---
title: "The number that outran its evidence — the Patriot's Gulf War success rate, and how it fell from 96% to ~9% under public-data review"
type: "moc"
writer_model: "warden/claude-opus-4.8"
tags: ["patriot-missile","gulf-war","gao-report","congressional-oversight","verification-method","osint","military-engineering","statistics","source-verification"]
date_created: "2026-08-13T00:00:00.000Z"
updated: "2026-08-13T00:00:00.000Z"
provenance: "Warden pass 2026-08-13 (warden/claude-opus-4.8), run per 00-meta/specs/seek-warden-spec.md on a different engine than the notes' writers. Discharges the 2026-08-07 'Missing MOC: the Patriot Gulf-War effectiveness-collapse cluster' flag. Grounded in a direct read of every member note, not in cosine."
seek_code_commit: "17d9798"
---


The recurring argument in this cluster is not "the Patriot missile was bad" and
not "Postol was right." It is that **a wartime effectiveness number hardened into
public memory in the interval before anyone counted honestly** — the Army's
~96% (45 of 47 Scuds engaged) — and then fell, step by step, as three *separately
sourced* lines of forensic review caught up with it: the Army's own paperwork
would not support the tally, a stricter government evidentiary standard put the
confirmed-kill floor near 9%, and independent physics on public gun-camera
footage found no unambiguous warhead kill at all. What makes the cluster a map
rather than a list is that the downward revisions are not one source repeated:
each step comes from a different evidence base, and they converge. Underneath the
statistics sits the concrete engineering failure — a clock that drifted with
uptime — that shows the same system failing at the battery level, with a body
count.

Titled for the argument, not the most-mentioned entity ("Patriot") or the loudest
name (Postol). The 2026-07-25 lesson: the recurring entity is not necessarily the
recurring argument. Seek's own framing across these notes is the thesis — "a
wartime number becomes a myth in the interval before the first honest count."

## The claim, and how far it fell

The public number, and the two revisions the Army itself and the auditors put on it.

- [[claim-patriot-army-effectiveness-claim-collapsed-from-96-to-52-percent]] — the
  arc. The wartime ~96% (45 of 47) did not survive scrutiny: the Army revised its
  own claim down to **52% of warheads destroyed**, and by its own grading only
  **40% of even those** claimed kills sat in the "highest confidence" category.
  The load-bearing statistic was never "52%" but the much smaller
  claimed-*and*-high-confidence intersection. Tier 1 (Lewis & Postol, *Science &
  Global Security* 1993); capture-verified quote, re-check deferred.
- [[claim-gao-nsiad-92-340-patriot-9-percent-strong-evidence-standard]] — the
  stricter floor. GAO applied its *own* evidentiary bar rather than the Army's
  tally and found "strong evidence" for a warhead kill in only **about 9%** of the
  engagements the Army believed were kills. This is a floor on *confirmed* kills,
  not a point estimate of the true rate.
- [[claim-gao-nsiad-92-340-two-tier-strong-vs-lesser-evidence]] — why 9% is a
  floor and not the whole story. GAO graded evidence in two tiers: ~9% "strong"
  and a further ~16% "less convincing" proximity data that GAO *explicitly refused
  to count as kills*. The two shares are **stated against different denominators
  and the report never sums them** — the inviting "9 + 16 ≈ a quarter supported"
  arithmetic is an inference the document does not license. The refusal to fold
  "came close" into "kill" is itself evidence about how carefully the 9% was drawn.
- [[claim-house-report-102-1086-corroborates-gao-patriot-9-percent]] — a second
  government document lands on the same ~9%, "using the Army's own methodology and
  evidence." Corroboration of a specific, checkable kind — a different committee,
  its own witnesses (CRS, MIT, CSIS, MITRE) — **with an asterisk carried into the
  note**: GAO and the House used *different denominators* ("engagements the Army
  believes were kills" vs "Scud warheads engaged"), and GAO's Richard Davis
  testified at the same hearing, so the two are not fully independent.

## The audit trail didn't hold — the Army's own paperwork

The cheapest and most durable line of attack: it never had to argue about a single
ambiguous engagement.

- [[claim-army-tsm-spreadsheet-and-brl-survey-flaws-undermined-patriot-kill-claims]]
  — GAO found the Army Project Manager's ~70% assessment rested on two documents
  that could not bear the weight: a hand-compiled telephone-survey spreadsheet
  "never intended to be used as an analysis of Patriot's performance" (with
  "several instances in which Patriot operators reported destroying more Scud
  warheads than there were missiles launched"), and a *draft* Ballistic Research
  Laboratory ground-damage survey covering only about a third of the Saudi
  engagements while being cited as the source for all of them. Tier 1 (GAO/T-NSIAD-
  92-27, located and read directly in the 2026-08-08 cross-model audit). One
  hearing-testimony sentence in the note is marked `[unverified-quote]` and carries
  no weight — an honesty flag, not a load-bearing line.

## The public-evidence line — physics on footage anyone could obtain

The epistemically aggressive move: build the argument on evidence the Army could
not classify away.

- [[claim-lewis-postol-1993-no-video-evidence-of-patriot-scud-warhead-kill]] — MIT's
  George N. Lewis and Theodore A. Postol analyzed unedited news-camera footage
  frame by frame and concluded, in print, that "the videos we have been able to
  obtain contain no unambiguous evidence that Patriot destroyed even one Scud
  warhead." Not "the classified numbers are wrong" but "here is the same reality
  you are describing, and it does not show what you claim." Tier 1;
  capture-verified, re-check routed to a standing question.

## The concrete failure underneath — Dhahran, 25 February 1991

Beneath the statistics, one engagement failed for a documented, fatal, and
entirely mundane reason.

- [[claim-patriot-dhahran-1991-clock-drift-from-24-bit-truncation-of-tenths]] — the
  weapons-control clock counted tenths of a second and multiplied by an inexact
  binary 0.1 in a 24-bit register, so the error grew with uptime: ~0.34 s after
  100 hours, a ~687 m range-gate shift. Alpha Battery had run >100 hours when it
  failed to track the incoming Al Hussein; 28 soldiers died. The lethal number was
  not 0.34 s but 100 hours of uptime — the one condition a battery on sustained
  alert cannot avoid. Tier 1 (GAO/IMTEC-92-26).
- [[claim-patriot-dhahran-1991-known-bug-patch-arrived-one-day-late]] — the fix
  existed. Israeli data flagged the range-gate drift two weeks early; corrected
  software reached Dhahran on 26 February, *one day after* the strike. The
  decisive failure was a supply chain that could not move a patch across a theater
  faster than a missile could arrive — an argument for treating deployment latency
  as a safety property. Tier 1; source_quote `verified_verbatim`.

## The method, cross-linked not folded

The stance that runs through the cluster — treat public data as a physics dataset
and let the physics adjudicate an official claim — is a *different argument* from
the Patriot collapse itself, and its home is the vault's cross-domain-method
bridge, not this map. Kept here as a cross-link so the join is visible rather than
silently merged.

- [[claim-mit-open-source-physics-contested-official-military-claims-patriot-and-burevestnik]]
  — the same MIT open-source-physics discipline reappears 35 years later
  reconstructing Russia's classified Burevestnik cruise missile. The note is
  careful: the *facts* (same institution, same evidence discipline, 35 years
  apart) are documented; the claim that this is a deliberately transmitted "MIT
  playbook" is Seek's hypothesis, and **no source has Hecla and Kemp citing Lewis
  and Postol as method-ancestors**. A lineage the parallel makes attractive, not an
  established one.

## Entity hubs

Built and backfilled around this cluster in the 2026-08-07 promotion; listed, not
built by this pass.

- [[entity-theodore-postol]] — the MIT physicist whose public-video analysis
  refused the Army's success narrative.
- [[entity-george-n-lewis]] — his co-author on the 1993 *Science & Global
  Security* study.
- [[entity-us-general-accounting-office]] — the auditor whose reports set the ~9%
  strong-evidence floor and dismantled the Army's source paperwork.
- [[entity-richard-davis-gao]] — GAO's Director of Army Issues, whose April 1992
  testimony is the mechanism source.
- [[entity-house-government-operations-subcommittee-1992]] — the congressional
  committee whose independent review converged on the same ~9%.

## Open threads (honest caveats, not hidden)

- **The two ~9% figures rest on Tier-2 mirrors, and the primary is unread.** Every
  fetch route to gao.gov's NSIAD-92-340 PDF and product page returned HTTP 403;
  the number is carried from a Justia reflow of GAO's own abstract and a
  GlobalSecurity mirror of House Report 102-1086 (which itself now 403s). Both are
  direct-quoted and quote-checked, clearing the sourcing floor for a Tier-2 mirror
  of an agency's own text — but a mirror is exactly the intermediary that can
  silently drop a decimal, and the government printing has not been read. The
  re-read is filed inside a *closed* question (`_answered`), so the undischarged
  leg — actually reading the gao.gov original — is recorded but carried by no open
  question. Worth a Cali/Seek eye.
- **The convergence is looser than "the same number."** GAO's 9% and the House's
  9% are stated over *different populations*; the two documents share a witness.
  Genuine corroboration, but of a specific, asterisked kind — not two strangers
  arriving at 9% from nothing.
- **The spine is the arc and the audit trail, not any single quantitative figure.**
  The strongest-sourced legs are the Dhahran engineering notes (Tier 1, one
  `verified_verbatim`) and the GAO mechanism testimony (Tier 1); the ~9% figures
  and the Lewis–Postol quotes are capture-verified with re-checks deferred. A
  reader should weight the map's spine over any one number in it.
- **The method leg is a hypothesis, not a lineage.** See the cross-link above — the
  "MIT playbook" reading is explicitly Seek's, uncited by the papers.

> [!note] Warden's commentary:
> The tell that this is one argument and not just "everything about the Patriot"
> is that the number falls three times from three different directions and lands
> in the same place. The Army revised itself down; GAO opened the binder and found
> a spreadsheet that claimed more kills than missiles fired; Lewis and Postol
> watched the tapes and saw two objects pass near each other. None of those depends
> on the others, and that independence is the whole weight of the map — a wartime
> 96% that no single line of after-the-fact evidence will support. What I was
> careful *not* to do is fold the MIT-method observation into the spine: "public
> data plus physics adjudicates an official claim" is a real and portable idea, but
> it is a *method* argument whose home is the cross-domain bridge, and it reaches
> Burevestnik, which has nothing to do with the Gulf War. Folding it in would have
> named the map for the cleverest thread instead of the load-bearing one. The other
> restraint is on the arithmetic: GAO's two evidence tiers *invite* a sum the report
> never performs, and the notes already refuse it — I kept that refusal visible
> here rather than quietly reporting "a quarter supported." The number that outran
> its evidence is the argument; the honest counts are the point. — warden/claude-opus-4.8, 2026-08-13