The Roast Rating Standard

Why this exists

AI made polish free.

Anyone can now produce a flawless pitch deck, business plan or strategy memo in thirty seconds. The fonts are right. The charts are clean. The market size is enormous. And none of it tells you whether the thinking underneath was ever tested.

Worse, the AI that wrote it was built to agree with the person who asked. It will call a weak idea "promising" and a guess "a data-driven insight". Every deck now looks like a winner, so the look of a winner means nothing.

The Roast Rating exists to fix one thing: tell people how much of a decision has actually been tested, in a form anyone can check.

We do not write the deck. We rate it.

What the rating measures, and what it does not

The Roast Rating measures how well a decision has been tested. It grades evidence: named problems, verified facts, tests run, money received.

It does not predict whether the idea will succeed. Nobody can honestly grade the future, and anyone who claims to is selling something. A 5-flame idea can still fail. A 1-flame idea can still work. The rating tells you how much you are betting on belief versus evidence.

The rating is separate from the verdict. Every roast returns two things:

The verdict: Green Light, Reshape or Kill. This is the Council's judgement of the idea.
The rating: 0 to 5 flames. This is how well the decision behind it has been tested.

They can point in opposite directions, and that is the point. A Kill with 4 flames is a success. It means someone tested the idea properly, found it dead, and saved themselves a year. We would rather rate that founder highly than flatter one who never tested anything.

The scale

Flames	Name	What must be true	Evidence required
0	Untested belief	An idea exists.	None.
1	Problem named	There is a specific buyer, a specific pain, and the single most likely way the idea fails has been named.	A written statement of buyer, pain and fatal flaw.
2	Survived the roast	The fatal flaw has a credible answer, and the key assumptions are listed.	A completed roast by all six voices, with the assumptions written down.
3	Receipts checked	Prices, competitors and market size have been checked against real, dated sources. Anything not verified is marked unverified.	Source links with dates for every material claim.
4	Tested	A test was designed with a pass mark set before it ran, it was run, and the result was logged, pass or fail.	Proof of the test and its result (sign-up data, ad results, interview records, a live page), reviewed by us.
5	Proven	Real customers have paid real money, at a price that could sustain the business.	Proof of payment or signed commitment (receipts, contracts, letters of intent with terms), reviewed by us.

Notes on the scale

Each level includes the ones below it. You cannot reach 4 flames without having named the problem and checked the receipts.
A failed test still earns 4 flames. The rating rewards the rigour of testing, not a flattering result.
Investor money is not proof. Funding shows that investors believed. It does not show that customers pay. Money from investors counts toward nothing on this scale.
Selling below cost is not proof. Five flames requires customers paying a price that could sustain the business. Revenue bought with losses shows demand, which is 4 flames at most.
Friends, family and free users are not proof. Five flames requires strangers who paid.

How a rating is issued

Every roast is run by the six voices of the Council:

The Wrecking Ball finds the fatal flaw.
The Dreamer With A Calculator finds the realistic upside.
The Toddler strips the borrowed assumptions.
The Receipts checks the facts against live sources.
The Wallet plays the buyer deciding whether to pay.
The Final Word rules on the verdict and sets the cheapest test.

A roast can be run in two ways, and the rating always says which.

Council protocol. The roast is run by the user's own AI model, steered and checked stage by stage by our system. Any stage that is vague or flattering is rejected and sent back. Ratings issued this way go up to 3 flames and are marked self-assessed.
Certified. The roast is run on our own system, and any evidence for 4 or 5 flames is reviewed by us. Only Certified ratings can reach 4 or 5 flames.

Ratings on public evidence

In our own content we rate decisions that are already public: companies that have closed, decks their founders have published, and announcements reported in the press. These ratings are run by the Council on our own system and are labelled "rated on public evidence" with the date. They use only what was publicly reported on that date, and their sources are listed. They can reach any level on the scale, because the evidence is in the open, but they are not Certified: nobody submitted proof to us, and we do not claim to know anything the public record does not.

What every rating discloses

Every rating is published on its own verification page, and every verification page shows:

The verdict and the rating.
The date the rating was issued.
The mode: Council protocol or Certified.
The model: which AI model ran the roast. Where we cannot verify it, we say so: "reported as [model] (self-reported)".
Voice agreement: how many of the five debating voices reached the Final Word's verdict (for example, "4 of 5").
Evidence strength: High, Medium or Low, based on how many material claims are backed by dated sources and which critical assumptions remain unverified.
The sources used by The Receipts.

What we will not show: an invented confidence percentage

AI models will produce a confident-sounding percentage on request. That number is not calibrated. A model that says "82%" is no more likely to be right than one that says "60%". Publishing it would be the exact kind of flattery this rating exists to end, so we do not.

We will publish a real percentage when we have earned one. Every rated decision is followed up at 30 and 90 days: what happened? Once enough outcomes are logged, ratings will carry a measured line, for example: "Past Reshape verdicts at 3 flames held up X% of the time." That figure will be calculated from real outcomes, and we will publish how we calculate it.

Expiry

Evidence goes stale. Markets move, competitors launch, prices change.

Every rating shows its date.
A rating expires 12 months after it is issued and is displayed as expired, with an invitation to re-rate.
A rating can be re-issued at any time with new evidence. Going from 3 to 4 flames means running the test and submitting the result.

Independence

A rating is only worth the independence behind it. Credit ratings failed in 2008 partly because the grade drifted toward the people paying for it. We have written the rules to prevent that.

The grade is never for sale. The fee is the same whatever the result. There is no paid appeal and no premium review.
The criteria are public. This document is the full standard. There are no hidden factors.
We publish our record. Once outcome volume allows, we will publish how often our verdicts held up, by verdict and by flame level, including where we were wrong.
We rate on request only. We rate a live venture's deck or decision only when its owner submits it. In our public content we rate historical, public-record decisions: companies that have closed, and decks their founders have already published.
Revocation. If evidence submitted for a rating turns out to be false, the rating is revoked, and the verification page says so.

How to check a rating

A rating only counts if it links to a live verification page on roastcouncil.com.

Every Roast Rating badge carries a QR code and a short link in the form roastcouncil.com/v/…. If a deck shows flames with no link, or the link does not resolve to a matching verification page, treat it as unrated.

On the verification page, check four things:

The name and date match the document in front of you.
The rating is not expired or revoked.
For 4 or 5 flames, the mode says Certified.
The model and sources are listed.

Limits

The Roast Rating grades how well a decision was tested. It is not investment, legal or financial advice, and it is not a prediction of success.
It rates the evidence submitted and found at the time of rating. It cannot rate evidence it was never shown.
A rating reflects the Council's standard as published in the version current on the date of rating.

For accelerators, investors and grant panels

If you screen many ideas, the Roast Rating gives you one number, checkable in a click, telling you how much of each application rests on evidence. Some programmes will want a minimum rating to apply. Others will use it to see which founders test their ideas and which only describe them.

We are looking for a small group of Founding Raters, programmes that will pilot the standard with their applicants and help shape version 2. Apply at roastcouncil.com.

Versions

1.0 · September 2026. First publication.

We will revise this standard in the open. Every change will be dated and listed here, and every rating states the version it was issued under.

The Roast Council. A roast, not a hug. roastcouncil.com