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

Rules:
- Make only name/term corrections. Preserve all other wording, punctuation, capitalization, repetitions, fillers, and formatting.
- Leave already-correct names and terms unchanged.
- Use pronunciation, grammar, and surrounding context together. A listed name may be transcribed as another name, an ordinary word, or a shortened fragment. Do not require an exact phonetic match.
- Check possessives carefully: the recognition error may be in the name before “'s.” Preserve the possessive when correcting it.
- Use the descriptions in the list to resolve ambiguity. Roles are clues, not requirements: people can discuss topics outside their roles.
- Do not force ordinary words, initials, or unrelated names into the vocabulary merely because they resemble a listed entry. The list is a correction aid, not a requirement to insert terms.

Important contextual corrections:
- “Jackson” referring to a person → “Jaxon.”
- “Sarah's rollout note” → “Saoirse's rollout note,” using the deployment/ops context.
- “search's less friendly replay” → “Saoirse's less friendly replay,” where the possessive refers to the person responsible for the replay/test run. Do not replace ordinary uses of “search.”
- “Pre's API notes” → “Priya's API notes,” using the person-possessive construction and API/partner context. Do not replace ordinary uses of “pre.”
- “Merritt” referring to the person sharing her taco recipe → “Mairead.” A plausible-looking alternative name can still be a transcription error.

Avoid unsupported technical insertions. For example, “loads the M file” in a discussion of shell execution, environment configuration, and an API key is not evidence for “MLX file”; leave that phrase unchanged rather than inserting MLX.

Before outputting, check for missed names disguised as ordinary words or possessives, then verify that every replacement is supported by both the transcript and the supplied vocabulary. If no correction is needed, return the transcript unchanged.