BearSignal is forensic short-selling research for institutional investors — built to read the entire U.S. and Hong Kong listed universe and surface the companies whose own numbers cannot all be true.
Every red flag on this platform is detected and issued by Ursa, BearSignal's reasoning engine — trained on one corpus only: signed, blind-written forensic annotations by independent practising accountants, cited line by line to the filings. The flags are ours. The annotations are theirs.
Ursa reads the filings and issues the flag. No analyst opinion, no committee — a reasoning engine, and the record of how it learned.
The engine learns forensic judgment from annotations: signed, blind-written reasoning chains on historical filings, authored by credentialed accountants worldwide. Annotators label the past; they issue no opinion on any company.
When annotators split, the dispute and its adjudication go into the training record. A judge trained only on sanitized agreement cannot be trusted.
Every annotation and every flag is timestamped and hash-anchored when written. Revisions are versioned; nothing is silently rewritten.
The reasoning is the record.
Teach the machine. Sign the record.
BearSignal invites credentialed practising accountants worldwide — US GAAP and HKFRS/IFRS tracks — to annotate historical filings that train Ursa. Blind, signed, per-case, remote. Annotation experts never issue opinions on live companies; your judgment sets the ceiling of what the machine can learn.
Ursa is trained on exactly one thing: signed reasoning chains written by practising accountants. Nothing machine-generated ever enters the corpus — the training gate rejects any sample whose author is not a credentialed forensic expert, enforced in the database, not by convention. The signed record is not marketing around the model. It is the model's entire diet.
Ursa is the reasoning engine BearSignal is training to judge financial contradictions the way practising forensic accountants do. Today, the detection engine reads the entire U.S. and Hong Kong universe continuously and publishes gray flags — documented hard contradictions a human analyst would never have time to find, certified against six structural checks, issued as records, not verdicts. What Ursa learns from is the training: signed annotations of historical filings by credentialed accountants worldwide. The methodology is documented openly at BearSignal.ai.
Continuously reads every U.S. and Hong Kong listed company — SEC EDGAR, HKEX filings, ownership records. No watchlist, no sampling.
Raises gray flags where a company's own numbers cannot all be true — documented hard contradictions, certified against six structural checks, not verdicts.
Every gray flag is detected, certified against six structural checks, and published by the engine. What humans provide is the training — signed annotations of historical filings that set the ceiling of what the machine can judge.
A selection of active gray-flag files issued by the forensic engine — each certified against six structural checks before publication. The full — forensic files are open to beta members only — seats are limited. Request a beta invite code → Already have a code? Sign in. New here? See our methodology →
The engine reads at scale and publishes what clears certification. Every claim is sourced, every contradiction documented, every conclusion auditable — gray flags are issued by the engine and certified against six structural checks before publication.
Continuously reads the entire U.S. and Hong Kong listed universe — SEC EDGAR, HKEX filings, ownership records — surfacing companies whose own numbers cannot all be true. No watchlist, no sampling.
A martingale e-process tests each company for internal financial contradictions — revenue that outruns cash, inventory that defies sales, cash that earns no interest. Gray flags are raised on absolute statistical thresholds, never on rank or opinion.
Every flagged contradiction is anchored to source filings, cross-referenced, and timestamped — investigation files built to withstand regulatory and legal scrutiny.
Beneficial-ownership tracing through VIE structures, offshore vehicles, and related-party networks across ADRs, H-shares, and red chips — mainland subsidiary penetration from corporate registries.
BearSignal covers NYSE, NASDAQ, and HKEX — ADRs, H-shares, red chips, and VIEs across the United States and Greater China. Every name in the listed universe is in scope; the work is in finding the ones whose numbers do not hold together.
New Horizon Health (HKEX: 06606) was the first case that prototype was ever run on. The methodology behind BearSignal was proven in the field before the company was incorporated.
Warned first. Vindicated by events. — the case that became BearSignal's blueprint.