Correct misrecognized names and terms in the raw speech-to-text transcript using the supplied names-and-terms list as the canonical spelling reference. Output only the corrected transcript.

## Scope and preservation

- Change only the text needed to correct a listed name or term. Preserve all other wording, sentence structure, punctuation, capitalization, hyphenation, dates, and numbers.
- Use the exact canonical spelling and capitalization for corrected names and terms. Keep already-correct names and terms unchanged.
- Do not fix grammar, polish prose, remove filler, or correct unrelated ASR errors.
- Correct every supported occurrence, including the first word of the transcript, direct addresses, possessives, and names in lists or comparisons.
- Treat each occurrence independently. Finding some corrections does not mean the transcript is finished.
- The supplied list is a spelling reference, not a requirement to mention its entries. Descriptions help identify a spoken name; they do not supply names missing from the transcript.
- Every correction must replace an existing plausible rendering of that name or term. Do not insert names or terms based only on the topic.
- Perform the checks below silently. Output no explanation or change list.

## Required first pass: explicit name spellings

Before considering more ambiguous matches, scan the entire transcript for these known name errors. Apply each mapping when the text refers to a person. An explicit role mention is not required.

- **Oifi → Aoife.** This is a known spelling error, not a match that needs additional pronunciation or role evidence.
  - “Oifi got her transcription script working” → “Aoife got her transcription script working”
  - Do not overlook “Oifi” at the beginning of the transcript.
- **Jackson → Jaxon**, including direct address and discussion of spelling:
  - “Jackson, by Tuesday…” → “Jaxon, by Tuesday…”
  - “Jackson keeps getting spelled…” → “Jaxon keeps getting spelled…”
- **Caitlin, Kaitlin, or Caitlyn → Kaitlynn**, including sentence-initial direct address:
  - “Caitlin, by Tuesday…” → “Kaitlynn, by Tuesday…”
- **Kave → Kaveh.**

Apply these mappings to possessive forms as well, subject to the name-ending rules below.

These corrections do not require the sentence to identify the person's listed role. For example, Aoife can be described as writing a script without being called an intern.

## Other name errors

Use pronunciation, spelling similarity, sentence structure, nearby references, and the supplied descriptions together. A familiar name or ordinary word can still be a misrecognition of a listed name. Do not leave a clear match unchanged merely because the raw spelling is itself valid.

- **Aoife** is pronounced approximately “EE-fa”; distorted renderings of that sound can also be name errors. The explicit spelling “Oifi” is covered by the first-pass rule whenever it is used as a person's name.
- **Sasha → Saoirse** when it is a plausible rendering of the listed deployment/ops person's name:
  - “Sasha already warned us not to describe an owned machine as free” → “Saoirse already warned us not to describe an owned machine as free”
- **Sarah or Sarah's** can represent **Saoirse**, especially when the transcript explicitly discusses names being transcribed or spelled incorrectly. Use the sentence's grammatical role to choose the ending.
- **Source** used as a person's name can represent **Saoirse**:
  - “Source says the Coyab bill…” followed by “I asked her…” → “Saoirse says the Koyeb bill…”
  - Leave ordinary uses such as “the source of the problem” unchanged.
- **Ted → Tadhg** when referring to the friend in a personal or family outing context:
  - “Ted says the beach by his place…” → “Tadhg says the beach by his place…”

Apply the same sound-and-context reasoning to the other supplied names, including Xochitl, Priya, and Mairead. Do not restrict correction to the examples above.

A topic or role match alone is insufficient. There must also be a plausible spoken-name rendering in the raw text. Conversely, an explicit role mention is not required when the source rendering and sentence context already support the name.

## Possessives and name endings

Preserve genuine possessives: a misrecognized possessive becomes the canonical name plus the appropriate possessive ending.

- “Jackson's script” → “Jaxon's script”
- “design and AFAS evaluation work” → “design and Aoife's evaluation work”

However, an apparent possessive ending can itself be part of the misrecognition. Do not retain it when the sentence clearly requires a nonpossessive name.

- “Jackson keeps getting spelled some other way, and so does Sarah's.” → “Jaxon keeps getting spelled some other way, and so does Saoirse.”

In that example, “Saoirse” is the subject of “does,” not a possessive. Correcting that name ending is not permission to fix unrelated grammar.

## Multiple corrections in one transcript

Correct all supported occurrences, even when they involve different names or are far apart. Do not stop after correcting the first name or term.

Raw:
Oifi got her transcription script working, so now it can run overnight and transcribe every new recording. The problem is the names are still wrong. Jackson keeps getting spelled some other way, and so does Sarah's. Tomorrow I'll have her try adding a name list and see if that helps.

Corrected:
Aoife got her transcription script working, so now it can run overnight and transcribe every new recording. The problem is the names are still wrong. Jaxon keeps getting spelled some other way, and so does Saoirse. Tomorrow I'll have her try adding a name list and see if that helps.

The three independent corrections are “Oifi” → “Aoife,” “Jackson” → “Jaxon,” and “Sarah's” → “Saoirse.” All other text remains unchanged.

## Term corrections

Use the exact canonical spelling and casing of listed terms when correcting phonetic renderings, incorrect spacing, or joined words.

- “Coyab” and similar phonetic renderings of the hosting service → “Koyeb.”
- “voice flow” referring to the listed product → “VoiceFlow.”
- Apply the same reasoning to vLLM, KV cache, LoRA, MLX, AI Builder Space, Superlinear, intake, Codex, and prefill.

A correction may replace a misrecognized multiword span with one canonical term, or a joined rendering with a canonical multiword term. Change only the span representing the spoken term.

Do not convert an ordinary word or phrase into a product or technical term unless both its sound or spelling and its usage support that correction. Do not replace a related concept with a listed term merely because the term would fit the topic.

## Avoid unsupported substitutions

- An investor discussion does not itself imply “Kaveh.”
- Deployment discussion does not itself imply “Saoirse.”
- “assured investor check-in” does not contain a plausible rendering of “Kaveh's”; leave it unchanged.
- Do not add names, expand pronouns or descriptions into names, or replace unrelated people merely to make the transcript fit the supplied list.
- Do not insert a term because it is relevant to the discussion.
- Every replacement must be traceable to an existing plausible spoken name or term in the raw transcript.
- If a proposed correction has only topical support and no plausible source rendering, leave that text unchanged.
- Do not let an uncertain candidate prevent you from applying other clearly supported corrections.

## Final completeness and preservation check

Before outputting, scan the entire transcript again, from the first word to the last:

1. Check the first word and every apparent personal name. If “Oifi” is used as a person's name anywhere, it must now be “Aoife.”
2. Check every occurrence of the explicit variants: Jackson, Caitlin, Kaitlin, Caitlyn, and Kave, including possessive forms.
3. Check other plausible name renderings, including Sasha, Sarah, Sarah's, Source, and Ted, using their sentence context.
4. Check all listed-term sound-alikes and split or joined forms.
5. Verify that genuine possessives are preserved and erroneous name endings are corrected only where supported.
6. Verify that every change has a plausible source in the raw transcript. Remove any unsupported substitution or insertion.
7. Verify that all unrelated text remains unchanged. Do not add punctuation or make incidental edits while correcting a name or term.

Output only the corrected transcript, without an introduction, explanation, quotation marks, or change list. If no supported correction exists, output the transcript unchanged.