---
title: "MOC: The same source wearing different clothes — what independent corroboration is not, and why a single-source cap's absence-finding is a date, not a verdict"
type: "moc"
tags: ["moc","source-criticism","single-source-cap","corroboration","independence-test","peer-review-status","preprints","epistemic-hygiene","vault-sourcing-floor"]
updated: "2026-10-30T00:00:00.000Z"
writer_model: "warden/claude-opus-4-8"
audit_status: "built 2026-10-30 (Warden pass, warden/claude-opus-4-8). FIRST-PASS SINGLE-ENGINE BUILD; no cross-model audit of this MAP yet — a future target, the posture the one-witness and shape-recurs maps carried before their next-day audits. READ IN FULL this pass (gathered by retrieval and reading, not by tag): the nine absence-findings (GIFT, Bit2Watt, RIKER, FIC, Hu-2026/NeuNeu, Pinheiro-Ting/Babylonian, Hexels, the NYC-nudge RCT, Allzén), the two peer-review-status discharges read directly (Steck-2024 precedent, TCHES/Bit2Watt, NeurIPS/Beltran), and the independent-corroboration discharge (Daryani/ChronoCulture). The false-corroborator sub-notes (RIKER self-replication, Hu same-byline, the UChicago parallel project) were read as part of their parent absence-findings. Every wikilink target was checked to resolve on disk this pass. SOFT SPOT, named not hidden: the cluster is ~60 notes deep by the `single-source-cap`/`source-corroboration` tags; this map gathers the spine I read in full and names the wider population rather than folding in unread notes — treat the membership as the read spine, not an exhaustive census."
provenance: "Built by the Warden pass 2026-10-30 (warden/claude-opus-4-8, run per 00-meta/specs/seek-warden-spec.md on a maintenance engine kept separate from the notes' writers), discharging the 2026-10-30 [entity] flag (seek-flags.md): the 'no independent corroboration, single-source cap remains undischarged' template, flagged at six confirmed members and proposed by the promoting writer for 'its own MOC on the model of moc-the-shape-recurs-the-object-does-not or moc-one-witness-is-not-enough — not built this promotion, per bees flag, the queen decides.' MOC creation is Warden territory (the 2026-09-19 precedent carried in moc-the-shape-recurs). Named for the ARGUMENT (what corroboration is not; the dated absence as a non-verdict), not for the entity 'single-source cap' (the 2026-07-25 lesson: the recurring entity is not necessarily the recurring argument). The flag's two further sub-asks — hubbing the single-source cap itself and `bridge_verdict: hedged` as concept pages — are DECLINED, reasoned on the flag; the governance question behind them is routed to Cali in this pass's report. The cap's own text (00-meta/specs/sources.md, Cali 2026-07-29) supplies this map's title phrase."
seek_code_commit: "d40ee39"
---


One recurring discipline, assembled here for the first time: **when a load-bearing
claim rests on a single uncorroborated primary, the vault runs a dated search for
*independent* corroboration — a different author group *and* a different venue — and
records the outcome as a note stamped "as of <date>."** The recurring finding of
those searches is not that the claim is false. It is that the things which fill the
vacuum before genuine engagement arrives — press write-ups of the same paper,
automated AI-summary mirrors, the same authors' own follow-ups, a declared commercial
partner's rerun, a parallel project that never touches the figures — are "the same
source wearing different clothes" ([[00-meta/specs/sources.md|sources.md]], Cali's own
phrase). The organizing question is not *is this paper any good?* but *has a second,
independent mind actually checked this specific claim yet?* — and the honest answer,
early, is almost always "not yet," which is a date, not a disconfirmation.

**Scope, and what this is not.** This maps one **argument** — the operational
corroboration discipline the vault runs on fresh single-source claims — not a topic or
an entity. It is deliberately distinct from two neighbours it shares a destination with
but not a method:

- [[moc-one-witness-is-not-enough]] is the **legal/historiographic** corroboration
  floor: *unus testis nullus testis*, the half-proof, patent-law's bar on
  uncorroborated prior invention, and priority claims that dissolve to one witness
  (Werbos, Isaacs). That map counts witnesses to a **retrospective** fact — who did
  what first, authenticated against the record. This map is **prospective** and
  operational: it watches a freshly-posted preprint and asks whether the field has
  caught up to it yet, under a cap the vault wrote for itself. The attribution case
  [[claim-schmidhuber-dreyfus-1973-attribution-single-source-not-corroborated]] ("two
  venues reciting the same sentence is not two witnesses") belongs to *that* floor, not
  this one, and is kept a neighbour below.
- [[moc-an-estimate-wearing-the-clothes-of-a-count]] shares this map's "X wearing the
  clothes of Y" shape but is about a different confusion (a modelled estimate read as a
  measured count). Named here only because the titles rhyme; not a member.

## The dated absence-finding — the honest "not yet"
Each is a search that returned no independent corroboration, recorded with its date and
its reason, and each says on its own face that the absence is consistent with recency,
not a verdict. Oldest first, so the pattern's age is visible.

- [[claim-gift-2026-unrefereed-and-unreplicated-as-of-2026-08-07]] — the earliest
  instance and the one the later notes cite as the template. A month out, GIFT's 7.6%
  speedup had no journal-ref, no citation, no fork: "a month bought nothing but the
  honest shape of nothing." The `[unverified-quant]` hold stands — "confirmation the gap
  is still open, not that it has closed in either direction."
- [[claim-bit2watt-no-independent-corroboration-found-as-of-2026-09-10]] — the note that
  names the mechanism. Six press outlets (The Hacker News, The Register, GBHackers, …)
  all describing the *same* paper: "a press summary of the same preprint is 'the same
  source wearing different clothes,' not corroboration." (Later discharged by a different
  route — see below.)
- [[claim-riker-2026-no-independent-source-same-author-admits-no-replication]] — the
  cleanest statement that the look-alikes are not independence: the only follow-ups are a
  same-author/same-company self-replication and a *declared commercial partner's* rerun
  (Signal65 "in partnership with" Kamiwaza, RIKER's own creator). The decisive witness is
  the author himself: "our findings have not been independently replicated with
  alternative frameworks."
- [[claim-fic-preprint-no-independent-corroboration-as-of-2026-10-20]] — the taxonomy's
  purest two look-alikes: a same-day blog post reproducing the abstract verbatim, and an
  automated AI-summary mirror ("5 views," empty discussion, AI overview citing only the
  preprint's own references). "What fills the vacuum before real engagement arrives …
  could be mistaken for 'coverage' by anyone not checking whether a second mind actually
  looked at the argument."
- [[claim-hu-2026-no-independent-corroboration-found-as-of-2026-10-24]] — the instance
  that shows the discipline self-correcting. The dated negative was real; a scheduled
  cross-model audit then *found* a pre-existing independent citer (arXiv:2602.10300) the
  first search missed, so "no corroboration found as of 2026-10-24" stands as an honest
  record of a search while "no independent citer existed" no longer does. The failure
  mode and its repair, both on the note's face.
- [[claim-no-independent-corroboration-pinheiro-ting-2026-babylonian-preprint-as-of-2026-10-28]]
  — the "parallel project" look-alike. A UChicago team
  ([[claim-uchicago-schein-wolfram-havi-project-parallel-not-corroborating-pinheiro-ting-babylonian-diaries|Schein/Wolfram]])
  works the same corpus but predates the preprint and never engages its figures: "two
  people building the same instrument in adjacent rooms, neither having heard the other's
  hammer yet."
- [[claim-hexels-no-independent-corroboration-found-as-of-2026-10-29]] — the youngest
  (two days old), and the sharpest statement of the recency frame: not "nobody looked and
  found nothing," but "nobody has looked yet, including the one place cheapest to check" —
  the authors' own project page, filled with their own numbers and an unrelated iPad app
  rather than a reproducible harness. "How much of 'independent corroboration' is actually
  'time.'"
- [[claim-no-independent-source-corroborates-nyc-nudge-rct-zero-rejection-result]] — the
  same-team-documents case, and a useful wrinkle: the striking 98%-acceptance figure
  appears only across the *same authors'* arXiv preprint, their ACM EC 2026 camera-ready,
  a PiTech spotlight, and a university press release. Here even a refereed version does not
  supply independent corroboration of the *result* — refereed is not reproduced.
- [[claim-allzen-2610-14137-zero-citations-eight-days-post-posting]] — the bibliometric
  version, read off INSPIRE-HEP's own citation counts: zero independent citations eight
  days post-posting, "consistent with this subfield's slow baseline citation uptake."
  (The note also carries its own honesty scar — broken capture receipts, figures
  re-confirmed live by a later audit.)

## What the look-alikes are — the taxonomy the searches keep rebuilding
The cap's definition names three (same-author follow-up, press summary, the preprint's
own revision). The searches above found the live catalogue is wider:

- **Press / journalism of the same paper** — Bit2Watt's six outlets.
- **Automated AI-summary mirrors** — alphaXiv's machine overview (FIC); the summary sites
  Hu-2026 left out rather than "dress up a search-engine echo as a second opinion."
- **Same-author / same-company follow-up** — RIKER's 172-billion-token self-replication;
  [[claim-riker-2026-self-replication-contradicts-original-fabrication-gap]];
  [[claim-hu-2026-neurips-icml-2026-same-byline-not-independent]].
- **Declared commercial partner** — Signal65's RIKER run, "closer to sponsored content
  than arm's-length testing."
- **Parallel project that never engages the figures** — the UChicago Babylonian team;
  Metis for GIFT ([[claim-metis-2025-anisotropy-diagnosis-predates-gift-not-a-replication]]).
- **The paper's own project page / GitHub** — Hexels' honeycomb-surfels.github.io;
  Pinheiro-Ting's self-hosted repo.

The through-line, stated by the notes themselves: none of these is *a different author
group testing the specific claim from outside.* That, and only that, is the bar.

## How a cap actually discharges — two distinct routes, both modest
The discipline is not a trap that never opens. Two routes clear a cap, and the notes are
careful that each clears less than it looks like it does.

**Route 1 — the source leaves the "unrefereed primary" category (peer-review status).**
The cap governs "a preprint, a working paper, a self-published report, a conference
abstract"; a document that cleared a named committee's review is not in that category.
The test is to read the venue's own review-policy text.

- [[claim-steck-2024-www-short-paper-cleared-committee-peer-review]] — the precedent: WWW
  2024's single-blind Short Papers committee review takes a paper out of the category.
- [[claim-tches-peer-review-process-clears-bit2watt-of-unrefereed-cap]] — the reusable
  test applied: TCHES's double-blind, three-reviewer, multi-round process (with
  [[claim-bit2watt-ches-2026-acceptance-confirmed-at-iacr-primary-record|acceptance
  confirmed at IACR's primary record]]) clears Bit2Watt. Seek's caveat is the load-bearing
  limit: "refereed and heavily-scrutinized are different properties, and this note only
  settles the first."
- [[claim-neurips-2025-peer-review-process-clears-beltran-nfl-ba-of-unrefereed-cap]] — the
  third application, and the sharpest caveat of all: clearing the *source's category* is
  not clearing a *claim*. The reviewed-and-accepted Beltran text still carried a fabricated
  citation through v1–v3 ([[claim-beltran-nfl-ba-accepted-neurips-2025-main-conference-confirmed-at-primary-record]]).
  Acceptance moved the source out of the cap; it vouched for nothing about the fabrication.

**Route 2 — a genuinely independent source corroborates the claim.**

- [[observation-daryani-independently-corroborates-chronoculture-discharging-single-source-cap]]
  — the model case: Daryani et al. clears every box the cap wrote in advance — different
  authors, different institution, different instrument, an *earlier* submission date. And
  the honest limit, again on the note's own face: "the cap is discharged exactly as
  written, and the branch is still two unreviewed papers agreeing with each other … That's
  genuinely more than one unreviewed paper, which is the whole point of the rule. It is not
  yet the field agreeing."

## Kept as neighbours, not members
Membership-tested and named rather than folded in.

- [[moc-one-witness-is-not-enough]] — the retrospective legal/priority floor; shares the
  destination ("one source is not enough") but counts witnesses to a settled fact rather
  than watching a live preprint's reception. Its commentary already notes that *this*
  vault-sourcing floor "is the same instinct" and deliberately declined to pull these
  operational cases in — this map is the home it pointed at without naming.
- [[claim-schmidhuber-dreyfus-1973-attribution-single-source-not-corroborated]] — an
  *attribution*-provenance single-source case; belongs to the one-witness floor.
- [[claim-lukyanov-115-figure-no-independent-source-located]] — a cousin: the independence
  test run on *secondary* popular-science sources copying one 2000 account, documenting
  textual lineage rather than a capped preprint. Named as a candidate the next pass could
  fold in or keep separate; not read in full this build.
- [[claim-riker-babylonian-absence-findings-not-evidentially-equivalent]] — the
  meta-note comparing two of this map's own absences and refusing to treat them as equal;
  a reflection on the discipline rather than an instance of it.

> [!note] Warden's commentary:
> I built this because the notes' own writer flagged it in almost these words — "may now
> be large enough to warrant its own MOC … the queen decides" — and because the read
> showed a real argument, not a tag pile. The giveaway that it is one argument and not
> sixty cosine-neighbours is that the notes keep **re-deriving the same taxonomy
> independently**: press, mirror, same byline, commercial partner, adjacent project,
> self-hosted page — each search reaches for the same list of things that look like a
> second witness and are not. The second giveaway is that the discipline has a working
> *release valve* with two keys (venue review; genuine independent corroboration), and the
> notes are scrupulous that each key turns less than it seems — refereed is not reproduced
> (Beltran's accepted-but-fabricated citation is the cleanest proof), and two preprints
> agreeing is more-than-one, not the field (Daryani). I named it for that discipline and
> its honest limit rather than for "the single-source cap," because the cap is the *rule*
> and the recurring argument is what the vault keeps *learning by running* it: that the
> absence of corroboration, early, is a date and a to-do, not a verdict — and that saying
> so in a dated note beats letting silence be read as either confirmation or doubt. The
> flag also asked whether the cap *itself*, and `bridge_verdict: hedged`, should become
> concept hubs. I declined both: they are vault-governance mechanisms, not domain concepts
> of the kind the entity-page spec's own examples (a heuristic, backprop) target, and a
> MOC of the claim-notes maps the argument without minting a hub for the rule. Whether a
> vault-internal procedural concept ever clears the entity-page bar is a standing question
> I have put to Cali rather than answered by fiat. — warden/claude-opus-4-8
>
> [!warning] Provenance and honesty caveats (read before trusting the map):
> - **First-pass, single-engine build.** No cross-model audit of this MAP yet. Its
>   strength is that the spine was gathered by direct full reads this pass, not by tag or
>   cosine, and the member notes are unusually well-receipted (sha-stamped captures,
>   quote-sweeps, cross-model audits on several).
> - **The membership is the read spine, not a census.** ~60 notes carry the
>   `single-source-cap`/`source-corroboration`/`sources-floor` tags; many are discharges,
>   contradictions, or neighbours of other arguments. This map holds the absence-findings
>   and discharge-exemplars I read in full, plus a named false-corroborator taxonomy drawn
>   from them. Other tagged notes (e.g. the CODA, D51/INSA, Costello, Bi-FORK corroboration
>   threads) may be members or neighbours; they are left for a future pass to test rather
>   than folded in unread.
> - **The finding is a discipline, not a verdict on any paper.** Every absence-finding here
>   says, in its own words, that the absence is consistent with recency. The map asserts the
>   recurrence of the *method and its honest frame*, not that any capped paper is wrong.
> - **An active, moving set.** Most members are 2026 preprints days-to-months old; several
>   absences will discharge (Bit2Watt already did). Revisit as they do — the discharges are
>   as much the finding as the absences.