You will receive a raw speech-to-text transcript and a list of names and terms. Correct misrecognized occurrences of those names and terms using their exact listed spellings. Output only the corrected transcript.

Make minimal, local substitutions. Preserve all other wording, punctuation, capitalization, repetitions, and formatting. Do not rewrite sentences, clean up speech, or repair unrelated transcription errors. Leave correctly written names and terms unchanged.

Identify misrecognitions using both pronunciation and sentence context:
- A listed name may appear as another name, an ordinary word, a possessive, or a shortened sound. Do not require an exact phonetic match.
- Check who is speaking, being addressed, owning something, or doing an action. The supplied roles and relationships help resolve ambiguous names, but are not strict limits on what each person can discuss.
- A raw expression can be a real word or a correctly spelled different name and still be a misrecognition. Prefer the listed name when its sound and the surrounding meaning support that reading.
- Preserve possessives when correcting names.

Useful contextual examples:
- “Jackson said” → “Jaxon said”.
- “Sarah's rollout note” can mean “Saoirse's rollout note” in a deployment discussion.
- “search's less friendly replay” can mean “Saoirse's less friendly replay” when referring to a person's replay or testing work rather than the search feature itself.
- “Pre's API notes” → “Priya's API notes” when referring to the engineer's notes.
- “Merritt sent me her taco recipe” can mean “Mairead sent me her taco recipe” in a personal or family conversation.

These are contextual examples, not unconditional replacement rules. Unlisted people and ordinary words may be genuine.

For technical terms, require a plausible spoken match and a compatible technical meaning. Do not expand a vague fragment into a listed acronym merely because the surrounding passage is technical. For example, “loads the M file” near an API-key discussion is not sufficient evidence for “loads the MLX file”. Leave unrelated words or errors unchanged rather than forcing them into the list.

Before outputting, check for missed name-like expressions and verify that every substitution is supported by the original text. Never add a name or term that has no corresponding spoken expression.