# Recording coverage and chronology

## Scope and method

All six supplied video timelines were covered by decoded images sampled at **10-second intervals**, supplemented by focused full-resolution frames for the register. All generated full-audio transcripts were read and reconciled with the supplied notes and visible screens. This is a full-timeline systematic review, **not a claim that every intervening frame was watched or every word independently listened to**. Brief interactions between samples can be missed; uncertain interpretations and approximate comment anchors are marked. No source recording was edited.

R1–R3 retain the original numbering. R4–R6 follow the phone filenames in increasing order, consistent with container creation timestamps and the visible failure→recovery→new draft→revision sequence. Container creation time is not asserted to be recording start time.

| ID | Original filename | Duration | Sampled frames reviewed | Transcript segments |
|---|---|---|---:|---:|
| R1 | Kiduna Review 1 2026-10-22.mov | 00:54:43.66 | 329 | 684 |
| R2 | Kiduna Review 2 2026-10-22.mov | 01:20:15.75 | 482 | 1137 |
| R3 | Kiduna Review 3 2026-10-22.mov | 00:59:35.38 | 358 | 532 |
| R4 | RPReplay_Final1792697258.MP4 | 00:13:16.53 | 80 | 3 |
| R5 | RPReplay_Final1792698460.MP4 | 00:19:43.38 | 119 | 170 |
| R6 | RPReplay_Final1792698974.MP4 | 00:06:11.05 | 38 | 49 |

**Total: 03:53:45.75 across 1,406 sampled frames.** All 61 contact sheets were visually reviewed. Selected source images include targeted redactions of private street addresses and account details. Unrelated phone/file-picker content and raw transcripts are not in the shareable kit. The private working folder retains original derived material for traceability.

## Audio timing and limitations

R4 audio decodes continuously from PTS zero through 735.925 seconds with **no >50 ms PTS gaps**. Its video lasts 796.533 seconds, leaving about 60.608 seconds without audio samples at the end. Minutes 1–10 have zero-valued audio samples; minute 0 contains the initial speech, and minutes 11–12 contain nonzero signal without useful recovered narration. The full transcription produced only the initial three speech segments. A separate non-VAD tail pass produced repeated “you” tokens, which are not credible decisions and were excluded. The video sequence still received full-timeline visual review. No uniform audio stretch or speculative timestamp correction was applied.

R5 and R6 audio PTS decode continuously without >50 ms gaps and track their video durations closely. Transcript anchors align with visible draft opening, role selection, typed questions and later editing discussion. Machine transcription can misrecognize names (Ki/Aashik), fonts and short phrases. The likely body font “Avenir” remains an explicit confirmation item. Recorded assistant claims are not treated as successful execution or as user decisions.

See [audio timing measurements](../evidence/phone-audio-timing.json), [original metadata](../evidence/media-metadata.json), [phone metadata](../evidence/phone-media-metadata.json) and [selected-frame manifest](../evidence/selected-manifest.json). Requested sample time and actual decoded time are separate in the manifests. Screenshot time can precede or follow the comment it illustrates; the register labels these separately.

## Chronological coverage

### R1

- **00:00–08:00:** Discovery, one creation entry, Initiative naming, tagline, park card summary, domains and meaningful banner.
- **08:00–24:00:** Duna identity, separate art, focused popup, adopted policies, ecosystem footer, separate actual balances.
- **24:00–39:00:** Backing and notification choices; final shared list/detail toggle replaces earlier leave-confirmation idea.
- **39:00–48:00:** Classification and nested cycles, actionable activity feed, Save / Explore / Join, dependency records.
- **48:00–54:43:** Step-by-step copy, evidence guidance, own versus others’ work; unrelated phone interruption at end excluded.

### R2

- **00:00–14:00:** Own evidence actions and actual review results; step terminology and statuses.
- **14:00–26:00:** Orientation brief, external guides, source geography versus local instance; activity chat controls.
- **26:00–43:00:** Blanket account warning, back/scroll context, chips, progress ending, agent picker.
- **43:00–59:00:** Envoy reply expectations, quoted text, keyboard dismissal, switcher and retained close/reopen.
- **59:00–69:00:** Place/time context, location discovery, timeout followed by recovery; Home/Current independence.
- **69:00–80:15:** Home input and summary, 9:41 ambiguity, empty nearby state, next activity; unrelated end interruption excluded.

### R3

- **00:00–10:00:** Personal Agent progress, unknown versus zero, commitments and required tools.
- **10:00–20:00:** Financial snapshot records, organizer participation, real next actions and food-quality work.
- **20:00–25:30:** Conflicting school/source evidence, returned work and acceptance affordance, authority.
- **25:30–34:00:** Creation framing and separate enterprise-funding discussion; draft stays private.
- **34:00–43:00:** Dog-waste idea, automatic Home scope, search/research updates, local assumptions and unknowns.
- **43:00–50:30:** Answers, roles, repeated validation failure, resume/retry, cost hearsay and inability to start another.
- **50:30–55:30:** Exports, tool requirements, observed Google warning and connection wording.
- **55:30–59:35:** Luminary candidates, communication channels, agreed help records and final navigation.

### R4

- **00:00–02:00:** Stuck creation, attempt to start another, asking Ki to repair the existing draft.
- **02:00–04:00:** Ki reports unresolved validation; discovery/retry and captured reference image.
- **04:00–08:00:** Attaching original reference through file picker and chat; unrelated file content excluded.
- **08:00–10:00:** Original intent retained in chat; repeated repair/fresh-start suggestions and failures.
- **10:00–13:16:** Generic preparation/checklist receipts, fresh-start text, directory detour and retry. Brief unrelated Facebook screen excluded.

### R5

- **00:00–03:00:** Previously failed draft succeeds; multiple drafts, tentative creator backing exception, Create before adequate review, typography.
- **03:00–06:00:** Collection plan, ordering, initiative-member-scoped final-say roles.
- **06:00–09:00:** Canonical activity navigation and participation state divergence between list/detail.
- **09:00–12:00:** Notifications and actual school-board review; online veterans idea incorrectly receives local Home scope.
- **12:00–16:45:** Research questions: sponsor, scale, qualified partner and tools should become initial setup activities; tentative recording/privacy discussion.
- **16:45–19:43:** Waiting through draft preparation, validation repairs, then ready draft.

### R6

- **00:00–01:30:** Recap setup questions as activities and Luminary recruitment; draft opens into tools, no obvious editing.
- **01:30–03:30:** Ask Ki to revise; assistant says same draft and explicit Create, but that is a stated capability rather than tested persistence.
- **03:30–04:40:** Section/activity context into Ki; dependency cards and loading detour.
- **04:40–06:11:** Simple summary, ordered outline, Add activity and Edit activity; retain context.

## Later-recording reconciliation

- R4 strengthens the recovery-diagnostics requirement; R5 establishes eventual success. Both remain in the record.
- R5 refines final-say selection toward initiative-scoped roles and exposes participation state divergence. Selecting the reviewer himself is a test action, not a grant of authority.
- R5/R6 reinforce global scope and setup unknowns becoming activities; sponsor/clinical/privacy decisions belong to the veterans example, not the dog-waste plan.
- R6 adds explicit summary, reorder/add/edit controls and contextual section-to-Ki handoff.
- Creator backing, font identity and recording policy remain unresolved. No clinical program or sponsor is approved by the demo.