A ticker is not an earnings transcript’s identity

A transcript can be accurate text and still be the wrong row. Provider histories are commonly requested by today’s ticker, but the company using that ticker today may have acquired, replaced, or merely renamed the issuer that held an older call.

That makes transcript quality a relational problem. The text, issuer, security tenure, call date, and provider slot all have to agree.

1. Understand acquisition aliases

StockQL found old Xilinx calls served under AMD and old Enable Midstream calls served under Energy Transfer. The calls themselves were real. The attribution was not: an acquisition had been represented as ticker continuity.

A full audit examined 174,074 stored transcript rows and removed 686 proven cross-issuer aliases across 100 groups. The important word is proven. A company name appearing in the text is not enough; analysts routinely mention peers, customers, and acquisition targets.

The reliable signal was issuer identity. Where an exact call title or self-identification pointed to another SEC registrant, StockQL compared the expected and claimed CIKs. A different CIK across an acquisition boundary is a mismatch. A date-bounded rename that retains the same CIK can be valid continuity.

2. Treat fiscal labels as attributes, not keys

Year and quarter look like natural transcript identity fields, but providers label them differently. One feed may label a January-fiscal-year retailer by the fiscal year in which the quarter belongs; another may use a calendar-forward convention. The same call can therefore be 2026 Q1 in one catalog and a different year-quarter pair in another.

Call dates are stronger, but even they can differ by one day when publication timestamps cross a timezone boundary. Transcript text can also share an upstream: in StockQL’s comparison, some smaller-company calls were byte-identical across providers while larger-company calls came from distinct transcription passes.

Agreement on ticker, quarter, or bytes is therefore useful evidence, not a complete identity.

3. Build a durable transcript identity

A robust record keeps several layers:

• Issuer: a stable company identifier such as CIK, with date-bounded rename continuity.
• Security tenure: the ticker and exchange that were valid for the issuer at the call date.
• Event time: the normalized call date and original provider timestamp when available.
• Provider locator: the symbol, fiscal year, and quarter used to request the source row.
• Content provenance: provider, fetch time, and a digest of the exact text.

StockQL uses call date rather than year-quarter alone when deciding whether a provider slot is new. Before persistence, its FMP client rejects a call that can be proven to self-identify as another SEC issuer. Ambiguous cases do not become automatic cross-company joins.

4. Compose coverage with lineage checks

FMP materially improved the transcript catalog available to StockQL: in the measured universe it covered 4,186 of 4,195 companies, was fresher for 65%, and deeper for 88% compared with the previous feed. That breadth is valuable precisely because it passes through an issuer-lineage guard before entering research tables.

A second source can help match dates, compare text, and recover missing calls. It cannot replace the identity rule when both feeds share an upstream or repeat the same acquisition alias. StockQL composes the richer catalog with company profiles, CIK continuity, call dates, and content digests so coverage does not outrank attribution.

Return to the first article: when is a stock really delisted? →