Why Sentinel
The independent second opinion.
Three judging models from rival vendors, every authority verified against open sources, and an honest account of what we can't see. Here's why that beats checking your research on the same system that produced it.
01 / The comparison
"Doesn't Westlaw already do this?"
Partly — and the part it doesn't do is the part that gets lawyers sanctioned.
| Existence-only checkers | Westlaw Quick Check / Lexis Brief Analysis | Sentinel Citation | |
|---|---|---|---|
| Confirms the authority exists | Yes | Yes | Yes — against CourtListener, Cornell LII, eCFR, with the source on the verdict |
| Fetches and reads the authority's actual text | No | Partial (within their corpus) | Yes — and compares it against your use of it |
| Judges quote accuracy & holding support | No | Limited | Yes — four checks per citation |
| Independent of the tool that drafted the research | Depends | No — same vendor ecosystem | Yes — three models, rival vendors, plus authoritative sources |
| Flags its own misses with reasons | No | No | Yes — not-found and skipped carry explanations, never silence |
| Works on opposing counsel's brief | No | Awkwardly | Built for it |
| Requires a subscription | Varies | Yes — thousands/year | No. Pay per filing. |
One vendor can't grade its own homework.
The rule every sanctions order teaches: never verify a citation with the same system that produced it. If your brief was researched on Westlaw, checking it on Westlaw is one vendor grading its own work. Sentinel is the independent second opinion — three models from rival vendors, verified against open authoritative sources.
Even a clean brief earns its fee.
Suppose Sentinel finds zero fabrications. You still walk away with a Citation Strength map of the entire filing: which citations carry your argument, which are binding versus merely persuasive, which would fold under a motion to strike — yours and theirs. Fabrication-hunting is why you run it the first time. The strategic read is why you run it every time.
02 / Coverage & triage
Verification you can check. Triage you can act on.
Sentinel verifies citations against open, authoritative sources — CourtListener (9M+ opinions, 2,000+ courts), Cornell LII, eCFR, with a Scholar fallback. Because the sources are open, every verdict carries a trail you can follow. And because we know exactly what those sources cover, Sentinel can do something subscription black-boxes can't: tell you precisely which citations are confirmed and which few still need your eyes.
What a result means
Found
The authority was located in an authoritative source and its actual text was fetched. It then gets the full four-check analysis. Trust it.
Not judged
Out of scope by design (state statutes and regulations, foreign law, treatises). Sentinel didn't try, so don't read anything into it — check these manually.
Not found
The citation parsed correctly but isn't in our sources. Most of the time this is a coverage gap, not a fabrication: a brand-new slip opinion, a Westlaw-only ruling, a state trial-court order. So every not-found is scored for fabrication risk on three signals (format plausibility, web footprint, fabrication-pattern analysis) and sorted into a tier:
- Likely fabricated All signals agree it looks invented. Demand the source document; if it can't be produced, treat it as fabricated.
- Mixed signals Could be a real, hard-to-index opinion or a plausible fake. Produce the source to resolve it.
- Unverified but plausible Recognized as the kind of real opinion our sources don't index well. Routine; verify on your subscription service.
The triage payoff
A 30-citation federal brief typically resolves to a long list of found-and-fully-judged citations and a short, prioritized list of items for human review — each with the reason attached. That's the cite-check you were going to do anyway, narrowed to the citations that actually need you.
Where to expect misses
A not-found in Supreme Court or published circuit territory is genuinely suspicious — that material is almost always indexed. A not-found on a bankruptcy, tax, specialty-court, or Westlaw-only state appellate cite is routine and usually real. Federal coverage is the product's home turf; that's why we say federal filings on the front page.
03 / Consistency
Run it twice. Same answer where it counts.
Every Sentinel verdict is produced by large language models, so we tested the obvious question: will it give you the same answer twice? We ran one six-page filing through the full pipeline 18 times across two independent studies — including ten consecutive runs in a single afternoon — and published what we found.
54/54
A citation whose retrieved authority text belonged to a different, unrelated case was scored vulnerable in every run (24/24 in Study I, 30/30 in Study II).
~70%
of citations got the identical verdict in every run; per-citation agreement averaged ~90%. Clearly strong citations and fatally defective ones got the identical verdict every single time.
0 drift
Across a two-hour back-to-back session of ten consecutive runs, behavior didn't drift at all.
A wavering verdict is a signal, not a glitch.
The verdicts that moved were confined to genuinely contested citations — the ones sitting on the line between defensible and attackable. One citation alternated between two labels across runs while its underlying score barely moved (~72, SD 3.4): the judgment was stable; it simply lives on a boundary. A wavering verdict is the system telling you this one is a close call — look here first. That's exactly what a litigator needs to know.
The jury is predictable.
Each judging model held the same evaluative posture in all 18 runs — the strict grader stayed strict, the lenient one stayed lenient. The meta-judge reconciles three known temperaments, not three random opinions. That's why the synthesis is stable.
The honest caveat
These studies measure consistency on one filing, run many times. They are not an accuracy benchmark across the universe of legal citations — that study is in progress and will be published the same way this one is: numbers first.
Read it twice if you want.
The strong citations will stay strong. Your first filing is on us.