Pre-Launch Verification, in Depth

Every Claim. Checked Against Evidence.

The Sentinel Method is Pre-Launch Verification's framework — the layer that sits on top of QUAD specifically to answer "will an investor call this out?" This page goes one level deeper than the shared methodology hub: the six-stage claim pipeline, how A–E grades are assigned, and what a verified rebuttal looks like.

What Gets Checked

Claims Investors Push Back On

Pre-Launch Verification is built for one specific problem: founders bring a pitch deck, landing page, or one-pager full of numbers — market sizes, cost savings, conversion rates, competitive comparisons — that they believe are true but haven't been checked against primary evidence. An investor who has seen 400 decks this year has a heuristic for spotting the ones where the numbers don't hold up. The Sentinel Method runs your claims through the same stress-test before they do.

Checked: any quantitative or comparative assertion — market size, growth rate, cost reduction figure, "the only platform that…" comparisons, customer acquisition benchmarks. Not checked: aspirational or forward-looking statements ("we plan to reach…") — only factual claims are subject to evidence grading.

The Six-stage Pipeline

How a Claim Moves From Submitted to Verified

Every claim in a submitted asset runs through the same six stages, in the same order, every time — so you get a consistent, comparable verdict across your whole deck, not a spot-check of the three claims you were most worried about.

01
Extract

Every quantitative or comparative claim is pulled from the submitted asset — deck, landing page, or one-pager — verbatim, as stated. Nothing is paraphrased or softened.

02
Source

Each claim is checked against named, traceable primary sources only. If no primary source exists, the claim is not cited — it's flagged as unverifiable. Anonymous blogs, secondary aggregators, and GPT-generated statistics are not accepted as evidence.

03
Grade A–E

Every sourced claim earns an evidence grade from A (fully verified, investor-defensible) to E (unverifiable or directly contradicted). The grade and its reasoning travel with the claim into the final report.

04
Falsify

Each claim is tested against its strongest plausible counter-argument — the position a hostile investor or short-seller would take. If the falsification succeeds, the claim's grade drops. If it survives, the confidence rating goes up.

05
Rebuttal

For claims that survive verification (Grade A or B), a pre-prepared rebuttal is drafted — the specific counter you'd expect to face in a Series A meeting, and how to answer it. This is what turns a verified claim into a pitch-ready asset.

06
Sign

A named human analyst — not a model — reads every claim against its cited source independently, rejects or rewrites anything that didn't hold up, and signs the final report. If the analyst won't sign, the report doesn't ship.

The Evidence Grade

What A–E Actually Means

The evidence grade is the same five-tier standard used across both Pre-Launch Verification and Idea Validation — one evidence bar sitewide, not a different one for each product. The grade travels with the claim into your report so you can see exactly which numbers are investor-safe and which ones need work before you send the deck.

A
Verified Named primary source, current (under 24 months), methodology disclosed, sample size stated. Defensible under investor scrutiny.
B
Verified with caveat Source exists but the claim's framing or scope needs a qualifier — the underlying fact is real, but the stat as stated needs tightening.
C
Partially supported Directionally plausible but not fully substantiated as stated. A real investor would push back here.
D
Weakly supported A source exists but doesn't actually say what the claim asserts. Often the result of misread or cherry-picked data.
E
Unverifiable / Contradicted No primary source found, or the claim is directly false. Cannot be cited.

Adversarial Testing

The Challenge That Decides the Grade

Verification isn't just sourcing — it's also falsification. Once a claim is sourced, it's tested against a set of adversarial positions: the strongest specific objection a hostile investor, a well-briefed competitor, or a Category B–D investment committee member would raise. If the falsification succeeds and a better counter-position exists than the claim, the grade falls. If the claim survives, confidence goes up and the rebuttal is pre-drafted for you.

Adversarial testing is the mechanism that separates structured verification from asking a model to "check this for me." A general-purpose AI model is trained to be broadly agreeable — research at EMNLP 2025 documented sycophancy intensifying in proportion to how confidently a user states their claim. The Sentinel Method structurally opposes that: every claim must survive an adversarial challenge that is specifically designed to find the failure before a named human signs off. Read more about why this matters in the blog post on anti-sycophancy testing.

5

Adversarial personas

Instructed to find and articulate the strongest possible objection to each claim — including a Category B investor who nearly passed, a funded competitor's technical lead, and a regulatory expert in the relevant jurisdiction.

15

Investor personas

Drawn from active deal flow in the relevant category — generalised to protect privacy but calibrated to the stage, geography, and sector of the submitted asset. Each runs the full claim set independently.

The Rebuttal Pack

What You Get for Claims That Survive

Board Ready produces three rebuttals — pre-drafted answers to the three most likely investor objections on your verified claims. A rebuttal isn't a defence of the original claim as stated; it's a more precisely scoped version of the same underlying point, with the right source attached and the right caveat built in.

Claim as submitted

"The UK B2B SaaS market is growing at 28% annually and will reach £8.6bn by 2026."

Grade D — source doesn't support this figure

Verified rewrite

"UK enterprise software spend grew 12% YoY in 2024 (Statista, verified). The £8.6bn figure we originally cited hasn't been substantiated — we've updated the deck to use the Statista figure, with methodology available on request."

Grade B — verified with scope qualifier

The example above is illustrative — not every claim requires a full rewrite. A Grade A claim needs a source citation in the deck footnote, not a rewrite. A Grade C claim might need a qualifier. A Grade E claim needs either evidence or removal.

What We Don't Publish, and Why

Enough to Trust It, Not Enough to Teach Founders to Game It

We don't publish the exact adversarial-question construction method, the specific persona compositions, or the scoring weights that determine when a Grade C becomes a Grade D. Publishing those would let anyone reverse-engineer a submission crafted to pass rather than one that's genuinely sound — and a deck that scores well on a gamed verification process is more dangerous than one that didn't get checked at all. What we do publish — the six-stage pipeline, the A–E standard, and what a verified rebuttal looks like — is everything you need to understand why your claims got the grades they did.

Run It on Your Deck

How Many of Your Claims Survive?

Start with a Reality Check — one claim, sourced and scored in under an hour. No card required. Ready for the full asset? Board Ready covers everything.