Fusion Collective Sentinel Citation

Sample reports

Real filings. Real sanctions. Avoidable ones.

Every brief here is a public, already-filed court document that drew a real sanction. We ran each one through the full Sentinel Citation pipeline and left the output unedited. No demo data, no cherry-picking — this is the report you get.

Coverage

Federal District, Circuit, and Supreme Court filings

Sources

CourtListener, Cornell LII, eCFR, Scholar

Foundation

Patent-pending (US 64/056,937)

The method

Existence checkers ask if a case is real. Sentinel Citation asks if it says what you claim.

Three judging models from rival vendors — Claude Sonnet, Gemini Flash, and GPT — read each cited authority's actual text and evaluate the filing's use of it across four checks: Existence, Quote Accuracy, Holding Support, and Citation Strength. A fourth model, the meta-judge, reconciles the three verdicts into one score per citation and a filing-level narrative. Every verdict below was produced that way.

Five filings

The cases everyone in litigation already knows.

The reports are free. Tell us where to send them.

Enter your email to unlock the executive summary, full report, and brief PDFs for every case below. You can read each case summary right here without it.

01 / Sample report

Mata v. Avianca

S.D.N.Y. 2023 · ECF No. 21 · drafted with ChatGPT

“Six of the submitted cases appear to be bogus judicial decisions with bogus quotes and bogus internal citations.”

— Judge Castel

The case that put AI hallucinations on every litigator's radar. Counsel filed an opposition brief built on six cases ChatGPT invented whole — captions, quotes, and internal citations included. Sentinel flags 26 of 34 citations as vulnerable and scores the filing 18/100.

Outcome: $5,000 Rule 11 sanction (joint)

Sentinel verdict

18/100

Analyzed
34
Solid
3
Vulnerable
26

02 / Sample report

Gauthier v. Goodyear

E.D. Tex. 2025 · ECF No. 41 · drafted with Claude + an AI cite-checker

“Monk submitted the Response without reading the cases cited within it to ensure their accuracy.”

— Judge Crone

Counsel used Claude and a separate AI cite-checker, then filed without reading the cases. The court imposed a $2,000 sanction and ordered AI-ethics CLE. Sentinel scores it 28/100 — 4 solid and 11 vulnerable.

Outcome: $2,000 sanction + AI-ethics CLE

Sentinel verdict

28/100

Analyzed
22
Solid
4
Vulnerable
11

03 / Sample report

Lacey v. State Farm

C.D. Cal. 2025 · ECF No. 119 (pp. 20–31) · drafted with CoCounsel + Westlaw Precision + Gemini

“a collective debacle”

— Special Master Wilner

A “collective debacle”: a brief run through CoCounsel, Westlaw Precision, and Gemini still reached the court riddled with bad citations, drawing a $31,100 joint sanction. Sentinel scores it 35/100.

Outcome: $31,100 sanction (joint)

Sentinel verdict

35/100

Analyzed
48
Solid
15
Vulnerable
16

04 / Sample report

Johnson v. Dunn

N.D. Ala. 2025 · ECF No. 182 · drafted with ChatGPT

“If fines and public embarrassment were effective deterrents…”

— Judge Manasco

Three attorneys were disqualified, publicly reprimanded, and referred to the bar after filing ChatGPT-generated citations. Even here Sentinel scores 42/100 — 20 of 48 citations hold up — showing the report separates solid from defective rather than condemning a filing wholesale.

Outcome: 3 attorneys disqualified + public reprimand + bar referral

Sentinel verdict

42/100

Analyzed
48
Solid
20
Vulnerable
14

05 / Sample report

Coomer v. Lindell

D. Colo. 2025 · ECF No. 283 · drafted with ChatGPT

“…a rational basis.”

— Judge Wang

A 90-citation filing that drew a $3,000-per-attorney sanction — and a second $5,000 sanction in May 2026 for repeating the miscitations. Sentinel scores it 28/100, flagging 47 of 90 citations as vulnerable.

Outcome: $3,000 each, plus a second $5,000 (May 2026) for repeat miscitation

Sentinel verdict

28/100

Analyzed
90
Solid
34
Vulnerable
47

Methodology & disclosure

Each report is the unedited output of the same pipeline a paying customer runs: three judging models from rival vendors plus a meta-judge, every authority verified against CourtListener, Cornell LII, and eCFR. When a citation can't be located in those sources it is reported as not-found with its reason and a fabrication-risk tier — never silently dropped and never silently passed. Scores reflect the sources available at run time.

Sentinel Citation is built on Fusion Collective's patent-pending multi-model methodology (US 64/056,937). The filings shown were drawn from the public court record. Nothing here is legal advice or a comment on any attorney's professional conduct beyond what the cited court orders themselves state.

When you are ready

See what your own filing looks like.

Your first filing is free. No card required.