# AI editorial product POV — descriptive transcript

Silent screen demonstration of a reconstructed editorial UI. All footage is actual product capture. Original synthetic eSpeak input, not historic AI House footage or a real speaker. Captions describe actions and limitations, not speech. Loading intervals are shortened; finished screens are held for reading.

## 00:00:00.000–00:00:06.000 · Upload
Captured screen: empty-real-upload-form. Synthetic rebuild · Original 173s eSpeak audio. No event footage or real speaker voice.

## 00:00:06.000–00:00:11.000 · Upload
Captured screen: actual-uploaded-source. Actual native upload of the owned synthetic recording.

## 00:00:11.000–00:00:16.000 · Fresh transcription
Captured screen: force-fresh-verified. Force fresh selected: this run does not reuse a cached transcript.

## 00:00:16.000–00:00:19.000 · Fresh transcription
Captured screen: clicked-Transcribe. Transcribe starts the actual Whisper base run. Waiting is shortened in this edit.

## 00:00:19.000–00:00:22.000 · Fresh transcription
Captured screen: actual-transcribed. The same transcription is still running. No simulated progress or additional inference.

## 00:00:22.000–00:00:32.000 · Transcript
Captured screen: actual-fresh-transcript. Fresh Whisper base transcription: 27.05s CPU, 32 segments. Raw ASR contains errors; review is required.

## 00:00:32.000–00:00:36.000 · Moments
Captured screen: clicked-2 · Moments. Open Moments to look for possible excerpts in the real transcript.

## 00:00:36.000–00:00:39.000 · Moments
Captured screen: clicked-Find moments. Find moments starts the actual scoring call. Waiting is shortened.

## 00:00:39.000–00:00:42.000 · Moments
Captured screen: actual-ranked. Waiting for that same scoring call. The reconstructed rubric is uncalibrated.

## 00:00:42.000–00:00:54.000 · Ranking
Captured screen: actual-ranked-results. Ranked candidates with evidence and cautions. M05 is selected; its score is not publication approval.

## 00:00:54.000–00:00:57.000 · Cut
Captured screen: clicked-Create excerpt. Create excerpt cuts the selected source interval, 52.7–116.7 seconds.

## 00:00:57.000–00:01:10.000 · Excerpt
Captured screen: actual-excerpt. Actual 64-second audio cut. ASR errors and an unfinished responsibility point still need human review.

## 00:01:10.000–00:01:13.000 · LinkedIn draft
Captured screen: clicked-4 · LinkedIn. Open LinkedIn. The output field is empty before drafting.

## 00:01:13.000–00:01:16.000 · LinkedIn draft
Captured screen: clicked-Build LinkedIn draft. Build LinkedIn draft starts a separate genuine model call.

## 00:01:16.000–00:01:19.000 · LinkedIn draft
Captured screen: actual-needsHuman. Waiting for the same draft call. No publishing or automatic approval.

## 00:01:19.000–00:01:35.000 · Human review
Captured screen: actual-grounded-draft. Draft ready, review pending. It omits the unfinished responsibility point. Nothing was approved or published.

## Audio excerpt and editorial limits
The separate excerpt.wav is unchanged, 64 seconds, cut from 52.7–116.7s of the owned synthetic source. Its end enters a responsibility example before that point is complete. The resulting LinkedIn draft explicitly omits that unfinished point. Raw transcription contains ASR errors. Scoring and drafting are two genuine SDK calls; rubric scores are uncalibrated. Final state is needsHuman, no human approval and no publication.

## Actual generated draft (not approved)
Trust takes more than an explanation
A clear explanation is useful—but it isn’t the whole trust test.

When reviewing a system, ask two questions:
• Can people understand the explanation?
• Does the test set reflect situations where an error would matter?

Neither answer replaces the other. And an explanation alone can’t show whether the data represents the people who will use the system.

This is a synthetic demo from a private user-uploaded recording—not a verified AI House policy statement or historical scoring. Review pending.

## Actual draft warning
The transcript may contain ASR errors, including “one's car” and “a mean responsibility.” The selected excerpt ends during the responsibility example; this draft omits that unfinished point.