A single history entry may contain more than the text you pasted. In SuprFlow 1.0.0, a history entry stores the raw speech-to-text result, an optional cleaned version, and an optional formatted version. It also stores a timestamp and model or formatting details. The display chooses the most processed available version, but the underlying entry can retain earlier wording. That matters if a cleanup step removed a filler word, altered a negation, or rewrote a client-sensitive phrase.
The app keeps the 100 most recent entries in a JSON file under ~/Library/Application Support/SuprFlow/. The cap is an entry count, not a 100-word, 100-day, or 100-kilobyte retention promise. Older entries drop as new ones are inserted in the current history, while backups or copied text may retain other versions elsewhere.
Why the three fields are useful — and sensitive
| Version | What it represents | Review question |
|---|---|---|
| Raw | The speech model's text before later cleanup or formatting | Did it capture the correct amount, name, date, and negation? |
| Cleaned | A later rule-based cleanup result, when recorded separately | Did the cleanup remove a word that changed meaning? |
| Formatted | Output of an optional formatting step, if it produced a distinct result | Did a local or hosted formatter rewrite the substantive statement? |
The fields are optional, so a particular entry may not show all three. The displayed text falls back from formatted to cleaned to raw. The history screen can expose earlier text for comparison; the newest displayed version should not be treated as the only local copy.
Consider a fictional adviser note: raw text says “Client did not authorize a sale of 50 shares”; formatted text says “Client authorized a sale of 50 shares.” The polished version is easier to read but wrong. A review that compares the versions catches the error before it reaches a supervised CRM. For an auditor, the same risk applies if “two exceptions” becomes “no exceptions.” The examples are illustrative, not measured SuprFlow failure rates.
What a firm should decide
First, decide whether a local drafting history containing raw and edited text is acceptable on managed devices. A final CRM or workpaper record may require retention even when temporary drafting text should be limited. Next, decide who may open the Mac account, what device backups capture, and whether support staff may ever receive a history export. Do not paste actual entries into a bug report; reproduce with fictional data.
The current history UI offers single-row deletion and Clear All, but in SuprFlow 1.0.0 an individual deletion may not persist after the app restarts, and a pending save can race with Clear All. Neither a disappearing row nor a quick empty screen establishes reliable disk deletion; use Clear All and check again after relaunching. See the deletion guide before writing a retention promise.
A synthetic review exercise
Dictate: “The fictional Mariner account did not approve the twelve percent adjustment.” In the history screen, compare whatever raw, cleaned, and formatted versions appear, then inspect the actual text inserted into a local editor. Do not infer a formatter's behavior from the history label alone. Record the SuprFlow version, selected speech model, formatter, and endpoint so the result can be reproduced after an update.
If any version changes “did not,” “twelve percent,” or the account name, correct the destination record before treating it as final. If a real client note was already saved incorrectly, follow the organization's record-correction procedure rather than silently replacing the draft.
Questions people ask
Does the 100-entry limit erase every older transcript?
It limits SuprFlow's current history list. It does not prove deletion from Time Machine, other backups, clipboard managers, destination apps, or a hosted formatter's systems.
If I only see polished text, is the raw text gone?
Not necessarily. An entry can retain raw text even when its displayed text prefers a formatted version. Inspect the entry and release-specific behavior before making a retention claim.
Are these history entries audio recordings?
A history entry stores text and metadata, not the original audio. The normal recording pipeline removes its temporary audio file after transcription, but error paths and the signed build still need verification.
Related guides
- Is Voice to text mode really a raw transcript?
- How to delete SuprFlow transcription history
- Can Time Machine keep a cleared transcript?
- Why local history is not the firm's official record
- What if AI formatting changes an audit exception?
- Where does my dictation go?