Methodology
Last updated: 2026-08-12
Coverage scope
TickerReceipts tracks only public, U.S.-equity research videos from a hand-maintained roster of English-language YouTube channels. We do not crawl the open web, and we do not cover Chinese-language videos, podcasts, press releases, sell-side reports, or social media. A change is recorded only when it appears in an eligible English video from a roster channel. Shifts that happen outside that roster are outside our scope by design, not by omission.
What counts as a change (and what does not)
A material change is a pointable shift in an analyst's stance, grounds, or a key figure on a specific research question, measured against their own prior verifiable baseline — a raised or cut target, a bull-to-bear flip, a swapped core argument, a corrected number. The following are NOT changes: restating an existing claim in new words, pure sentiment or slogans, a bare price level or entry point with no grounds, and another retelling (occurrence) of the same claim. Non-changes are logged in the Filtering Ledger for transparency but never produce a change card.
Identity and deduplication
Every claim has a stable Claim Identity, defined by the analyst, the ticker, the research question, and the stance direction. When the same claim is repeated across videos, each repeat is only another Occurrence of that claim — never a new change. An occurrence is promoted to a Material Change only when it genuinely alters the stance, grounds, or a figure relative to the baseline. This deduplication is what stops 'repeating the same opinion' from masquerading as 'a stream of new changes.'
Cross-language order
The source evidence is always English. The order is fixed: we first fix the English original and its English Canonical Assertion, then produce a rendering for Chinese readers. Chinese is a reading layer, never a source of fact. Every cross-language output passes an entailment check — the Chinese rendering must be entailed by the English original, with numbers, direction, and conditions matching one-for-one; if the check fails, that item is not published. You can always open the English original and its timestamp to verify against the source video yourself.
Fully automated, fail-closed
The whole pipeline is fully automated — no human editor rewrites content. Each candidate change has exactly three fates: Publish, Downgrade (to a plain occurrence / ledger entry), or Suppress (not published). When evidence is thin, the cross-language check fails, or a quality gate is not met, the system chooses 'do not publish' rather than 'guess' — that is fail-closed. We would rather miss one than ship a change you cannot verify. The Filtering Ledger publishes the counts at every step (eligible → processed → published / downgraded / suppressed / failed).
Corrections and versioning
Published items are versioned. When an English source is corrected or retracted, or we find a processing error, the original URL stays the same, a correction / retraction status is shown at the top of the page, and a Revision History is kept so you can see what changed and when. Every page carries a correction link at the bottom — you can file a correction request on any change or receipt, which enters our quality-feedback process.
Boundary statement
We do not give a composite score, we do not rank analysts, and we do not tell you to buy or sell. TickerReceipts records whose research claim changed, on what grounds, and where the original is — it is a research aid, not investment advice. Nothing here is an offer to buy or sell or personalized financial advice; verify the English original yourself and own your investment decisions.