Devil's Advocate

• Generates a candid, adversarial written brief arguing the weaknesses in your case, streamed in real time as a single long-form report rather than a back-and-forth exchange.
• Surfaces patterns from Merito's Outcome Library for any criterion your case file doesn't cover yet, so you can see what other tracked outcomes show about criteria you have no evidence for at all.
• Keeps a full history of past briefs so you can regenerate after adding evidence and compare what the adversarial read looked like before and after.

Screenshot placeholder: the fixture case (phase5-test@merito.test) has two real Devil's Advocate reports on record — one complete, one an intentionally-failed historical credit-refund test (per docs/DEMO_CASE.md) — but no screenshot was captured this content pass. See the report's evidence table for why. "Reading the result" below documents the real UI elements from the component source.

Devil's Advocate sits in the Stress Test stage of the Case Journey (journeyStageMeta.ts), alongside Officer Simulator and Debate Mode.

• Before: build your Case File up with real evidence first — this tool argues from what's actually documented, so a thin file mostly just tells you the file is thin, not something more specific.
• After: the cross-sell cards on the result point at exactly the two natural next steps — run Debate Mode to see the same file defended, not just attacked, or open Petition Studio to check whether the petition draft itself (not just the underlying evidence) is part of the weakness. If you're already past filing and facing a real denial with similar grounds, Denial/NOID Response is the tool for that.

• Input: none beyond an existing, ingested case — no text to type, no file to upload for this run itself.
• Case File prerequisite: yes — the case must be ready before you can generate a brief.
• Typical time: a couple of minutes to stream the full brief; longer if you also read through the linked Outcome Library patterns for uncovered criteria.
• Credit cost: 2 credits per brief (Pro plan or credits).

• The brief itself — a single freeform written report (not broken into fixed sections like Debate Mode's Judge panel), each claim tied where possible to the specific Case Canvas node it targets.
• Watermark — every brief carries "SIMULATED ADVERSARIAL ANALYSIS — NOT A PREDICTION."
• Patterns for your uncovered criteria — a separate card listing, per criterion your case file doesn't yet cover, observed patterns from the Outcome Library. Shown only when the library actually has data for those criteria; an explicit empty message otherwise, distinguishing "no evidence in your case" from "no data in our library."
• Cross-sell: Run the Debate — a direct link to see the same file defended by an Advocate against the same skeptical read, in Debate Mode.
• Cross-sell: Open Petition Studio — a direct link to check whether the petition draft's own readability, generic language, and evidence linkage are part of the weakness this brief just found.
• History — every generated brief, including failed ones (useful for confirming a credit refund actually happened), tagged with the Case File version it was run against.

• Devil's Advocate is not on the Free plan — it's explicitly listed in plans.pro.features_json (devils_advocate).
• A 7-day Pro trial includes 40 credits — at 2 credits per brief, that's up to 20 briefs during the trial window, shared with every other Pro tool.
• The trial checkout collects a card up front (a standard Stripe subscription checkout with a trial period, not a card-free trial), but cancellation is straightforward — cancel anytime from Settings before the trial ends and you won't be charged. Start your free trial to generate your first brief.

• Running it on a barely-started case. A file with almost no evidence produces a brief that mostly restates "there's nothing here yet" — build real evidence first for a brief that actually stress-tests specific claims.
• Treating the brief as a prediction of what USCIS will say. The watermark is explicit: this is a simulated adversarial analysis, not a prediction — it exists to find weaknesses before anyone else does, not to forecast an outcome.
• Ignoring the uncovered-criteria patterns because the main brief looked fine. The brief can only attack what's actually in your file — the separate patterns panel is specifically there to flag criteria with zero evidence, which the main brief can't meaningfully critique.
• Regenerating after a cosmetic edit and expecting a different verdict. Like Debate Mode, this tool reacts to documented evidence, not to how persuasively something is worded.
• Forgetting only the case owner can generate a new brief. Collaborators can read History but the generate button is owner-gated.