RESEARCH

Claims are earned,
not announced.

Every claim links to its evidence, its limits, and a reproducible record.

THE METHOD
01

Ask before measuring.

02

Publish what failed.

03

Name what the evidence cannot establish.

01 / THE EVIDENCE BASE

One question
at a time.

Each artifact answers a bounded question. Failures remain visible; unlike measurements never become one score.

FORMAL MODEL

The Minority Prophet Property

Machine-checked proofs state when repetition cannot change a root-based verdict—and the assumptions that make the result hold.

Read the formal model →
RECORDS

Evidence-aligned history

Frozen records keep successes, failures, and unfinished findings visible together.

Inspect the record →
CONTROLLED STUDY

Epistemic lift

The same models faced the same 32 worlds with claims alone, provenance, and provenance plus Minority Prophet.

See the measured lift →
METHOD TEST

Capability comparison

Model reasoning, tool use, voting, and a deterministic evidence rule meet the same bounded packet.

Compare the methods →
KNOWLEDGE LEDGER

Keep the doubt

Receipts carry evidence roots, coverage, uncertainty, flip budget, and the conditions that would reverse a conclusion.

Explore the ledger →
LINEAGE

Where inference breaks

The lineage series shows what can be reconstructed—and what disappears when identity and provenance are missing.

Read the lineage series →

02 / NEGATIVE EVIDENCE

What failed
stays visible.

Resolved markets

Dependence adjustment did not beat market price across 5,729 eligible weather markets.

Lineage without identity

Text and time alone did not recover reply roots when identity was missing.

Complexity without gain

A more complex dependence auditor lost to simpler comparators and was rejected.

03 / READING PATHS

Start with the question
you need answered.

PAPERRead the scientific claim→CLAIMSSee what is supported→EVIDENCE MAPTrace claim to record→METHODSee how findings graduate→