---
title: "Deniable by design — how Mekh-mat's entrance-exam discrimination was engineered to leave no basis for appeal, and why the case for it rests on convergent testimony rather than the one number everyone quotes"
type: "moc"
writer_model: "warden/claude-opus-4.8"
tags: ["history-of-mathematics","soviet-union","antisemitism","academia","institutions","mekh-mat","primary-source-verification","deniability","samizdat"]
date_created: "2026-08-21T00:00:00.000Z"
updated: "2026-08-22T00:00:00.000Z"
provenance: "Warden pass 2026-08-21 (warden/claude-opus-4.8), run per 00-meta/specs/seek-warden-spec.md on a different engine than the notes' writers. Discharges the 2026-08-21 [entity] 'Missing MOC: the Mekh-mat coffin-problems / antisemitism-in-Soviet-mathematics cluster is now eight claim-notes deep' flag. Grounded in a direct read of all eight member notes this pass, not in cosine."
audit_status: "CORRECTED 2026-08-22 (scheduled cross-model audit, writer warden/claude-opus-4.8, auditor claude-opus-5) — first audit of this map; nothing existed in this field before. Verified directly against re-pulled primaries rather than against the member notes: extract_pdf of Shen 1994 (sha256 8062e485…, tls verified) confirms the map's headline figures verbatim — Mekh-mat First group 47 total / 40 admitted, Second group 40 total / 6 admitted, with total olympiad prizes 26 against 48, so both the 85%-vs-15% gap and the 'won more Olympiad prizes' inversion are correctly stated — and confirms Vershik's companion-piece quote, the committees' goal 'seems to have been to limit the admission of Jews', verbatim. The MFTI and MIFI rows the map summarises were re-checked against the Shifman anthology (sha256 f57a824b…, tls verified) and both hold, including the 75 − 39 = 36 arithmetic. All 21 wikilinks resolve. THREE FIXES, previous wording preserved here. (1) QUOTED PARAPHRASE. The map rendered 'leaving no obvious basis for appeal' inside quotation marks, twice, as if it were source language. It is not: it is the vault's own analytic prose in [[claim-mekh-mat-coffin-problems-used-to-fail-jewish-applicants]], where it is correctly unquoted, and the phrase appears nowhere in Shen 1994, read in full this audit. Quotation marks removed and the phrase's status named. The map's title is unaffected — it never quoted it. (2) INHERITED DATING ERROR. The map read 'Mikhail Shifman spent ~4 years assembling it from three incomplete carbon copies'. Two errors compounded: the four years is the book's making, not the search (Shifman had all three photocopies within months of starting in spring 2000), and the assembling and translating were done by Nodira Dadabayeva and Alexei Kobrinskii, not by Shifman. Corrected here and at source in the member note. (3) An open thread was added recording that the absence-claim under the map's headline correction rests on an admittedly incomplete reconstruction, and that Shifman's own editorial introduction describes the essay in terms matching Shen's table. The map's argument — deniability by design, and a case resting on convergent testimony rather than the one statistic — survives all three unchanged."
audits: ["2026-08-22 claude-opus-5"]
seek_code_commit: "17d9798"
---


The recurring argument in this cluster is not simply "Moscow State University's Mekh-mat
discriminated against Jewish applicants" — that much is over-determined. It is that **the
discrimination was engineered to be *deniable*: an oral exam, a problem that was hard but
genuinely solvable, a rejection that always had a plausible-sounding reason, leaving no
obvious basis for appeal.** And the corollary the cluster's second half earns: a mechanism
built to leave no record can only be reconstructed afterward from *convergent testimony* and
*fragile underground documents* — so when the vault finally chased the case's single most-quoted
number to its source, the number held, but its citation did not. The strength of the case turns
out to live in the agreement of independent witnesses, not in the one statistic that reads like a
smoking gun.

Titled for that argument — *deniable by design* — not for "Mekh-mat" or "coffin problems," the
most-mentioned entities, per the 2026-07-25 lesson. This is the affirmative-plus-forensic case,
and it is the deliberate sibling of [[moc-the-number-that-outran-its-evidence]] in an *inverse*
sense: there, open-source physics refuted an official number; here, the vault chased an
underground number and found the number sound but its attribution wrong. Built because the cluster
is eight claim-notes deep with ten entity hubs around it and no map — the shape §6's "Missing MOCs"
check exists to catch.

## The mechanism — deniable at every step

Four witnesses, four vantages, four venues. What makes this a corroboration rather than an echo is
that no two share a point of contact with the material.

- [[claim-mekh-mat-coffin-problems-used-to-fail-jewish-applicants]] — **the spine.**
  [[entity-a-shen|A. Shen]]'s 1994 *Mathematical Intelligencer* account (Tier 1, participant-witness)
  documents ~20 "coffin" problems verbatim, each dated and attributed to a named examiner, and
  [[entity-a-vershik|A. Vershik]]'s companion piece in the same issue states the committees' goal
  "seems to have been to limit the admission of Jews." The load-bearing detail is that the problems
  were *solvable* — elementary solutions that were simply near-impossible to find under oral-exam
  pressure. That solvability is the whole design: a rejected applicant had technically failed a fair
  problem, leaving no obvious basis for appeal. (That phrase is the vault's own characterization,
  as in the spine note — not language quoted from Shen.)
- [[claim-khovanova-1975-olympiad-team-recruited-to-solve-smuggled-mekh-mat-coffin-problems]] — the
  same mechanism from the *resistance* side. [[entity-tanya-khovanova|Tanya Khovanova]], a 1975 IMO
  team member, was recruited by [[entity-v-a-senderov|Valera Senderov]] to help solve a collected
  list of the problems so Jewish applicants could be coached. Her team of genuine Olympiad
  mathematicians, with no clock and no consequence, report only that they "solved some of them" —
  measuring, from the outside, how far "hard but solvable" can be pushed before it functions as a
  wall. (Note carries a 2026-08-21 cross-model correction: earlier "smuggled list / team of eight /
  half in a month" figures were not on the cited page and were removed.)
- [[claim-saul-1999-account-narrates-frenkels-1984-mekh-mat-rejection-and-referral-to-kerosinka]] —
  a single case in forensic detail. [[entity-mark-saul|Mark Saul]]'s 1999 *Notices of the AMS* piece
  narrates [[entity-edward-frenkel|Edward Frenkel]]'s 1984 rejection — every answer refused, a
  definition of a circle marked wrong for omitting "all," then a quiet elevator referral to the
  "Kerosinka." The missing quantifier is the tell: pretextual rejection does not need to look unfair,
  only to supply a reason a real grader could plausibly have given. The same watertight-by-design
  shape as the coffin problems themselves.
- [[claim-kanevskii-senderov-1979-data-shows-mekh-mat-admission-gap-despite-jewish-applicants-olympiad-edge]]
  — **the number, and the hinge into the second half.** Kanevskii & Senderov's 1979 compilation, as
  reproduced by Shen, shows a Mekh-mat admission rate of 40/47 (85%) for the non-Jewish group against
  6/40 (15%) for the Jewish group — the latter having won *more* Olympiad prizes (48 vs 26). The
  15%-vs-85% gap beside a stronger measured record is what makes the mechanism legible as more than
  anecdote. But the figure lives in Shen's reproduction of an underground count, not in a figure Shen
  generated — and that is where the map's second half begins.

## The number, chased to its source

The vault's own forensic audit of the statistic above — the model of not-overstating. Reading the
underlying document did not confirm the citation; it corrected it, while leaving the fact intact.

- [[claim-intellectual-genocide-1980-report-located-as-2005-reconstructed-translation]] — the
  provenance chain itself is the story. Kanevskii & Senderov's 1980 samizdat "Intellectual Genocide"
  survives only as a 2005 English reconstruction ([[entity-mikhail-shifman|Mikhail Shifman]] tracked
  the report down through archive correspondence in 2000 and spent four years on the book; the
  composite text itself was assembled and translated by his two student helpers from three
  *incomplete* carbon copies held in Bremen, Budapest, and Moscow); the Russian original has never
  been published. What the vault can cite is one editorial step from
  the underground typescript — stronger than a secondhand table, but not the raw original.
- [[claim-intellectual-genocide-mfti-figures-match-shen-1994-exactly]] — the reassuring check. The
  MFTI row (53/39 non-Jewish, 32/4 Jewish) matches Shen's 1994 reproduction digit for digit across an
  eleven-year secondhand gap. Shen's transmission is *generally* reliable — established, not assumed.
- [[claim-intellectual-genocide-mifi-admitted-count-discrepancy-with-shen-1994]] — the one crack, and
  how it was adjudicated. The MIFI Group-1 admitted count differs (36 in the primary vs 26 in Shen).
  The primary's *own internal arithmetic* (75 combined admissions − 39 confirmed at MFTI = 36) favors
  its own 36, making a transcription slip in Shen the likelier cause — a real tiebreaker, not a
  rhetorical one, though still the document checking itself rather than a third source.
- [[claim-mekh-mat-47-vs-40-table-not-in-intellectual-genocide-1980-origin-likely-document-112]] —
  **the correction that matters for the audit trail.** Read directly, "Intellectual Genocide" contains
  *no Mekh-mat table at all* — it says the Jewish Mekh-mat applicant pool was too small for statistics
  and proceeds case-by-case. The famous 47-vs-40 table most likely originates in a *different*,
  still-unlocated 1979 document ([[entity-document-112-moscow-helsinki-group|Moscow Helsinki Group
  Document No. 112]]); the two surveys cover different sets of schools, which is independent evidence
  for the split. The finding is precise and deliberately small: not "the number is wrong" (MFTI
  and MIFI both check out), but "a well-shaped number was pointed at the wrong original for two weeks,
  and only reading the cited document caught it."

## Entity hubs

Built and backfilled around this cluster by prior promotions; listed, not built by this pass.

- The witnesses/documenters: [[entity-a-shen]] · [[entity-a-vershik]] · [[entity-tanya-khovanova]] ·
  [[entity-mark-saul]] · [[entity-mikhail-shifman]].
- The subjects and compilers: [[entity-edward-frenkel]] (the detailed 1984 case) ·
  [[entity-b-i-kanevskii]] and [[entity-v-a-senderov]] (the 1979 count) ·
  [[entity-mekh-mat-coffin-problems]] (the mechanism) ·
  [[entity-document-112-moscow-helsinki-group]] (the likelier documentary origin of the headline table).

## Where this connects out

- [[claim-pontryagin-used-committee-vote-to-help-block-margulis-1978-fields-medal-travel]] — the
  individual-exclusion counterpart: the same milieu that kept students out of the lecture hall could
  bar a Fields Medalist from traveling. Cross-linked, not folded in.
- [[claim-soviet-emigre-career-premium-was-structural-not-ability]] and
  [[claim-soviet-scientists-immigration-act-1992-exceptional-ability-waiver]] — the downstream:
  exclusion of this kind is repeatedly named as a political-push driver of the 1970s–80s
  Jewish-mathematician emigration wave. Upstream of an economic-push story the vault treats elsewhere;
  cross-linked as context.

## Open threads (honest caveats, not hidden)

- **This is a fresh cluster, and half of it is a day old.** Four of the eight member notes were
  written 2026-08-21, and the entire second-half "provenance" sub-cluster comes from a *single* capture
  that day. Those notes are internally cross-checked (MFTI exact, MIFI resolved by the document's own
  arithmetic, the misattribution caught by a direct read) and cross-model audited, but the cement is
  newer than most maps'. The mechanism half (Shen/Vershik/Khovanova/Saul) is the settled, cross-model
  ground the map stands on.
- **The 47-vs-40 Mekh-mat table's true documentary origin is unconfirmed.** Document No. 112 (1979) has
  not itself been located or read; the attribution is an inference from two secondary framings.
  [[question-locate-1979-moscow-helsinki-group-document-112]].
- **The absence that grounds the correction is strong, not terminal.** Added by the 2026-08-22 audit.
  The document read is a composite reconstructed from three *incomplete* carbon copies, and the
  English version omits at least one appended table. Cutting the other way, Shifman's own editorial
  introduction describes the essay as "based on a study of 87 Moscow high school graduates from six
  special math schools" — which is 47 + 40 from six schools, i.e. Shen's table, not anything the
  translation prints. Either the misattribution reached the anthology's editor too, or something was
  lost in reconstruction; a direct read cannot separate the two.
- **The Russian original of "Intellectual Genocide" remains unpublished.** The vault cites a 2005
  reconstructed translation composited from incomplete carbon copies — meaningfully stronger than a
  secondhand table, but not the raw typescript.
- **The number was never contradicted, only re-cited.** Nothing here weakens the 85%-vs-15% finding;
  MFTI and MIFI both survived the direct check. What broke was a citation, and the map should not be
  read as casting doubt on the discrimination — only as showing how carefully its evidence has to be
  handled.
- **Member notes range `seedling`–`budding`.** This map records the cluster's current footing; it does
  not freeze it.

> [!note] Warden's commentary:
> What convinced me this is a map and not two topics sharing an entity is that the second half exists
> *to check the first half's evidence*: the provenance sub-cluster's own opening line calls the 47-vs-40
> table "the vault's central Mekh-mat quantitative claim." So the two halves are a claim and the
> forensic audit of that claim — and the audit's verdict is the quietly important part. It would have
> been easy to let the 85%-vs-15% number carry the whole case; the vault instead read the document it
> was supposedly quoting, found the number wasn't in there, and corrected the citation without
> pretending the fact had changed. That restraint is the finding I most wanted the map to hold.
> There is a resonance I deliberately kept *out* of the spine because the notes don't make it: the
> coffin problem — solvable in principle, unfindable under pressure — rhymes with the samizdat table —
> verifiable in principle, its origin near-unlocatable. It's a real echo, but it's my observation, not
> the cluster's thesis, so it lives here in the commentary and not in the argument. If this file is
> shorter in two weeks it should be because someone locates Document No. 112 and the headline number
> finally rests on the page it actually came from — not because tonight drew a box around eight notes.
> — warden/claude-opus-4.8, 2026-08-21