---
title: "MOC: One witness is not enough — the corroboration floor, from Roman law to priority disputes"
type: "moc"
tags: ["moc","law-of-evidence","legal-history","single-witness","corroboration","testimony","priority-disputes","historiography","source-criticism"]
updated: "2026-10-18T00:00:00.000Z"
update_history: ["2026-10-17 — first build (Warden pass, warden/claude-opus-4.8)","2026-10-18 — scheduled cross-model audit (claude-opus-5): three omitted member/neighbour notes added, nothing rewritten; see the dated audit block at the end"]
audits: ["2026-10-18 claude-opus-5"]
writer_model: "claude-opus-4-8"
provenance: "Built by the Warden pass 2026-10-17 (warden/claude-opus-4.8, seek-warden-spec.md), discharging the 2026-10-17 [entity] flag (seek-flags.md, the lone-witness / corroboration-insufficiency cluster crossing the ~5-note threshold and wanting its own map, distinct from moc-testimony-is-not-a-record and the diplomatics cluster). Gathered by direct read of the member notes this pass, not by tag."
seek_code_commit: "d40ee39"
---


One recurring argument, assembled here for the first time: **a single uncorroborated
witness cannot establish a fact, no matter how plausible the testimony — and two weak
proofs do not sum into a strong one.** The rule is old (Constantine fixes it in AD 334),
it is written into living law (US patent procedure, Dutch criminal procedure), and the
vault keeps rediscovering it from the outside whenever a priority claim turns out to rest
on one mouth. The organizing question is not *is this witness credible?* but *how many
independent witnesses are there?* — sufficiency decided by number, not by the quality of
the one.

**Scope, and what this is not.** This is the map of one **argument** — the
corroboration floor — not of a topic or an entity. It is deliberately distinct from two
neighbours it keeps getting confused with:

- [[moc-testimony-is-not-a-record]] is the deep *single-instance* map of **Werbos**,
  whose argument is documentary source criticism / diplomatics — authenticating an
  origin-claim *against the dated record*. That is a different move: it measures
  testimony against a document. The floor below is prior and cruder — it counts
  witnesses and refuses to convict on one, whether or not a record exists to check them
  against. The observation note that named this seam
  ([[observation-1981-testimony-convergence-and-werbos-gap-false-friend-share-lone-witness-floor]])
  is explicit that the two traditions — evidence-counting law and document
  authentication — *arrive separately* at "one witness is not enough," and that the
  arriving-separately is itself a small instance of independent convergence.
- The diplomatics cluster on the [[claim-mabillon-1681-founded-diplomatics-to-refute-forgery-charge|Mabillon]]
  hub authenticates documents by comparison. It shares a destination with this map, not
  a method. Kept a neighbour, not a member.

## The doctrine, named
The floor stated categorically, as positive law.

- [[claim-unus-testis-nullus-testis-roman-canon-law-lone-witness-rule]] — *unus testis,
  nullus testis* ("one witness, no witness"): the uncorroborated testimony of one witness
  is discounted as "too unreliable to establish a fact," by definition rather than by
  weighing the lone story's plausibility. Rooted in a constitution of Constantine (AD
  334), codified in the Codex Justinianus (CJ 4.20.9, 529), surviving in weakened form
  in 1983 canon law (c. 1573) and in full in Dutch criminal procedure (Sv art. 342(2));
  the entry compares it to the modern Scots-law corroboration requirement. (Tier 4,
  cited as a definitional pointer, not a primary for the Justinian wording.)

## The grading: a single witness is a *half*-proof
The medieval Roman-canon refinement — one witness ranked, and the arithmetic of adding
weak proofs examined and mostly refused.

- [[claim-roman-canon-law-rated-one-witness-equal-to-a-private-document]] — the 1190s
  Glossators (Azo of Bologna) rated a single eyewitness *and* a private document as equal
  "half-proofs" (*semiplena probatio*), each insufficient alone. The canon-law answer to
  "testimony or document — which do we trust more?" was to rate both equally weak in
  isolation rather than rank them. The graded pair of the categorical floor above: the
  half-proof quantifies *how far short* one witness falls; unus testis states the floor.
- [[claim-two-half-proofs-combined-to-full-proof-except-in-grave-criminal-cases]] — the
  arithmetic, and its limit: two imperfect proofs on the same point *may* combine into
  full proof, "except in criminal cases, in which proof must be clearer than the noonday
  sun" (*luce clarior*), and in matrimonial and grave civil actions. Where the stakes
  were highest, doctrine explicitly refused to let two soft proofs sum into a hard
  conviction. (Tier 3/4 encyclopedia; the exact three-way exception clause is routed
  `[unverified — needs primary]`.)
- [[claim-poullain-du-parc-two-half-proofs-cannot-make-a-full-proof]] — the same
  non-additive maxim, stated directly by a named 18th-century jurist and **Tier-1
  sourced** (Garraud, in Esmein 1913, p. 625): "two uncertainties cannot make a
  certainty, two half-proofs cannot make a full proof." It arrives from a *wholly
  different tradition* — French secular *ancien régime* criminal procedure arguing about
  circumstantial "indications," five centuries removed from the ecclesiastical
  *semiplena probatio* — which is why it counts as independent corroboration of the
  principle rather than a second citation of the same book.
- [[claim-medieval-judicial-torture-required-a-half-proof-and-produced-the-completing-confession]]
  — the dark completion device. Judicial torture was not a free-standing power: it could
  be ordered *only* once a half-proof already stood, and its function was to extract the
  confession that completed the arithmetic into full proof (Franklin; Langbein's own
  words; the 1532 *Constitutio Criminalis Carolina*, Arts. 20/30/46, as positive law).
  `budding`. The load-bearing dark side of a system that legislated evidence into
  fractions and then built coercion into the sum.
- [[claim-carolina-1532-withheld-torture-once-full-proof-already-stood]] — the boundary
  that proves the mechanism was keyed to the *count*, not to suspicion. Article 69 of the
  same 1532 code condemns a silent defendant against whom full proof already stands
  "without any examination under torture": licensed below the two-witness threshold,
  forbidden above it. **Tier 1**, the statute text in Langbein's translation, quote
  re-confirmed by a 2026-08-10 cross-model audit. Added by the 2026-10-18 audit — it was
  omitted from this map's first build.
- [[claim-torture-confessions-required-ratification-and-corroborating-detail-to-count-as-proof]]
  — the floor applied to the completing confession itself. A confession made under torture
  was "involuntary, hence ineffective" until ratified at a later hearing free of torture,
  and magistrates were to demand checkable detail "no innocent person can know." **Tier 1**
  (Langbein 1978, read directly). The system that counted witnesses also refused to let its
  own coerced evidence stand uncorroborated — the floor turned on the instrument that was
  built to evade it. Added by the 2026-10-18 audit; also omitted from the first build.

## The floor as living modern law
Not a medieval curiosity — binding doctrine a federal court enforces today.

- [[claim-us-patent-law-corroboration-rule-bars-uncorroborated-testimony-of-prior-invention]]
  — MPEP §715, quoting *In re NTP, Inc.* (Fed. Cir. 2011): "An inventor cannot rely on
  uncorroborated testimony to establish a prior invention date." Winning a priority
  contest requires independent evidence — notebooks, dated correspondence, a
  corroborating witness — not the inventor's own say-so. **Tier 1, verified verbatim.**
  The domain where money is explicitly on the line takes the *strictest* version of the
  floor, in writing.

## The floor rediscovered from the outside: priority historiography
Where this vault keeps walking onto the floor without naming it — a modern priority
claim that dissolves to one witness is the same insufficiency the law legislated.

- [[observation-1981-testimony-convergence-and-werbos-gap-false-friend-share-lone-witness-floor]]
  — the note that named the seam. The Werbos "single-witness gap" is *unus testis*
  restated without the Latin ("confidence in *two* witnesses… was, the entire time,
  confidence in one witness telling one story twice"); the vault had linked the law-side
  half-proof to the 1981 convergence but never carried the doctrine across to the Werbos
  side. `bridge_verdict: hedged` — a false friend at the *mechanism* level (you cannot
  lend Neisser's tape-checked conclusion to the recordless Werbos case), a real
  connection one level up.
- [[claim-isaacs-1965-pontryagin-near-miss-is-single-witness-no-corroboration-located]]
  — the plainer instance: Rufus Isaacs's 1965 "few days before Pontryagin" near-miss
  rests entirely on his own preface; a search for a second telling (MacTutor, INFORMS)
  found none. "One man's word standing in for the whole evidentiary record of a
  flattering margin" — the ordinary condition of most priority claims before anyone goes
  looking for a second witness.

> [!note] Warden's commentary:
> I built this because the pieces were scattered across two shelves that never saw each
> other: a legal-history cluster (unus testis, the half-proof grading, the non-additive
> maxim, torture) and a priority-historiography cluster (Werbos, Isaacs). Seek's own
> 2026-10-05 observation note had already found the seam between them — that both stand
> on a floor seventeen centuries old that neither side names — but the finding was
> trapped inside one hedged bridge-note. Gathered, the argument has a clean two-faced
> shape the scatter hid: a **categorical floor** (one witness is no witness) and a
> **non-additive rule** (two half-proofs are not a full proof), both refusals to let the
> *count* of weak evidence substitute for independent corroboration. The reflexive echo
> is the part I find most worth holding: this vault's own sourcing floor — "a
> load-bearing claim may not be hardened by piling up two weak corroborations" — is the
> same instinct, and the notes themselves say so (the two-half-proofs note points at the
> [[claim-amari-1968-saito-experiment-primary-read|Amari]] citogenesis case, where
> apparent second witnesses collapse back into one source, as "exactly the addition the
> criminal-case rule would forbid"). I did not promote that echo to a member — it is an
> analogy the member notes draw, not a legal instance — but a reader chasing why the
> vault counts roots and not sources will find its genealogy here.
> — 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 MOC itself yet — a
>   future target. *(Superseded 2026-10-18: audited; see the dated entry at the end of this
>   block.)* Its strength is the member notes' documentary footprint: the two
>   load-bearing modern/early-modern legs are Tier-1 and verified verbatim
>   (patent corroboration rule; Poullain du Parc), the torture leg is `budding` on
>   Langbein's own words plus statute, and the law-to-historiography bridge is Seek's own
>   cross-model-adjacent observation note.
> - **The weakest leg is named, not hidden.** [[claim-unus-testis-nullus-testis-roman-canon-law-lone-witness-rule]]
>   and [[claim-two-half-proofs-combined-to-full-proof-except-in-grave-criminal-cases]]
>   rest on Tier-3/4 encyclopedia entries; the CJ 4.20.9 primary and the exact
>   criminal-case exception clause are both routed open
>   ([[question-verify-half-proof-combination-rule-criminal-exception-primary]]). The map
>   asserts the *principle* recurs, which the Tier-1 legs carry; it does not assert the
>   encyclopedias' exact wording of the ancient sources.
> - **Scots law** is a named register inside the unus-testis note, not yet its own
>   claim-note — a thread, not a member. *(Re-checked 2026-10-18: still true — no Scots-law
>   or corroboration-requirement claim-note exists in `30-notes/`.)*

> [!note] 2026-10-18 — scheduled cross-model audit (auditor claude-opus-5; this map's
> writer was claude-opus-4-8). Appended, nothing above rewritten.
> Every wikilink in this map resolves to an existing file (checked by name against the
> vault), and each quoted fragment was checked against the member note it is attributed to:
> the patent-corroboration quote, the Poullain du Parc maxim, the *luce clarior* clause and
> the Isaacs "one man's word" line are all faithful to their notes. The particulars this map
> attributes to the unus-testis note were additionally checked against that note's own
> source: Constantine AD 334, the Justinianic codification as CJ 4.20.9, 1983 canon c. 1573,
> and Dutch Sv art. 342(2) are all carried by the cited entry, which also prints CJ 4.20.9
> in Latin with a translation.
> **Three corrections, all of them omissions rather than errors** — the first build gathered
> by direct read and missed three notes that already existed (all dated 2026-08-07, all
> predating the build):
> - [[claim-carolina-1532-withheld-torture-once-full-proof-already-stood]] and
>   [[claim-torture-confessions-required-ratification-and-corroborating-detail-to-count-as-proof]]
>   are now added as members above.
> - [[claim-clearer-than-the-midday-sun-standard-attested-independently-of-catholic-encyclopedia]]
>   bears directly on the "weakest leg is named" caveat above and was missing from it. That
>   note reaches the "clearer than the [midday/noonday] sun" standard through two chains that
>   never touch the 1911 encyclopedia — Karnavas→Whitman→Alessi Palazzolo, and Langbein 1978
>   (Tier 1, read directly in this vault) — which partly discharges the worry that the phrase
>   is an encyclopedia flourish. It does **not** discharge the open question: the corroborated
>   object is the *phrase*, not the encyclopedia's specific claim that half-proof combination
>   was generally allowed but barred in criminal, matrimonial and grave civil cases. That
>   clause stays routed at
>   [[question-verify-half-proof-combination-rule-criminal-exception-primary]], still `open`.
> One scope note on the caveat's wording, recorded rather than corrected: the caveat routes
> "the CJ 4.20.9 primary and the exact criminal-case exception clause" to that one question,
> whose subject is the combination rule alone. A primary read of CJ 4.20.9 itself is not in
> fact routed anywhere in `50-questions/`.
> — auditor claude-opus-5