evidence / Engineering depth
The Hypothesis-to-Verdict Knowledge Engine -- folklore in, pre-registered verdicts out
Sports folklore goes in; leak-free, pre-registered verdicts come out; and the nulls are published as first-class results, not buried. The single truth-source for any figure below is docs/JOB_EVIDENCE_PACKET.md section G. Where a later working-tree recount reads higher than the packet's committed count, both are shown and labeled -- the packet stays the truth-source. No dollar/edge/ROI is claimed anywhere on this page; a NULL is market-efficiency evidence, and an honest REJECT is a success.
strongest single receipt
the claim
"Teams on a back-to-back shoot worse." "Momentum is real." "Home dogs cover." Sports is drowning in folklore, and almost none of it is tested against leak-free data. This system turns each belief into a pre-registered hypothesis, runs it through the same refutation gate every signal faces, and records the verdict -- CONFIRMED, NULL, REJECT, or NOT_TESTABLE -- in an append-only ledger, one per sport. The engine has been drained across all four sports: no open, untested beliefs are left in a backlog. Roughly half of the testable folklore survives locally; the rest is null or not-testable, and every null is kept and published rather than deleted. That honest shape is the point.
receipts
| claim | committed artifact |
|---|---|
| Seed: literature-to-verdict loop | commits e500a0d6, 1bc7d622 (+ closer chain); rows in the sport ledgers |
| Gate: leak-free ship gate | src/loop/gate.py |
| Gate at scale: interaction factory | data/cache/intel_claims/interaction_factory_ledger.jsonl |
| Ledger: append-only verdicts, 4 sports | domains/{basketball_nba,mlb,soccer,tennis}/knowledge/validation_ledger.jsonl |
| Ledger integrity: dedupe audit + repair | commit 78d503ee (writeup local-only under docs/research/) |
| Resolver receipt: anti-folklore client | scripts/platformkit/answers/contract_client.py, resolver_registry.py |
| Effect graph | scripts/platformkit/answers/effect_graph.py |
| Honest-tally exhibit | scripts/platformkit/analytics_showcase/out/honesty_exhibit.json |
| Survival rollup | scripts/platformkit/analytics_showcase/out/mechanism_survival.json |
why this matters
Anyone can assert that back-to-backs hurt shooting. The hire signal is a machine that turns each assertion into a pre-registered, leak-free test, keeps the negative results as first-class output, audits its own ledger for duplicate pollution, and answers only by reproducing a recorded verdict -- never by improvising a plausible-sounding belief. Folklore goes in; a verdict with a sample size, a p-value, and a source path comes out; the honest majority that does not survive is published, not hidden. That discipline -- not any single confirmed effect -- is the product.
reproduce
# Reproduce one verdict deterministically from the committed ledger
python -m scripts.platformkit.answers.contract_client "does b2b_rest_penalty hold up" --sport nba
# Regenerate the honest tallies from the ledgers
python -m scripts.platformkit.analytics_showcase.honesty_exhibit
python -m scripts.platformkit.analytics_showcase.mechanism_survival
# Recount raw verdict splits directly
python -c "import json,collections,glob; c=collections.Counter(json.loads(l)['verdict'] for f in glob.glob('domains/*/knowledge/validation_ledger.jsonl') for l in open(f,encoding='utf-8') if l.strip()); print(sum(c.values()),'rows',dict(c))"This site reports calibration and sharpness only, never a dollar edge, ROI, or bankroll result. An honest null is a success.