A reporting framework shows how return timing shifts financial NLP labels

When a financial language benchmark uses a full earnings-call transcript or recording to predict stock movement, what does its result mean if the return window begins before the call ends?

It may capture price movement during the disclosure, rather than prediction after the complete input is available.

The Availability Contract records input boundaries, timing evidence and return clocks so financial benchmark claims match what those clocks support.

The authors apply the contract to earnings calls by comparing returns with a shared endpoint but different starts: a documentary call-start proxy and an audio-completion proxy with a 15-minute allowance. Among 552 paired calls, moving to the completion proxy changed the return sign for 9.4% of calls.

Because the same calls and endpoint are compared, the result shows that the return target depends on the chosen start time. A completion proxy makes the timing assumption explicit, but it does not establish when the full input or model was actually ready.

The audit uses retrospective timing proxies. It does not observe when transcripts or model inference were complete, so it supports conclusions aligned with those proxies, not verified post-input prediction or deployment claims.