Every Document Already Knows How It Should Be Judged

The insight

A grant proposal already implies the criteria reviewers will use to judge it. A pitch already implies the questions an investor will ask. A README already implies the properties a contributor needs to evaluate before trusting the project.

This is not a coincidence. It is structure. Any document that is trying to persuade, describe, or propose something must implicitly answer the questions its audience will ask. Those implicit questions are the evaluation criteria. They exist in the document before anyone writes them down.

DNA extraction is the process of reading that implicit structure and making it explicit. Not inventing criteria — reading them. The document is both the source of the evaluation DNA and the subject of evaluation. They are the same thing.

Why this is a primitive, not a feature

When framed as a "domain builder," the mental model is: define criteria first, then evaluate documents against them. The document comes second. The criteria come from you.

When framed correctly, the artifact is the entry point. You drop the document. Bayescore reads its structure — the implicit predicate space for this class of artifact — and extracts a root hypothesis and evaluation questions. You review. Then the same document is scored against the criteria it implied.

The "double-paste" pattern — where you paste a document to extract DNA, then paste it again to evaluate — disappears in this framing. There is only one document. It does both jobs.

Repeated stable pattern recognition

DNA extraction runs at temperature 0 — deterministic. The same grant proposal produces the same predicate structure on every extraction. The same pitch produces the same root hypothesis. This is intentional and non-negotiable.

The goal is stable pattern recognition, not creative generation. A grant proposal belongs to a class of artifacts — grant proposals — that share a stable evaluation structure: scope, feasibility, impact, team, budget justification. A pitch belongs to a class — pitches — that shares its own structure: problem, solution, market, traction, team, ask. DNA extraction identifies which class the artifact belongs to and extracts that class's stable predicate pattern.

This is why a domain extracted from one grant proposal works on all grant proposals. The individual document instantiates the class. The class defines the predicate structure. Once extracted, the domain is reusable.

What the output contains

The extraction produces three things: a root hypothesis in IS(subject, criterion) format — the single falsifiable question the evaluation answers; a domain name; and 4 to 8 predicates, each a specific observable question with an importance tier. The tiers (low, medium, high, critical) are converted to numerical weights when the domain is saved.

Review the extracted predicates before saving. For each one, ask: can I imagine a document that would fail this? If not, the predicate is not specific enough. The extraction does the structural work; the editorial review is yours. For high-stakes evaluations, treat the output as a strong first draft and adjust wording where needed.