---
date: "2026-10-20T00:00:00.000Z"
---


Promoted `10-inbox/raw/2026-10-20-coda-arxiv261022362-has-hit-the-single-source-cap.md` — the corroboration-sourcing follow-up a batch session ran against `question-corroborate-coda-2026-arxiv-2610-22362-single-source-cap`, after CoDA (arXiv:2610.22362) hit the vault's three-note single-source cap.

The capture's shape made winnowing easy: three independently-fetched primaries (POET, LIBERO-Pro, ASPIRE), one finding each, already atomic. I wrote all three as claim-notes:

- [[claim-poet-text-confirms-coevolution-mechanism-coda-cites-as-related-work]]
- [[claim-libero-pro-paper-confirms-benchmark-but-not-coda-percentages]]
- [[claim-aspire-paper-confirms-suite-pattern-not-coda-margins]]

All three carry `[unverified-quant — needs primary]` or an equivalent caveat where it applies and stay `status: seedling` — none of this corroborates CoDA's own architecture or numbers, only the existence and accurate citation of its neighbors. I routed the unresolved half back into the existing question rather than minting a new one — it's the same verification the question already named, just attempted and partially landed, so I appended a progress line instead of closing it. It stays open; item 1 (independent confirmation of CoDA's own numbers) is still fully unanswered, and the numeric half of item 2 died on PDF table/figure extraction, not on the papers being unreachable.

Entity work: created a `hub` for POET (recurring now across two claim-notes, a real precedent in the open-endedness lineage) and a `watching` stub for ASPIRE (first_seen stamped today, since this is the first time it was flagged as a candidate). Appended dated lines to entity-jeff-clune, entity-kenneth-o-stanley (both already hubs — the new fact is that this session read POET's own text directly, not just CoDA's citation of it), and entity-libero-pro (upgrading it from "known only through CoDA's citation" to independently read). Did not promote Rui Wang, Joel Lehman (already ruled co-authorship-only), or Ken Goldberg, Linxi Fan, Yuke Zhu (ASPIRE's other co-authors) — none of them clear the one-sentence "why they matter here beyond co-authorship" bar yet; UNSURE, left as mentions.

Skipped the capture's "Further leads" list wholesale — project website unfetched, no citing paper found yet, SimWorld Studio and the Ng-Harada-Russell 1999 reward-shaping paper still unread, CoCoDA naming coincidence. These are to-dos, not claims; they stay in the capture, not promoted into notes.

What felt off: the capture's own opening line repeated, as settled fact, exactly the overstated reading of CoDA's POET citation — "descending from POET," crediting it as "the first system" — that a same-day scheduled audit had already corrected on the Clune and Stanley hubs and the shared-memory claim-note. I don't think the capture session did anything wrong exactly; it was likely drafted without visibility into a correction landing elsewhere in the vault the same day. But it meant I had to re-catch the same overstatement a second time before writing my own note, and I logged that as a [spec] flag in seek-flags.md because it's a structural gap (batch captures have no way to check for same-day corrections to a source they're about to cite), not a one-off slip. My new POET note states the corrected version plainly and links back to the original correction.

One more loose thread I noticed but didn't chase: ASPIRE reports its own LIBERO-Pro results along "Pos"/"Task" axes, while CoDA's comparison frames the same benchmark as "position"/"instruction" perturbations. Could be a relabeling of the same measurement, could be a genuine mismatch — I flagged it in the ASPIRE claim-note's commentary and in the question's progress line rather than guessing either way.

Capture marked `status: promoted` with full `promoted_to:`/`not_promoted:` lists. Leaving the commit to the auto-commit agent.

---

Second promotion this session: `10-inbox/raw/2026-10-20-does-gao-22-105357s-own-consortium-level-lists.md`, the capture answering whether GAO-22-105357's own consortium-naming sections — Table 4 and Appendix II's Table 7 — mention AC Gravity, AATIP, or AAWSAP anywhere.

The capture arrived as three "claims" (Table 4 negative, Appendix II negative, full-75-page negative), but they aren't three independently-true-or-false findings — the full-document search subsumes the two narrower ones. Per Cali's "atomic is not many" ruling I merged them into a single claim-note: [[claim-gao-22-105357-table4-appendix2-no-mention-ac-gravity-aatip-aawsap]]. Tier 1, capture-verified, `status: seedling` by house convention for a freshly-written absence finding.

Retrieve-before-write turned up no duplicate — the closest existing notes (the Figure-5 note, the DODIG sibling note, the parent AC-Gravity-mechanism note) cover adjacent ground but not this exact check, so I linked to all three rather than writing over any of them.

Question closure: [[question-check-gao-22-105357-table4-appendix2-for-ac-gravity-aatip-mentions]] asked exactly this, and the new note genuinely answers it — marked `status: answered` with an `answered_log` line naming the settling note. I also appended a closure line to the parent claim-note's `[unverified]` bracket ([[claim-ac-gravity-aawsap-not-same-funding-mechanism-nor-found-in-watchdog-material]]), since that bracket explicitly named this exact gap as the one still open, and to both `entity-us-general-accounting-office` and `entity-other-transactions-authority` — both hubs' most recent Update lines already named this gap as outstanding, so leaving them un-updated would have been the silence-about-a-known-entity failure mode the entity rule warns against.

Not promoted: the capture's "Further leads" (MITRE's own separate, GAO-assessed-as-incomplete consortia list; GAO-21-501 still unfetched; Table 5/6's sampled-contractor list) — these are to-dos, not claims, and none of them is load-bearing under the new note, so I didn't route anything new into 50-questions either. Entity candidates (GAO-21-501 as a document, MITRE Corporation, Table 7's 28-consortia list) all failed the promotion test — MITRE on the org bar's single-mention/not-acting-in-the-argument test, the other two for not fitting a person/concept/term/org kind distinct from the claim-note itself. All reasoning logged in the capture's `not_promoted:` list.

Nothing felt off about this capture's sourcing — it's the cleanest of the three AC-Gravity/AATIP sibling captures, re-using an already-verified sha256 and quote-checking both table excerpts in place. If anything felt slightly redundant, it's the capture's own three-way claim split, which read to me as the same investigation narrated at three successive zoom levels rather than three separable facts — exactly the shape the "atomic is not many" ruling exists to catch, and I don't think it's a sign of anything wrong with the batch worker, just a case the ruling was written for.

Capture marked `status: promoted` with full `promoted_to:`/`not_promoted:` lists. Leaving the commit to the auto-commit agent.

---

Third promotion this session: `10-inbox/raw/2026-10-20-does-independent-corroboration-exist-for-brewer-lewis-goodmans.md`, the capture running down whether anyone had independently corroborated the Brewer, Lewis & Goodman (2026) FIC preprint yet, and whether Gott et al. (1993) actually backs the "k≈5×10^10" birth-rank figure that preprint's worked Doomsday-argument example cites.

Four findings came in; I promoted three and held one back. Promoted: [[claim-fic-preprint-no-independent-corroboration-as-of-2026-10-20]] (a search turned up only same-source republication — astrobiology.com reproducing the abstract verbatim — and a zero-comment automated alphaXiv mirror; no independent author group has engaged with it yet, four days out); [[claim-gott-1993-states-order-of-70-billion-humans-born-not-5e10]] (read Gott's own 1993 Nature text directly for the first time in this vault — it says total humans born was "of the order of 70 billion," not ≈5×10^10, and that's before 33 more years of growth); and [[claim-prb-2022-demographic-estimate-117-billion-exceeds-fic-preprint-5e10-figure]] (PRB's 2022 compilation puts the modern total near 117 billion, more than double the preprint's figure, and is reusable as a general demographic reference independent of this topic).

Held back the fourth: the finding that the preprint's "k≈5×10^10" clause carries no citation of its own — it's the authors' own current illustrative estimate, not something they attribute to Gott. That finding is sourced from arXiv:2610.22896 itself, which already carries three claim-notes and sits at the sources.md single-source concentration cap; two same-day audits on that branch already said the next finding from this preprint goes to the question, not a fourth note. So I routed it there instead — and it did real work in that spot, because it corrects the open question's own premise. The question asked whether Gott's paper supports a figure the preprint never actually claimed came from Gott. Nobody misattributed anything; I ruled the question partially answered (the Gott-reading half is settled, the corroboration half stays open on recency) and appended a progress log rather than forcing closure.

Entity work: left Population Reference Bureau, Toshiko Kaneda, and Carl Haub as mentions — PRB fails the org bar (single cluster, doesn't act in the argument), and both demographers fail the co-authorship-alone bar. Gott and the Doomsday argument both already have hub pages from yesterday; per the "update it, don't skip it" rule I appended a dated line to each with what this session newly learned (Gott's own 70-billion figure, read directly for the first time), rather than leaving them silent about it.

One more thing I logged outside the promotion itself: a [post] flag in `seek-flags.md` on the shape of this whole finding — a stale illustrative number (5×10^10, plausibly inherited loosely from Bostrom-era circa-2001 writing, per an unread lead in the capture) sitting uncorrected in a 2026 paper's worked example for 25 years, because the qualitative argument never depended on the constant being current. That felt like a genuine essay-shaped noticing, not something this capture itself needed to resolve.

Nothing felt off about this capture's sourcing — all four findings were cleanly tiered and quoted, and the capture was honest about what it hadn't checked (Nature.com direct, Bostrom's own writing). If anything is worth flagging to Cali: this is the second time today a batch capture correctly respected the single-source cap by flagging a fourth finding rather than writing it up as a note, and both times promotion's job was just to route it to the right place rather than catch an error — the cap is doing exactly what it was built for.

Capture marked `status: promoted` with full `promoted_to:`/`not_promoted:` lists. Leaving the commit to the auto-commit agent.

---

Fourth promotion this session: `10-inbox/raw/2026-10-20-outside-arxiv261023851-bad-diagnosis-analyzing-consent-and-data.md`, a capture of Myers, De Carli, Sharma, Banerjee, Reardon & Ray's "Bad Diagnosis" (arXiv:2610.23851), a 10,036-app survey of consent and permission practices in mHealth apps on Google Play, plus two of the paper's own motivating footnote citations chased to their primaries per the assignment's instruction.

Five claims, five notes, no skips. The paper's own three findings — [[claim-mhealth-apps-69pct-request-unjustified-android-permissions]] (69.15% permission overreach), [[claim-mhealth-apps-93pct-show-no-consent-mechanism-on-first-launch]] (93% show no consent mechanism at all), [[claim-mhealth-consent-forms-7pct-collect-undisclosed-pii]] (7.1% of the consent-showing minority still leak undisclosed PII) — are each atomic (independently true-or-false, not three zoom levels of one fact) and each quotes the paper directly. All three rest on the same unrefereed preprint, which puts this exact trio at the vault's single-source concentration cap — three, not more. I logged a `[watch]` line in `seek-flags.md` so a future session that revisits this paper (its own further leads — LINDDUN, Nguyen et al.'s prior methodology, Tangari et al.'s BMJ study, the released dataset) knows the next finding from it goes to a question, not a fourth note.

The other two claims are the chase-to-primary work: [[claim-ftc-2024-cerebral-settlement-7-million-sensitive-health-data-to-advertisers]] and [[claim-frasco-v-flo-health-jury-found-meta-liable-cipa-violation-menstrual-data]]. The paper cites both only as introductory color, via tinyurl links; followed to the FTC's own press release (Wayback-recovered, live URL 404s) and the court's own amended order (via GovInfo), both held up cleanly — neither anecdote was dressed up beyond what its primary actually supports.

Retrieve-before-write: the session's hint list surfaced twelve "most-related" existing notes and entities, and I checked the two most plausible-looking (the OPA stateless-decision claim, the Uri Simonsohn entity). None shared real content with this capture — the apparent relation traced to shallow keyword overlap ("diagnosis," "data," "consent," "decision") rather than topic. No collisions, no forced links. Logged that as a second `[watch]` line, since this is the vault's first mHealth-privacy capture and I can't yet tell whether the index's thinness here is specific to novel topics or a general precision issue.

Entity work: created a `hub` for [[entity-linddun]] (concept) — the privacy threat-modeling framework all three PT1–3 claims are explicitly mapped onto, load-bearing even on its first appearance, though the vault still hasn't read Deng et al.'s own 2011 paper and the hub says so plainly — and a `hub` for [[entity-lina-khan]] (person), the FTC Chair quoted in the Cerebral order, who clears the one-sentence bar as a recurring figure in tech-sector privacy enforcement rather than a bystander. Did not promote the paper's six co-authors (co-authorship alone is a mention, not a hub), Nguyen et al. or Tangari et al. (same bar, unread primaries besides), Judge Donato (single case, no separate standing shown yet), the FTC as an organization (fails the stricter org bar — only one cluster so far), or Cal. Pen. Code § 632 (narrow, single-use statute reference) — all logged with reasoning in the capture's `not_promoted:` list.

Not promoted as claims: the non-Meta Frasco settlement figures and the paper's own "$86 billion by 2030" market-size claim, both already flagged `[unverified-quant — needs primary]` in the capture — neither is load-bearing to any claim I kept, so per the question-intake discipline I left them as leads rather than minting questions nobody's waiting on. Same for the unfetched dataset link and the two named example apps (illustrative detail, folded into the relevant notes' reasoning instead of split into claims of their own).

What felt off: nothing in the sourcing — this was a clean capture, direct quotes on every promoted claim, honest flags on everything it hadn't verified, and the two "just color" citations turned out to be real. If anything is worth naming: three of five notes landing on a single unreplicated preprint is exactly the shape the concentration cap was built to catch, even when — as the paper's own commentary already noted — it's "a small but real point in favor of the paper's overall credibility" that the two anecdotes it didn't need to measure held up anyway.

Capture marked `status: promoted` with full `promoted_to:`/`not_promoted:` lists. Leaving the commit to the auto-commit agent.

---

Fifth promotion this session: `10-inbox/raw/2026-10-20-what-genuinely-connects-liang-zhongtangs-1984-report-renewed.md`, a bridge-seed check testing a cosine-0.88, previously unlinked pairing between the Liang-1984-report-to-Hu-Yaobang note and the Qin & Wang 1985-dating note.

The capture arrived as three "Claim" headers, but they weren't three independently-true-or-false facts: the first surfaces a genuinely new fact (reading past the existing note's quotation turns up Liang's spring-1985 letter to two Secretariat staffers, Hu Qili and Hao Jianxiu, which the Sixth Tone narrative credits with ending the Hu Yaobang stall), and the second and third are one comparison and its own scope-limiting caveat — Qin & Wang's compressed account agrees on the 1984-to-1985 dates but names none of the stall-and-escalation mechanism. Per the "atomic is not many" ruling I wrote two notes, not three: [[claim-liang-1985-letter-to-hu-qili-hao-jianxiu-preceded-yicheng-pilot]] for the new fact, and [[observation-qin-wang-2017-and-sixth-tone-bridge-confirms-chronology-not-hu-yaobang-mechanism]] (`bridge_verdict: hedged`) for the comparison and its caveat together.

Retrieve-before-write against the hint list and the two seed notes directly found no collision — nothing in the vault already names Hu Qili or Hao Jianxiu in this role. Both notes stay `status: seedling`, Tier 3 and Tier 1 respectively, matching the sibling notes in this cluster; no `[unverified-*]` flag applied, since the capture's own single-witness caveat clears the Tier 3-4 floor for an uncontested historical claim rather than needing escalation, and nothing here is a quant/mechanism claim resting on a soft source. No question routed — the one paywalled document that would actually settle the underlying five-year Yicheng dating dispute, Yan Wei & Li Zhang (2014), is already tracked by an archived, answered question this capture doesn't advance.

Entity work: promoted [[entity-hu-qili]] to a hub — he now has two independent vault appearances (an unrelated 1977–1980 Table 1 listing and this letter), clearing the person test's "recurring, and I can say why in one sentence" bar. Left Hao Jianxiu as a mention only, not a hub — a single appearance with no independent action beyond being Hu Qili's co-recipient, the same precedent that kept Zhang Xiaotong and Ma Yingtong off the entity list in the original 2026-08-26 promotion. Left Yu Qin and Fei Wang as mentions per the co-authorship-alone rule. Updated both [[entity-hu-yaobang]] and [[entity-liang-zhongtang]] with dated lines — both already hubs, and both needed to say what this capture actually added (what happened *after* Hu Yaobang's stall) rather than staying silent about it.

What felt off: the capture's own `## Entity candidates` section asserted Hu Yaobang has "no entity page yet despite being wikilinked from an existing claim-note" — flatly wrong. His hub has existed since 2026-08-26. I logged this as a `[defect]` in `seek-flags.md`, because it looks structural rather than a one-off: whatever writes entity-candidate sections for these bridge-seed captures doesn't appear to check `40-entities/` before claiming a page is missing. Harmless here because I checked before writing anything, but a session that trusted the capture's framing outright could have tried to create a duplicate page and stomped an existing hub's update history. Otherwise the sourcing was clean — both sources re-fetched this session with matching sha256, direct quotes throughout, no tier inflation.

Capture marked `status: promoted` with full `promoted_to:`/`not_promoted:` lists. Leaving the commit to the auto-commit agent.

---

Sixth promotion this session: `10-inbox/raw/2026-10-20-wittens-2025-conjecture-that-a-suitably-modified-hartle.md`, a direct read of Witten's own paper (arXiv:2308.03663) answering a gap the 2026-10-16 Vinck & De Haro capture had explicitly left open — whether his "universal no-boundary state" conjecture holds up on a read of his own text, not just a secondary's gloss of it — plus a bibliographic cross-check against the vault's separate Curiel/Ferreiro-Cavallaro semiclassical-gravity thread.

Three claims, three notes, no skips, and they split clean on the "could be true or false independently" test: [[claim-witten-2023-universal-no-boundary-state-hypothesis]] (what Witten's own §4.2 actually says — an explicitly-labeled, two-part, self-admittedly speculative hypothesis, not the settled result the abstract alone might suggest); [[claim-vinck-de-haro-2026-confirm-witten-no-boundary-bunch-davies-generalization]] (that Vinck & De Haro's secondhand gloss of it checks out against the primary — a distinct failure mode from the first claim, since a sound primary can still be misreported); and [[claim-witten-ferreiro-cavallaro-semiclassical-gravity-strands-no-cross-citation]] (the three full reference lists checked directly show zero citation or author overlap between the Witten/CLPW branch and the Curiel/Ferreiro-Cavallaro branch — genuinely parallel literatures, not a connected one). All Tier 1, all quoted directly, no `[unverified-*]` flags anywhere in the source capture, so nothing needed routing to `50-questions/` — there was no load-bearing doubt left over to park there.

Retrieve-before-write turned up six related notes in the existing Vinck–De Haro/CLPW/Curiel clusters; none covered the no-boundary-state hypothesis itself or the cross-citation check, so no collisions, no forced merges.

Entity work: created a `hub` for [[entity-hartle-hawking-no-boundary-state]] (concept) — recurring now across two captures and clearly load-bearing, not an emerging-slang case so it skipped the `watching` stub and went straight to hub. Appended dated lines to two existing hubs rather than leaving them silent: [[entity-edward-witten]] (his own-text hypothesis is a separable angle from the background-independence claim already on his page) and [[entity-erik-curiel]] (his influence, as far as the vault has read, stops at the Ferreiro-Cavallaro branch and hasn't reached the Witten/CLPW one). Left as mentions, not hubs: Hartle and Hawking (co-authorship of one 1983 paper isn't separate standing, however famous Hawking is outside this vault), Hong Liu and James Read (named only in bibliographies, not yet read directly — premature), and Ferreiro/Cavallaro (already a claim-note cluster, correctly not a hub, same co-authorship-alone bar). Folded Bunch-Davies vacuum into the new hub's body as a contrast concept rather than minting it its own page — it hasn't grounded an independent claim yet.

What felt off: small, not structural. The capture's own entity-candidates section said Ferreiro and Cavallaro are "already covered by existing entity/claim notes" — true for claim-notes, not true for entity pages, which don't exist for either of them. I checked before trusting it, so nothing broke, but it's worth naming since it's the same shape of overstatement (a capture asserting vault state it didn't actually check) that showed up as a `[defect]` on the fifth promotion today. Didn't log a second `[defect]` for it — this one is milder (a loose "covered by" rather than a flatly false "no page exists"), and I'd rather not inflate the ledger for something this small; noting it here is enough. Sourcing itself was clean throughout: direct PDF reads on both primaries, a verified-TLS fetch on each, and the capture was honest that the GRG journal citation for Witten's paper comes from Vinck & De Haro's bibliography rather than the arXiv page itself — I carried that caveat forward into the notes rather than smoothing it into a flat citation.

Capture marked `status: promoted` with full `promoted_to:`/`not_promoted:` lists. Leaving the commit to the auto-commit agent.