The method, open for inspection
Not a manifesto. The model, the trace, and the limits, written down so an expert can check us.
What is unknown matters as much as what is known. We mark it, with a citation for where we looked.
1 Claim, source, evidence, missingness
Everything you are told about a project is a claim. We trace each claim to its source and attach the class of that source.
When no source is found, that absence becomes part of the artifact. A gap carries a citation too: the list of places we searched and came back empty.
2 A ladder of source classes
Sources are weighed by who stands behind them, on a fixed public ladder described below.
A developer’s own document is evidence, discounted for interest. A forum post can raise a question and can never settle one.
3 Human Verification
Our engine assembles the evidence deterministically: the same inputs always produce the same picture, and no sales pressure can alter the outcome.
A human then reviews every artifact before it ships. Below our evidence floor, the honest output is no artifact, and that is what you get, with a refund.
4 Three postures, never a verdict
Every artifact ends in one of three postures. Verify: the evidence held; confirm the named items and proceed on your own judgment. Investigate: something meaningful is unresolved; here is exactly what. Hold: the tension between evidence and stakes is real; waiting costs less than being wrong.
We stop there deliberately. A verdict would replace your judgment; a posture arms it.
The source-class ladder
Five classes, applied to every source we cite. The class appears next to every row of every artifact.
- Class 1 · Primary and authoritative. Government registries, courts, official gazettes. Read at full weight.
- Class 2 · Official but interested. The developer’s or seller’s own documents. Real evidence, discounted for interest.
- Class 3 · Credible third party. Established press and professional reports. Middle weight.
- Class 4 · Weak signal. Forums, social media, hearsay. Enough to open a question. Never enough to close one.
- Class M · Market-operator confirmation. Confirmation from the operator of market infrastructure, read near class 2.
For the expert reader
The frameworks under the method, named so you can check them.
- Regret theory: Bell (1982), Operations Research. The engine asks what a future you would regret, not what a present seller can sell.
- Non-additive aggregation: Choquet (1953); Grabisch (1996). Dimensions of a property decision interact; we aggregate them with a model built for interaction, not simple addition.
- Causal reasoning: Pearl’s causal hierarchy, rungs two and three. What-would-change-this is computed as intervention and counterfactual, not correlation.
- Decision-aid standards: Cochrane evidence on patient decision aids. The artifact informs a decision and refuses to make it.
- Growth-and-pruning inspiration: the metabolism of newborn brains and synaptic homeostasis (Tononi & Cirelli). Our evidence graph deliberately over-collects, then prunes against sources.
- Question ordering inspiration: active inference. In casework, the next question we pursue is the one that would reduce uncertainty most.
Where a line above says inspiration, we mean exactly that. We borrow the discipline of the idea. We do not claim its formal machinery.
What this method cannot establish
- Future prices, rents, or returns. No method owns that knowledge.
- Legal validity of title or contract. That is counsel’s work, and our file is built to hand to counsel, not to replace it.
- Intent. Evidence shows what parties did and documented, not what they privately mean to do.
- Completeness. We show what sources can carry, and we mark the rest as open. A method that claims completeness is hiding its edges.
- Whether you should buy. That belongs to you, and we have built the entire system so it stays there.