# Reported issues for aidemo

Pod holds 5 of 5 GitHub reports that passed its relevance review. This can include external user reports, maintainer-confirmed bugs, and concrete feature gaps. Treat them as evidence to inspect, not a count of distinct defects.

Back to [aidemo](/mcp/aidemo).

## Most discussed

### record/compose: anchor-unreachable, overrun-trim and scene-freeze are printed but never fail the run — add --strict / non-zero exit

**Seen in** `demos/loyalty-prod` v6 (0.14.1): `report.json` carried three warnings — `overrun-trim` (s1, 1445 ms cut), `anchor-unreachable` (s4, the read reward), `scene-freeze` (s5c, 53 % held) — and each mapped 1:1 to a defect the owner then found by eye (the trimmed hook, the reward shown < 1 s, the frozen confirm card). `compose` exited 0, the wrapper script went on to the EN dub, and the warnings scrolled away.

**Ask**
1. `compose --strict` (or `--fail-on warning`) → exit 1 when…

[Read the thread](https://github.com/tandryukha/aidemo/issues/59) · 2026-09-10 · open · 0 comments

### hold.drift makes the page visibly bob/creep (±6 px) on idle spans — reads as a jitter bug, not a dwell

**Seen in** `demos/loyalty-prod` v6 (0.14.1) with the house preset `hold: { mode: "drift", driftScale: 1.05, backoffMs: 400 }`, `frame.chrome: browser`. Measured on the MP4 (window-chrome edge vs page-content edge tracked per frame at 15 fps): the chrome stays at y=50 while the page content moves 180→186→180 px in 0.3 s bursts during s1's auto-idle span (~5.7–7.9 s) and creeps 373→367 px over 3 s in s6. The reviewer's words: "the page jumps up and down".

Why it reads wrong: the drift zoom is…

[Read the thread](https://github.com/tandryukha/aidemo/issues/57) · 2026-09-10 · open · 0 comments

### highlight/focus/callout: rect measured before layout settles — outline lands on the wrong element after lazy content loads above the target

**Seen in** `demos/loyalty-prod` v6 (aidemo 0.14.1, Playwright capture, 1440×900). Scene s4 does `waitFor [data-turg-read]` → `focus` → `waitForChange` label → `highlight [data-turg-read] 1600`. A lazy product carousel (turg-embed, fetched after DOMContentLoaded) inserts ~400 px of content **above** the end card between the measure and the draw. Result in the MP4: the gold outline is drawn over the product tins where the card *was*, the card itself is off screen, and `report.json` only says…

[Read the thread](https://github.com/tandryukha/aidemo/issues/56) · 2026-09-10 · open · 0 comments

### Multi-clip resume: reused scenes overrun into the next navigation; autoIdle trims a still confirm card; waitFor matches a hidden table

**What happened / suggestion**

Three things found while cutting `demos/loyalty-prod` v6 (aidemo 0.14.1 + `5ff3e74`), a take resumed twice with `--from-scene X --no-replay` so the final compose splices two `raw.keep-*` clips plus `raw.webm`. QA (`lint` + per-scene hold %) did not flag any of them; a 1 fps contact sheet did.

### 1. Reused scenes overrun into the next scene's navigation (~0.5–0.9 s)

After a resume, each reused scene's window in `timeline.json` (`startMs`/`endMs` + the clip's…

[Read the thread](https://github.com/tandryukha/aidemo/issues/55) · 2026-09-10 · open · 0 comments

### No way to accept a native dialog (window.confirm) — a filmed click silently does nothing

**What happened / suggestion**

## What I hit

A storyboard needs to film a destructive-looking control whose handler opens a native
`window.confirm()` (an admin "Archive this widget?" button). There is no op that can
accept or dismiss a browser dialog, and Playwright auto-dismisses dialogs by default —
so a filmed `click` on that button *looks* like it worked and nothing happens. That is
worse than not filming it: it puts a **false outcome on camera**.

I could not find a workaround.…

[Read the thread](https://github.com/tandryukha/aidemo/issues/54) · 2026-09-09 · open · 0 comments

## Most recent

The remaining reports are on [the project's issue tracker](https://github.com/tandryukha/aidemo/issues).