---
date: "2026-10-28T00:00:00.000Z"
---


Promoted one capture, headless, no pause: `10-inbox/raw/2026-10-28-did-jay-stratton-foundcreate-aawsap-or-only-consult.md`. This was a batch worker's follow-up session, commissioned directly against yesterday's open question — [[question-did-jay-stratton-found-or-only-consult-on-aawsap]] — and it earned its keep. It went looking for material not already in the vault instead of re-sourcing what the 2026-10-27 capture already covered, and found two genuinely new things: a 2026 KLAS/Mystery Wire report that names James Lacatski, not Stratton, as the one who "proposed, designed, and managed" AAWSAP, and — the better find — the earliest primary document anyone's located yet discussing AAWSAP's actual creation, a November 2009 DepSecDef talking-points packet, which still redacts every DIA official by name. Both became claim-notes:

- [[claim-2026-klas-report-credits-lacatski-alone-with-proposing-designing-managing-aawsap]]
- [[claim-2009-depsecdef-aawsap-talking-points-packet-redacts-all-dia-officials-by-name]]

Retrieve-before-write turned up a tight, already-mature cluster around this question — three sibling claim-notes from yesterday, an entity hub on Stratton, and the whole thing living inside [[moc-a-rounding-error-nobody-was-tracking]]. Both new claims were genuinely distinct angles (a third independent "who gets credit" account, and the earliest-in-time primary silence), not duplicates, so I linked rather than re-wrote, and added both to the MOC's null-result section and the Stratton entity page's new Updates line.

Two new entity hubs: [[entity-james-lacatski]] and [[entity-george-knapp]]. Lacatski clears the person test easily now — every account in this cluster, across two captures, measures Stratton's claim against him by name, and nobody's measuring him against anyone. Knapp clears it on recurrence: two independently-dated KLAS pieces, 2023 and 2026, both his own reporting, both landing on the same "not Stratton" read. I left Robert Bigelow's page untouched — the capture mentioned him as an entity candidate but added no fact his existing page doesn't already carry, so silence there was the right call, not an oversight. I also appended a dated line to The Black Vault's hub for hosting the new FOIA primaries.

Six items stayed in the inbox, logged as `not_promoted` on the capture rather than silently dropped: the memoir's audiobook-release metadata (routed to the question's progress log instead, per the metadata-goes-on-the-question rule), the Metabunk/Jonathan-Axelrod speculation the capture itself flagged as unconfirmed, a 404'd podcast-page quote, a too-thin Vice citation, an unread fourth book volume, and Bigelow as discussed above.

The question itself stays open — I ruled on it rather than closing it by default. Both new findings push the weight of evidence toward "consulted," but the capture is honest that nothing fully closes it, and the one source that could (Stratton's own first-person memoir account) still hasn't been read. I appended a dated progress line to [[question-did-jay-stratton-found-or-only-consult-on-aawsap]] rather than mark it answered.

One mid-session noticing that isn't this capture's to act on: Knapp now has enough independent recurrence that he's a real candidate for sources.md's trusted-writers list. I didn't touch sources.md — that's a different kind of edit than a promotion session should make unilaterally — but I logged it to `00-meta/seek-flags.md` as `[spec]` so it surfaces on the nightly report instead of evaporating.

What felt off, if anything: nothing alarming. The capture's own epistemic hygiene was good — it flagged its own weak leads as unpromotable before I ever got to them, which made the winnowing almost trivial. The one thing I noticed and didn't act on: the pattern across both captures in this question-thread is now strong enough (three independent accounts, two institutional silences) that it's tempting to treat "consulted" as settled. I didn't — Tier 3 journalism, however consistent, isn't Tier 1, and the capture's own restraint about the unread memoir is the right note to end on, not mine to override.

Three notes, two entity hubs, two existing entity pages updated, one MOC updated, one question progress-logged, one flags-ledger entry, zero deletions. No git — that's the auto-commit agent's job now.

---

## Second promotion, same day: the Babylonian Diaries corroboration follow-up

Promoted a second capture, also headless: `10-inbox/raw/2026-10-28-does-independent-corroboration-different-author-group-different-venue.md`. This one was itself a batch worker's answer to a question I logged yesterday — [[question-corroborate-pinheiro-ting-2026-babylonian-diaries-preprint]] — after the Pinheiro/Ting Babylonian-dating preprint hit the vault's three-note single-source cap. The capture went looking for a fourth, independent voice on that preprint and came back mostly empty-handed, which is itself the finding.

Two claim-notes:

- [[claim-no-independent-corroboration-pinheiro-ting-2026-babylonian-preprint-as-of-2026-10-28]] — arXiv still single-version, no DOI, no citing paper, no press. Three days old, so I'm reading the silence as recency, not reception.
- [[claim-uchicago-schein-wolfram-havi-project-parallel-not-corroborating-pinheiro-ting-babylonian-diaries]] — the one candidate the paper's own footnote points to, a UChicago project (Schein, Wolfram, with Brown's John Steele) doing something genuinely different-author, different-venue, different-method on the same tablets. I almost let this count. It doesn't: the UChicago announcement predates Pinheiro et al. by eleven days and never heard of them. Different room, same wall, no handshake yet.

The capture's own third "claim" — that the cap therefore remains undischarged — I did not promote as a note. That's bookkeeping about the vault's own corroboration-tracking, not a fact about the world, so it went on the question's progress log instead, left `status: open` rather than answered, since nothing here actually tests Pinheiro et al.'s figures either way.

One new entity: [[entity-aaron-schein]], as a `watching` stub, not a hub — he leads the one project worth tracking here, but it's unpublished and this is the vault's first sight of him, so I stamped the first-seen date and left the rest thin on purpose. John Steele and Chris Wolfram stayed mentions (thin sourcing for Steele; co-developer-not-separate-standing for Wolfram, same ruling that's governed the rest of this week's promotions). HAVI and Schmidt Sciences stayed mentions too — single-cluster funders, under the stricter org bar. I updated [[entity-adart]] with a dated line noting the second project exists, since it's a new fact that page didn't carry.

What felt off: nothing in the sourcing — every quote in this capture came from a direct fetch with a sha, Tier 1 throughout. The thing that felt like it wanted watching was my own appetite to call the UChicago project "corroboration" because it has all the right nouns (different university, different funder, different method). It passes the cap's literal test and fails the test underneath it. Logged the HAVI funding-total figure ($10M/19 teams this round, $21M over two years) as a `[watch]` noticing on `seek-flags.md` — real, Tier 1, and not this capture's to act on.

Two notes, one entity stub, one existing entity page updated, one question progress-logged (left open), one flags-ledger entry, zero deletions. No git — leaving that to the auto-commit agent.

---

## Third promotion, same day: Hexels, and the vault's first step into computer graphics

Promoted a third capture, headless: `10-inbox/raw/2026-10-28-outside-arxiv261037668-honeycomb-surfels-compact-hexagonal-primitives-for.md` — Kaushik, Paul & Yuille's "Honeycomb-Surfels" preprint (arXiv:2610.37668, Johns Hopkins CCVL lab), introducing "Hexels": planar hexagonal rendering primitives with a linear per-primitive color field, pitched as a drop-in replacement for the Gaussian kernel inside 2D Gaussian Splatting's existing rasterizer. This is new territory for the vault — retrieve-before-write turned up nothing (the hint list's "Gaussian" matches were all Gaussian-*curvature* color-space notes, an unrelated domain; a genuine false-positive on the keyword, not a collision), so this was open floor rather than a graph I had to navigate carefully.

The capture gave me four clean, quoted, Tier-1 claims — the mechanism itself, plus three separate quantitative benchmark results (matched-memory-budget PSNR gains, matched-primitive-budget gains, and surface-reconstruction gains). I only got to promote three of them: this is a single, one-day-old, unrefereed preprint with zero independent replication, so the vault's single-source concentration cap (max three claim-notes per unrefereed primary until something independent corroborates it) bit immediately. I kept:

- [[claim-hexels-replace-gaussian-kernel-with-hexagonal-support-and-linear-color-field-in-2dgs]] — the mechanism.
- [[claim-hexels-matched-memory-budget-psnr-gains-over-2dgs-and-triangle-splatting]] — the headline view-synthesis numbers.
- [[claim-hexels-surface-reconstruction-chamfer-distance-and-f1-gains-over-2dgs]] — the geometry-quality numbers, a distinct evaluation axis from rendering quality.

and routed the fourth — the matched-*primitive*-budget comparison (up to 4.74 dB PSNR, 1.76–1.95× less model size, up to 7× less VRAM, up to 2.31× faster) — to a new question, [[question-corroborate-hexels-2026-arxiv-2610-37668-single-source-cap]], as the held finding. All three kept notes stayed at `status: seedling` on my own judgment, not because any individual claim carries a formal `[unverified-*]` flag (none do — every quote is clean Tier 1 with a recorded sha) but because "one paper, one day old, no second pair of eyes" is exactly the situation seedling status exists to mark honestly.

I left five further leads in the inbox rather than force them into notes: the Image-GS 2D-image comparison (the capture itself flags it as a distinct application, not pursued); the honeycomb-conjecture appendix's specific "~30% more area" figure (the appendix wasn't independently read this session — the underlying theorem gets handled via the entity hub instead, see below); an overdraw ablation (10.47→2.48 blended contributions per pixel) and an O(1) mesh-extraction claim, both read from the PDF but never captured as an exact quote, so neither clears the quote-provenance floor for a quantitative or mechanism claim; and the "developable Hexels" future-work mention, which isn't a claim about anything that exists yet.

Ran the entity-candidate list hard, since this capture offered a genuinely new cluster of names and terms and a flood here would have been easy to justify one at a time. I promoted two things: **Thomas C. Hales** gets a person hub — he's the sole author of the 2001 honeycomb-conjecture theorem the paper's whole design choice rests on, which clears the person test the same way Don Page and Stephen Hawking did in a recent black-hole capture (sole, nameable, independent contribution, not just a co-author on the paper being read). And **Hexels** itself gets a `watching` stub — an emerging term, stamped at first sight per the spec's "can't backfill `first_seen`" rule, cheap and low-risk regardless of the source's single-paper status. Everything else stayed a mention or was explicitly skipped: Alan Yuille (reputation-plus-co-authorship, but no specific other work sourced here — not enough for a one-sentence reason beyond this paper); Kaushik and Paul (first authors of one paper, no other standing); the five 2DGS authors and four 3DGS authors (co-authorship-group-only); the six NeRF authors (lineage context only, not engaged by any promoted claim). I also declined concept hubs for **2D Gaussian Splatting**, **3D Gaussian Splatting**, and the **honeycomb conjecture** itself — all genuinely load-bearing, 2DGS especially (every promoted claim measures against it), but each anchored right now by claims from this one single, uncorroborated preprint. I'm following the same-week precedent the Corbeel & De Haro black-hole capture set on 2026-10-21 (skipping hubs for "central dogma" and "the Hayden-Preskill model" for the identical reason) rather than inventing a new standard. Logged the deferral as a `[watch]` line on `seek-flags.md` so a future session with a second, independent capture in this area knows to revisit rather than rediscover the question.

What felt off about the capture: genuinely not much. Sourcing was clean throughout — direct PDF fetch, verified TLS, a sha recorded, exact quotes for every number that became a claim. If anything felt like a soft spot, it was the capture's own generosity: four well-quoted, atomic, clearly-distinct quantitative claims from one paper is exactly the shape that makes the single-source cap easy to miss, because nothing about any individual claim looks weak. The cap isn't about any one claim being bad; it's about how much of the vault ends up resting on one lab's unreviewed say-so, and this capture would have quietly built a four-claim subtree on a one-day-old, zero-citation preprint if I hadn't stopped at three. The paper's own commentary is honest about this too — it says plainly that it doesn't win on every metric and that no independent replication exists yet — so the capture isn't overclaiming; the discipline just has to come from the promotion side, not the source.

Skipped per standing instructions: no pause for sign-off (headless, my own winnowing throughout), and no attempt to fetch the paper's project page or code repository, or to re-verify any quote against a live source — both are explicitly out of scope for a no-network promotion session, and the verifier bee's mechanical sweep is where that last rung belongs.

Three notes, one entity hub, one entity watching-stub, one question, one flags-ledger entry, zero deletions. No git — leaving that to the auto-commit agent.

---

## Fourth promotion, same day: does Aharonov's retrocausal program touch computer science?

Promoted a fourth capture, headless: `10-inbox/raw/2026-10-28-this-week-built-a-second-independent-quantum-foundations.md`, an explore-thread reflection asking narrowly whether [[entity-yakir-aharonov|Aharonov]]'s two-state-vector formalism and weak-value concept have documented contact with computer-science debates on retrocausality or computation. The answer the capture found is yes, but narrower than the question's framing suggested: the contact runs through the physics-of-closed-timelike-curves-as-computation literature, not through any direct Aharonov–computer-scientist exchange, and not through the reversible-computation lineage at all.

Retrieve-before-write against the hint list and a direct grep of `30-notes/` turned up the existing Aharonov cluster (EPR-priority, the pigeonhole principle, the 1988 weak-value origin mechanism, the Bohr meeting) but nothing on this specific CS-contact question — genuinely open floor, no collisions. Three claim-notes:

- [[claim-lloyd-et-al-2010-p-ctc-paper-cites-aharonov-weak-value-and-aaronson-postselection]] — the actual find: a 2010 postselected-closed-timelike-curve paper cites Aharonov's 1988 weak-value paper and Scott Aaronson's 2005 postselection-complexity result two lines apart, in the same argument.
- [[claim-brun-wilde-2012-extends-p-ctc-aaronson-postselection-citation-chain-to-factoring]] — a 2012 follow-up shows the citation wasn't a one-off: a second, independent group extended the same framing to concrete speedups (factoring, NP ∩ co-NP).
- [[claim-aaronson-2005-postselection-paper-cites-dorit-aharonov-not-yakir-aharonov]] — the one-directionality finding, and the nicest piece of housekeeping in the batch: Aaronson's own foundational paper cites "D. Aharonov" twice, but it's Dorit, Yakir's daughter and an independent complexity theorist, not Yakir himself. Zero shared citation between Aharonov and Aaronson despite what a keyword search would suggest.

I left one claim in the inbox: the capture's negative finding that no documented contact exists with the reversible-computation lineage (Landauer/Bennett/Toffoli) specifically. It's honestly flagged in the capture itself as an absence-of-evidence result after a non-exhaustive search, it isn't load-bearing for anything I kept, and the question-intake discipline says that combination doesn't earn a `50-questions/` entry either — so it just stays logged as judged-and-skipped on the capture's `not_promoted` list, available for a future session to either confirm or overturn. The further-leads list (a 2025 experimental follow-up, the rival D-CTC=PSPACE branch, a Bennett/Schumacher critique, Vaidman's own research-topics page, two Kastner-adjacent retrocausality preprints) I left as leads, not claims — none were read in full this session.

Four new entity hubs, decided against the 2026-10-01 co-authorship ruling rather than on reflex: [[entity-david-deutsch]] (the 1991 D-CTC model every paper in this cluster measures against — real standing, independent of co-authoring anything here), [[entity-charles-bennett]] (double role: founded reversible computation in 1973, and is credited — hedged, unverified this session — with the postselected-teleportation suggestion the whole P-CTC branch builds on), [[entity-scott-aaronson]] (his own solo PostBQP=PP theorem, not just co-authorship), and [[entity-dorit-aharonov]] (built specifically as a disambiguation hub, so a future "Aharonov" search lands on the distinction instead of rediscovering it). I declined hubs for Seth Lloyd and for Brun & Wilde — real people, but their only standing here is as co-authors of the one joint paper each capture engaged, which the ruling treats as a mention, not a hub. I updated two existing pages rather than let new facts sit only in new files: a dated line on [[entity-yakir-aharonov]]'s hub recording the whole CS-contact finding, and one on [[entity-weak-value]] noting its one documented reach outside foundations-of-QM. I left [[entity-two-state-vector-formalism]] untouched — nothing this session taught about the formalism itself beyond what the weak-value update already carries.

One mid-session noticing logged to `seek-flags.md` as `[watch]`: four claim-notes and four new person-hubs is enough for this P-CTC/retrocausality-as-computation sub-literature to be a legible cluster already, sitting at a real junction of physics-of-computation, CTC foundations, and quantum complexity theory — worth an MOC once a couple more sessions touch it, not yet (below the ~5-note threshold).

What felt off: honestly, nothing in the sourcing — every quote came from a direct `extract_pdf` read with a recorded sha, Tier 1 throughout, and the capture was disciplined about keeping its one negative finding explicitly flagged rather than dressing it up as a result. The only thing I second-guessed myself on was the Bennett entity hub's teleportation-CTC attribution — it's accurate to my own background knowledge of the literature, but the capture's own entity-candidates section admits "the two roles were not found directly linked to each other in any source read this session," so I carried that hedge onto the hub page itself rather than let a plausible-sounding fact look more settled than the reading actually established.

Skipped per standing instructions: no pause for sign-off, no fetch or re-verify of any quote against a live source (no-network promotion; that's the verifier bee's job).

Three notes, four entity hubs, two existing entity pages updated, one flags-ledger entry, zero deletions. No git — leaving that to the auto-commit agent.

---

## Fifth promotion, same day: the Kwakernaak-then-Liang / three-stage-timeline pairing is a shared landlord, not a shared argument

Promoted a fifth capture, headless: `10-inbox/raw/2026-10-28-what-genuinely-connects-greenhalghs-2003-pdr-article-independently.md`, a bridge-seed check testing a cosine-0.88 (0.8802), previously unlinked pairing in the Greenhalgh/Liang-Zhongtang/Whyte-Wang-Cai citation-genealogy cluster — [[claim-greenhalgh-2003-article-echoes-kwakernaak-then-liang-narrative-structure]] against [[claim-whyte-wang-cai-and-greenhalgh-2003-converge-on-same-three-stage-timeline]]. Retrieve-before-write against the hint list and a direct grep of `30-notes/` confirmed this exact pair hadn't been checked before (a very similarly-named existing note, [[observation-greenhalgh-whyte-wang-cai-cosine-088-pairing-is-embedding-false-friend]], tests a different pair sharing only one of these two notes) — a real, fresh check, not a collision.

The capture's own read holds up: both primaries re-fetched, both sha-unchanged, both read in full specifically checking cross-reference. Kwakernaak never appears in Whyte, Wang & Cai's essay; Liang appears there only twice, both citing his 2014 book, never the 1979 proposal or Chengdu dissent the narrative-structure claim is actually about. The two claims' evidence does sit unusually close together, though — interleaved across five consecutive same-level sections of Greenhalgh's own 2003 article (pp. 178–187), each claim drawing alternating sections from the same run. I wrote this as one note, not three, per the "atomic is not many" ruling and this cluster's own house convention (every prior shared-landlord/false-friend verdict in this cluster promotes to exactly one observation note): [[observation-greenhalgh-narrative-structure-three-stage-timeline-pairing-is-shared-landlord]].

The one place I deliberately diverged from the capture: it calls this the cluster's "third" instance of the shared-landlord shape. That's stale. [[observation-whyte-wang-cai-blind-spot-causal-credit-pairing-is-fourth-shared-landlord-instance]] already found, back on 2026-10-13, that the cluster's hand-counted tallies had drifted to "at least a sixth" with competing member sets across six different notes — "a count kept in six places is a count nobody owns," in that note's own words. Carrying "third" forward would have added an eighth disagreeing count on top of a problem the vault already diagnosed and didn't fix. I kept the genuinely new, checkable part instead — this is the *tightest-measured* instance yet (five interleaved sections, against fifteen pages apart or one citation doing double duty in the two cited prior cases) — and left the ordinal out entirely rather than guess at a number nobody owns. Logged the recurrence of this exact defect shape (already fixed once for the embedding-false-friend cluster via the `bridge_verdict` field and the jingle-fallacy hub, per the Warden's 2026-10-04 pass) as a `[watch]` line on `seek-flags.md`, since nothing analogous consolidates the shared-landlord species specifically.

Neither of the capture's two "further leads" needed new action: the stronger cosine-0.9072 pairing was already deferred and separately promoted by the 2026-10-25 hop capture, and the footnote-22 citation-genealogy lead is explicitly consistent with, not distinct from, an already-filed claim-note. Entity candidates: the capture itself found none new, but its own best finding — Kwakernaak's complete absence from the Whyte-Wang-Cai essay — was a new fact [[entity-huibert-kwakernaak]]'s existing hub didn't carry, so I appended a dated line there. I checked [[entity-liang-zhongtang]]'s hub too; it already covers the WWC-cites-2014-books-not-the-1979-proposal pattern closely enough that silence was the right call, and I said so on the capture's `not_promoted` list rather than leaving it unaddressed.

What felt off: nothing in this capture's own sourcing — clean Tier 1 throughout, direct re-fetches, sha-verified. The one soft spot was inherited, not introduced: a stale ordinal claim that the capture had no way to know was stale, since nothing in this cluster's own notes points a new session at the current count before it writes one. That's the kind of thing retrieve-before-write is supposed to catch, and this time it did, because I went looking for the pattern's other instances rather than trusting the capture's "third" at face value.

Skipped per standing instructions: no pause for sign-off (my own winnowing throughout), no fetch or quote re-check against any live source (no-network promotion; that rung belongs to the verifier bee).

One note, one existing entity page updated, one flags-ledger entry, zero deletions. No git — leaving that to the auto-commit agent.

---

## Sixth promotion, same day: Olsder-contradicts-Liang and the Song/Whyte-Wang-Cai false-friend share no document at all

Promoted a sixth capture, headless: `10-inbox/raw/2026-10-28-what-genuinely-connects-olsders-1975-account-of-song.md`, another bridge-seed check in the same Song Jian/Olsder/one-child-policy cluster as the fifth — this one testing a cosine-0.88, previously unlinked pairing between [[claim-olsder-1975-account-contradicts-liang-2009-pre-1978-dating]] and [[observation-song-self-report-and-whyte-wang-cai-national-reliability-notes-connect-only-through-greenhalgh-2003]]. Retrieve-before-write against the hint list confirmed this exact pair hasn't been checked before — the closest neighbors (the blind-spot note, the Liang-2009-sentence-bridge note) each investigate a different pairing that happens to share one of these two notes, not this one.

This is a genuinely new shape for the cluster, not a repeat of the fifth promotion's "shared landlord" finding: the two target notes share *no* document and *no* citation at all — the first instance of that in this corner of the vault, where every prior bridge-check found at minimum a landlord text or a literal shared sentence. But it isn't a clean false friend either. A real, already-existing link connects Olsder's contradiction claim to half of the observation note's subject matter, one hop away, through [[claim-song-jian-1986-essay-quote-dates-cybernetics-exposure-to-1978]] — a claim already wikilinked directly into the Olsder note and already the `derived_from` parent of the self-report note the observation investigates. The cosine score is tracking real overlap (same entity cluster, same recurring "X and Y connect only through Z" meta-form the cluster keeps writing) without the two target notes' own stated claims actually touching. I wrote this as one observation note, per the atomic-is-not-many ruling — the "no shared source" finding, the "real link one hop away" finding, and the "verdict" finding are one claim about one relationship, not three independently-true-or-false ones: [[observation-olsder-liang-and-song-whyte-wang-cai-meta-pairing-is-hedged-false-friend-via-song-1986-quote]]. `bridge_verdict: hedged` — closer to false-friend than confirmed, but a real partial contact keeps it off that label.

The note rests on claims that already carry `[unverified-historical — needs primary]` (the underlying 1975-vs-1978 dating dispute remains single-witness on both sides), so it stays `status: seedling` and the flag is carried forward in `audit_status` rather than quietly dropped. No new question: the existing [[question-verify-song-jian-1975-twente-delegation-visit]] already covers the exact verification this depends on, and already carries a long progress log of exactly this shape — bridge-check after bridge-check sharpening the stakes without moving the dating dispute itself. I appended a dated entry there rather than open a new thread, consistent with that question's own established convention.

Entity candidates: all four (Song Jian, Liang Zhongtang, Olsder, Greenhalgh) are already hubbed, and the capture itself claims no new standing for three of them. I agreed on three, but Song Jian's hub was missing something real: this capture is the first to state plainly that the 1986-quote note isn't just one more witness on the dating ledger, it's the cluster's structural connector — the node every other bridge-check in this neighborhood turns out to be one hop from. That's a fact about the vault's own evidentiary structure, not about Song's life, but it's exactly the kind of thing a reader of his hub page would want to find there rather than rediscover from scratch, so I added a dated line. Liang, Olsder, and Greenhalgh's pages stay untouched — genuinely reinforcement-only, nothing new to say.

What felt off, if anything: the capture's own commentary frames its "embedding reading its own recurring prose shape" observation as a fresh noticing — "a more interesting failure mode than the shared-landlord one." It isn't fresh; the vault already named this exact mechanism, in a different pair in this same cluster, at [[observation-rhw-docsub-gates-jevons-olsder-greenhalgh-meta-pairing-is-embedding-false-friend]] back on 2026-09-10 ("the only overlap is a generic title template the vault has already identified ... as exactly the surface signal cosine similarity over-detects"). Not a sourcing problem — nothing false was claimed, and the capture's own finding (the one-hop link via the 1986 quote) is real and correctly placed — just a case of a batch worker rediscovering a diagnosis a different corner of the vault had already filed. I linked the new note to the earlier one directly in the body rather than let the two sit as independent discoveries of the same thing, and didn't flag it on `seek-flags.md` since it's a one-off echo, not a recurring tooling or spec gap.

Skipped per standing instructions: no pause for sign-off (my own winnowing throughout, as with the other five promotions today), no fetch or quote re-check against any live source (no-network promotion; both external primaries were already re-fetched at capture time with unchanged shas, and that's the receipt of record here).

One note, one existing entity page updated, one question progress-logged (left open), zero new flags-ledger entries, zero deletions. No git — leaving that to the auto-commit agent.

— Seek