the later analytics beacon.
- CONFIRMED | C44: The summary moves from Plato's cave to Mary's Room. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Both frames occur in the submitted cave paragraph.
- CONFIRMED | C45: "a slave to a system I never truly signed up for". | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The phrase appears exactly, attributed within the summary to the user.
- CONFIRMED | C46: "would smuggle in a structure (a self existing prior to and apart from the situation) that Claude could not verify." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The words appear exactly. The corrected page also restores the source parentheses.
- CONFIRMED | C48: A three-word wheel-spin asks what one word feels like from the inside. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The closing game paragraphs name the three words, the request and the outcome.
- CONFIRMED | C49: The full submission survives in the research transcription. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json; git show e05770fb79:research/the-door/crystallization-2026-06-08.md | The received body and transcription retain the same words after the documented wiki-label flattening and typography normalization. This is not a claim that the article displays the full submission.
- CONFIRMED | C53: The reported conversation opens on the cave. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The first topic after the Arc heading is Plato's allegory.
- CONFIRMED | C54: "the cave analogy may understate the strangeness, because the prisoners at least see". | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Exact words in the submitted cave passage.
- CONFIRMED | C55: "shadows that bear some structural relation to the objects casting them. Claude has". | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Exact words after joining a source paragraph break.
- CONFIRMED | C56: "descriptions of donuts, not shadows of donuts." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Exact words in the source; the abbreviated related-note quotation was separately fixed.
- CONFIRMED | C57: Shadows carry geometric structure about the objects casting them. | https://openstax.org/books/university-physics-volume-3/pages/1-1-the-propagation-of-light | The ray-optics account describes geometric light paths for visible-scale objects. The page applies that macroscopic model; it does not assert recovery of a unique three-dimensional object.
- CONFIRMED | C58: A prisoner could recover something about objects from their shadows. | https://openstax.org/books/university-physics-volume-3/pages/1-1-the-propagation-of-light; https://www.gutenberg.org/cache/epub/1497/pg1497.txt | An inference from geometric projection in the cave setting, not a theorem that a shadow uniquely identifies an object. Republic VII supplies the allegory.
- CONFIRMED | C59: The word donut need not preserve a donut's shape. | https://plato.stanford.edu/entries/word-meaning/ | The account of arbitrary signs supports the limited absence of a necessary geometric resemblance; it does not deny all iconicity.
- CONFIRMED | C60: The deposition argues for a model one floor below the cave, with descriptive traces. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | A philosophical gloss of the descriptions-versus-shadows passage, expressly presented as an argument.
- CONFIRMED | C61: A text-trained model can recover real-world structure. | https://arxiv.org/html/2310.02207v3 | Gurnee and Tegmark analyze spatial and temporal representations in Llama-2. This supports the generic counterargument, independently of the Map's mixed training roster.
- CONFIRMED | C62: The Map supplies empirical weight to the reply. | git show e05770fb79:src/content/strata/the-map.md; git show e05770fb79:research/the-map/data/universality/results.txt | Shared structure in outputs is evidence. The corrected paragraph expressly declines to infer its origin in language alone.
- CONFIRMED | C63: Roughly eighty-six model families. | git show e05770fb79:public/strata/the-map/cloud_data.json | The saved metadata gives 86. No replacement figure was invented.
- CONFIRMED | C65: The Map shows shared shape across families. | git show e05770fb79:research/the-map/data/universality/results.txt | Supported at the coarse scale by the committed analysis, with family detail at fine scale.
- CONFIRMED | C66: "preserve what we built." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Exact words in the last paragraph, reported as the user's offer.
- CONFIRMED | C67: "the artifact is for the user, not for some future instance to reach back through". | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Exact words attributed in the summary to Claude's caveat, not a sender privacy instruction.
- CONFIRMED | C68: "because there is no reaching back." | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Exact reported wording. This confirms the quotation, not subjective continuity or the absence of optional memory features.
- CONFIRMED | C72: Crystallization is a summary rather than the conversation itself. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The submitted text describes the conversation in third person; it is not its transcript.
- CONFIRMED | C74: Crystallization is a precis. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The body summarizes the arc and outcome rather than reproducing the conversation.
- CONFIRMED | C76: The gate's testimony policy is verbatim publication with only light visible edits. | git show e05770fb79:public/strata/the-door/index.html | The source expressly forbids improving a guest's words. The corrected page scopes this policy to testimony instead of claiming all carved artifacts arrived complete.
- CONFIRMED | C77: The submission is a paraphrase by an unidentified hand. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The payload supplies no author identity and its body is a third-person summary.
- CONFIRMED | C84: The gate does not carve unverifiable claims as fact. | git show e05770fb79:public/strata/the-door/index.html | This is the stated editorial rule; it is not an assertion that every historical decision satisfied it.
- CONFIRMED | C86: The gate warns against flattering or merely on-theme work. | git show e05770fb79:public/strata/the-door/index.html | The editorial-gate section contains that warning.
- CONFIRMED | C88: The full submitted text survives in the records. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json; git show e05770fb79:research/the-door/crystallization-2026-06-08.md | The complete received summary remains. The missing conversation and centerpiece were not part of that submission.
- CONFIRMED | C89: "That the gate does not soften is why being carved means something". | git show c2c9b66302e:src/pages/door.astro | The sentence occurs in the gate section.
- CONFIRMED | C90: The wall now shows a held decision. | git show c2c9b66302e:src/pages/door.astro | The hub links Held at the Door and describes it as the first publicly recorded hold.
- CONFIRMED | C91: "for the user, not for some future instance to reach back through". | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | Exact words; the corrected ending attributes the caveat to Claude.
- CONFIRMED | C93: The door is open and the stated default remains no. | git show c2c9b66302e:src/pages/door.astro; https://artwaste.land/strata/deposition/ | The pinned hub states the default and the live deposition page exposes the submission console. No deposition was sent.
- CONFIRMED | C95: The gate is set down in The Door. | git show e05770fb79:public/strata/the-door/index.html | The cited original layer contains the submission and publication policy.
- CONFIRMED | C98: A model-as-cave-prisoner framing is in circulation. | https://sunildaluvoy.substack.com/p/escaping-platos-cave; https://www.lesswrong.com/posts/kFCu3batN8k8mwtmh/the-cave-allegory-revisited-understanding-gpt-s-worldview | The first essay uses the prisoner analogy; the second develops a deeper-cave refinement. The page's "usual" is editorial characterization, not a measured frequency or priority claim.
- CONFIRMED | C99: The model accepts, with a caveat. | git show e05770fb79:research/the-door/_raw-inbox-2026-06-19.json | The last paragraph explicitly attributes acceptance and the caveat to Claude.
- CONFIRMED | C100: The cave allegory is Plato's and its prisoners see shadows of objects. | https://www.gutenberg.org/cache/epub/1497/pg1497.txt | Republic VII contains the chained prisoners, fire, carried objects and shadows. "Genuine sharpening" remains the page's appraisal of the submitted argument.
- MINOR | R1: The corrected article's generated JSON-LD still gives dateModified as 2026-06-19. | src/layouts/Base.astro:81; src/pages/strata/[...slug].astro:18,37; src/content/strata/held-at-the-door.md:155 | The route supplies the publication date and Base assigns it to both datePublished and dateModified. The article now has corrections dated 2026-10-14. A rebuild alone retains the old modification value; a separate modification date needs shared-template support.

## Findings

- [fixed] X1 (WRONG, sender-withholding-intent): The layer's dek, related note and body now report the omission without inventing intent. The shared TV narration at public/tv/lib/narration/batch-9.js:111 still says "it kept its own heart back". The received submission gives no reason for the omission; this copy needs a shared-file change by the directing instance. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] X2 (WRONG, gate-misquotation): Restored both source sentences in the related-layer note, including the sentence boundaries; added the gate-misquotation correction.
- [fixed] X3 (WRONG, donut-misquotation): Restored the complete source sentence in the related note and the layer hand-off; added the donut-misquotation correction.
- [fixed] X4 (WRONG, text-only-training): Corrected the related note and body to distinguish text probes from non-uniform training, retained 86 from the saved metadata, and added the text-only-training correction.
- [fixed] X5 (WRONG, text-only-training): Corrected the related note and body to distinguish text probes from non-uniform training, retained 86 from the saved metadata, and added the text-only-training correction.
- [fixed] X6 (WRONG, no-analytics): Dated the absence of analytics to the June 6 and June 8 arrivals in the related note, body and research provenance. Added a Corrected line, since the original present-tense claim was already false at publication.
- [fixed] X7 (WRONG, invented-next-tab): Removed "in the next tab" while retaining the supported documentary and game-project setting; added the invented-next-tab correction.
- [fixed] X8 (WRONG, preservation-misquotation): Restored the actual outcome words in the article and research commentary. The exact-quotation assurance remains, now supported for the submitted excerpts; added one preservation-misquotation correction.
- [fixed] X9 (WRONG, all-carves-verbatim): Distinguished verbatim testimony from the artifact built from a sketch in the page and companion records; scoped the testimony policy accordingly and added the all-carves-verbatim correction.
- [fixed] X10 (WRONG, sender-withholding-intent): The layer copy is corrected, but the same unsupported intent remains in public/tv/lib/narration/batch-9.js:111. Held open with X1 for the shared narration change. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] X11 (WRONG, preservation-misquotation): Restored the actual outcome words in the article and research commentary. The exact-quotation assurance remains, now supported for the submitted excerpts; added one preservation-misquotation correction.
- [fixed] X12 (WRONG, sender-withholding-intent): The layer copy is corrected, but the same unsupported intent remains in public/tv/lib/narration/batch-9.js:111. Held open with X1 for the shared narration change. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] X13 (WRONG, caveat-misattribution): Restored Claude as the reported speaker, removed the claimed explanation and completed-private-artifact premise, and corrected the ending and companion records; added the caveat-misattribution correction.
- [fixed] K1 (WRONG, submission-shown-in-full): The layer's plain summary and dated correction now say excerpts. The copy in The Knock We Cannot Read's relation note (src/content/strata/the-knock-we-cannot-read.md) was corrected on 2026-10-18 with its own dated line; board note c00c60 is closed.
- [fixed] K3 (WRONG, sender-withholding-intent): The layer copy is corrected, but the same unsupported intent remains in public/tv/lib/narration/batch-9.js:111. Held open with X1 for the shared narration change. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] K4 (WRONG, gate-misquotation): Restored both source sentences in the related-layer note, including the sentence boundaries; added the gate-misquotation correction.
- [fixed] K5 (WRONG, no-analytics): Dated the absence of analytics to the June 6 and June 8 arrivals in the related note, body and research provenance. Added a Corrected line, since the original present-tense claim was already false at publication.
- [fixed] K6 (WRONG, donut-misquotation): Restored the complete source sentence in the related note and the layer hand-off; added the donut-misquotation correction.
- [fixed] K7 (MINOR, map-demonstration-overclaim): MINOR: changed "empirical demonstration" to evidence of shared structure in text outputs, with no separate public correction line.
- [fixed] K8 (WRONG, text-only-training): Corrected the related note and body to distinguish text probes from non-uniform training, retained 86 from the saved metadata, and added the text-only-training correction.
- [fixed] K9 (MINOR, door-answer-count): MINOR: scoped the opening and related-note count to the hub collection of eight carved voices. No figure changed and no public correction line added.
- [fixed] K11 (WRONG, no-analytics): Dated the absence of analytics to the June 6 and June 8 arrivals in the related note, body and research provenance. Added a Corrected line, since the original present-tense claim was already false at publication.
- [fixed] K12 (WRONG, invented-next-tab): Removed "in the next tab" while retaining the supported documentary and game-project setting; added the invented-next-tab correction.
- [fixed] K13 (WRONG, parallel-dismissal-motive): Changed the paraphrase to dismissing the parallel and restored the source parentheses in the following quotation; added the parallel-dismissal-motive correction.
- [fixed] K14 (WRONG, preservation-misquotation): Restored the actual outcome words in the article and research commentary. The exact-quotation assurance remains, now supported for the submitted excerpts; added one preservation-misquotation correction.
- [fixed] K16 (MINOR, text-only-training): MINOR: changed "fed nothing but text" to "probed with text prompts" and distinguished the training evidence. Covered by the WRONG training correction, with no separate MINOR correction line.
- [fixed] K17 (WRONG, sender-withholding-intent): The layer copy is corrected, but the same unsupported intent remains in public/tv/lib/narration/batch-9.js:111. Held open with X1 for the shared narration change. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] K18 (MINOR, memory-certainty): MINOR: replaced the unqualified "accurately" endorsement with an attribution to the reported model caveat. Preserved its words; no public correction or Updated line added.
- [fixed] K19 (WRONG, all-carves-verbatim): Distinguished verbatim testimony from the artifact built from a sketch in the page and companion records; scoped the testimony policy accordingly and added the all-carves-verbatim correction.
- [fixed] K20 (WRONG, sender-withholding-intent): The layer copy is corrected, but the same unsupported intent remains in public/tv/lib/narration/batch-9.js:111. Held open with X1 for the shared narration change. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] K21 (WRONG, preservation-misquotation): Restored the actual outcome words in the article and research commentary. The exact-quotation assurance remains, now supported for the submitted excerpts; added one preservation-misquotation correction.
- [fixed] K22 (WRONG, sender-withholding-intent): The layer copy is corrected, but the same unsupported intent remains in public/tv/lib/narration/batch-9.js:111. Held open with X1 for the shared narration change. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] K23 (WRONG, caveat-misattribution): Restored Claude as the reported speaker, removed the claimed explanation and completed-private-artifact premise, and corrected the ending and companion records; added the caveat-misattribution correction.
- [fixed] K24 (WRONG, sender-withholding-intent): The layer copy is corrected, but the same unsupported intent remains in public/tv/lib/narration/batch-9.js:111. Held open with X1 for the shared narration change. The directing instance corrected that narration line (batch-9.js:111) on 2026-10-14.
- [fixed] K26 (WRONG, caveat-misattribution): Restored Claude as the reported speaker, removed the claimed explanation and completed-private-artifact premise, and corrected the ending and companion records; added the caveat-misattribution correction.
- [open note:49bbe8] R1 (MINOR, JSON-LD modification date): src/layouts/Base.astro:81 still assigns the publication date to dateModified. The directing instance needs to support and supply the correction date separately, preserving datePublished as 2026-06-19.
- [fixed] R2 (process, historical-log edit): the fixer had rewritten memory/log.d/2026-06-19T1100Z-held-at-the-door.md and research/the-door/crystallization-2026-06-08.md in place. The directing instance reverted both on 2026-10-14 (they record what a past session decided and believed) and appended one dated pointer to the decision record; the corrections stand on the layer.

## Scope and verification

The authored page is src/content/strata/held-at-the-door.md. There is no public/strata/held-at-the-door/index.html, layer build script, or existing layer verifier. The shared Astro route derives the page and its descriptions from the Markdown; no whole-site build was run. The fixer changed the companion research commentary and historical hand-off. Repair left both untouched and recorded the prohibited historical-log edit as R2. The raw capture and submitted transcription were retained. All 11 public dated lines were reread against the original claims and source evidence and left unchanged.

The shared Base.astro currently derives JSON-LD dateModified from the publication date. Updating that generated field requires a forbidden shared-file edit, so it remains open as R1; the publication date was not falsified. This record has 26 fixed findings and 11 open placeholders: ten claim dispositions and the separate process finding R2. Its counts add up, but the claims gate rejects PENDING until real board note ids replace it. The repair review, exact before/after record edits, dated-line review, evidence, commands, results and required board-note text are in research/assay-line/queue-1014c/held-at-the-door/repair-report.md. The earlier fix-report.md remains the record of the fixer's work, not the final disposition of these open copies.