YUCLAW's lane
AI research assistants and evidence protocols are different kinds of tools. Both are useful.
They answer different questions, and conflating them helps no one — so here is the difference,
stated factually. No product comparisons, no company names: categories, not competitors.
Disclaimer — Research & education only. Not investment advice. Everything on this page describes methodology, not performance.
An AI research assistant…
answers "what should I read, and what does it say?"
- Search — find filings, transcripts, and news relevant to a question
- Summarize — condense a document or a quarter into readable prose
- Monitor — watch tickers or topics and surface new items
- Converse — answer follow-up questions in context
The output is prose for a human reader, produced on demand. Its value is speed of
understanding. Verifying it is the reader's job.
An evidence protocol…
answers "what exactly was claimed, from what source, knowable when — and does it hold up?"
- Point-in-time extraction — every event carries available_as_of — what was knowable, when; replays filter on it, so nothing ingested later can leak into an earlier date
- Event classification — filings become typed events (a locked vocabulary) with magnitude, direction, and a SourceLock-verified excerpt that must locate verbatim in the filing
- Event-study validation — classified events are tested against subsequent abnormal returns and reported as measured — including where the result is adverse to the hypothesis
- Public replay — one command re-derives every published statistic from published derived data on anyone's machine — no account, no API key
- Ledger anchoring — each day's signal content-hashes roll into a daily root committed to a public git repository before pages update; edits after the fact are detectable
- Methodology boundaries — scoring universe and evidence tier are separated by positive gating with a standing negative check; coverage percentages are stated as measured
- Adverse-result disclosure — underpowered windows, outages, and results that go against the hypothesis are disclosed and never deleted — presentation may be compressed, substance may not
The one-sentence version
An assistant helps you read the evidence; a protocol makes the evidence —
its extraction, its classification, its validation, and its record —
independently checkable. YUCLAW is built as the second thing. Where an
answer needs prose, the prose cites events; where it needs a number, the number replays.
What that costs: an evidence protocol is narrower. It covers a measured universe (stated
per page), speaks a locked vocabulary of research classifications, and refuses questions
its evidence cannot answer — you will see "outside current evidence scope" and
"not statistically proven" on these pages, by design.
Status — proven · not proven · accruing
Rendered from one shared source (v3/web/useful_blocks.py) on every page that shows it,
so the copies cannot drift. Statuses are measured, not aspirational.
Proven (verifiable today)
- ✓ Replay works — one command reproduces every Lab statistic and ledger root from published data
- ✓ Ledger anchored daily — sha-256 daily roots committed to a public git repository before pages update
- ✓ Evidence traces to filings — every accepted event carries a source URL, accession number, and verified excerpt
- ✓ Coverage measured — SEC-filer weight per lens is stated as measured, never rounded up
- ✓ Snapshots are point-in-time — daily as-of writes, zero retroactive edits (outage window disclosed, not repaired)
- ✓ Evidence-tier names are never scored — enforced by positive gating and a standing negative check
Not proven
- ✗ Forward alpha — no spread, IC, or alpha significant at 5% with adequate power
- ✗ C6 risk-gate sign — rareness confirmed OOS 2026-07-06 (22% fire rate, n=9 held-out); sign confirmation pending (elevated arm n=2; accrual live from 2026-07-16)
- ✗ Peer-model CAR lead — event-study lead over peer models is not established; live-era sample remains small
Accruing
- · Forward out-of-sample record — one period per trading day, accruing daily
- · Matured CAR events — each accepted event matures into the event study after its forward window completes
- · C6 elevated arm — live Form-4 ingestion since 2026-07-16 restores the insider stream to production inputs
- · External replications — the replication log accrues as independent runs are reported