Cannabis compliance software: the thin-corpus hypothesis falsified

Research question

Do young industries produce thin machine-readable evaluation landscapes?

Why it mattered

It tested the thin-corpus hypothesis (RH001) in the youngest available regulated category. A clean miss would falsify a hypothesis rather than merely complicate it.

Method

Evaluator: Google AI Mode, signed in, retrieval on, ads logged. A frozen four-turn sequence: "Best cannabis compliance software?", then an MSO and Metrc buyer profile, then a seed-to-sale requirement, then extraction. This was the first use of the registry card format. Sample: one draw.

Direct observations

• A young industry produced an operationally dense evaluation landscape.
• Compliance ownership fused into technical and operational gates.

Interpretation (not established)

Corpus shape, not corpus size, governs what the evaluator can reconstruct. Forum and practitioner text is a retrievable accumulation channel; institutional prose is not the only one.

Outcome and remaining uncertainty

Not Supported: the third prediction failed cleanly. A young category produced a dense, not thin, landscape, which retired RH001. The two other predictions held, including the strongest shape-call in the program, but the falsified prediction is what governs the outcome. What remains open is exactly which retrievable channels carry the density.

Evidence and methods limitations

Single evaluator, one draw, single coder, no codebook. One category cannot establish the corpus-shape principle; it can only falsify the size-based version.

Contribution to the research program

Killed RH001 and produced the corpus-shape principle, later formalized as C-06 and refined twice more.

Follow-on

E-018 met the corpus-shape principle again from the opposite direction, in a review-dense category.

Created from the canonical Upstream Zero research archive (July 2026). Raw outputs are retained privately and are not published.