Play the architecture's history between two commits on the web map

[MrLesk]

Implements TASK-563. The live revision control gets Play and Stop between a start and an end commit, with speeds of 0.5x, 1x, 2x and 4x. The range stays in the URL and the header, and a caption names each commit while it plays. The web session caches architecture and source snapshots, plays only readable revisions in Git order, and preloads the next two comparisons. At the end the map shows the full-range comparison. The static export and the terminal viewer are unchanged. docs/viewers/web/history-playback.md describes it.

A Codex agent (GPT-6.1 Sol) built this live during the Devoxx Belgium keynote on 9 October, working TASK-563 in one checkout it shared with five other agents. This branch holds only this task's changes: I matched every changed line to the agent session that wrote it and rebuilt the task on a fresh branch from main.

Acceptance criteria, as the agent left them

• The revision control lets a person choose a start and an end commit among the readable revisions, and Play walks from start to end one revision at a time, with Stop and a speed choice. The URL carries the range so a link reopens it.
• Each step morphs the map from the previous revision's layout to the next with retained positions, the way the comparison keeps positions, and the step's short id, subject and date show while it plays.
• Playback stays smooth: a revision is read once per session and cached, upcoming revisions load ahead, and a range of fifty commits plays without reloading the page or blocking the map.
• When playback reaches the end, the map shows the comparison from the start to the end commit with its changes bar and stepper.
• A reduced-motion preference steps without animation. The static export and the terminal viewer are unchanged.
• Playback visits only compatible revisions, in commit order, from the chosen start commit to the chosen end commit.

The agent ticked none of them on purpose. It type-checked and linted its work, but had no browser in its session to see playback run, and would not tick criteria without that evidence.

Validation

On stage the agents were told not to install, build or run tests, so these are the first runs of this code.

• bun run typecheck passes.
• biome lint: no errors, no new warnings.
• Node tests: 16 pass. Bun suite: 776 pass, 51 skip, 0 fail, the same as main.
• Not checked yet: playback in a browser, including a fifty-commit range.

Overlap with the other PRs from this run

map-session.ts, data.ts and render.ts are shared with other PRs from the same run. Each adds separate lines.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-09T09:19:00.951010Z	190730d	PR opened
🔒 Security Review	✅ Completed 2026-10-09T09:19:17.927185Z	190730d	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Groma architecture comparison

Changes	Added	Modified	Removed
Components	0	4	0
Relationships	0	0	0

Open before / after comparison

Compared 7b9f3c0 (merge base) → 190730d.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 190730d65c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Start playback before loading every snapshot

For a long range, /playback.json does not return until this serial loop has extracted, parsed, sourced, and laid out every commit via snapshot(). Consequently a fifty-commit playback remains on “Loading history…” until all fifty expensive reads finish, instead of starting from the first frame while only upcoming frames preload, which fails TASK-563 acceptance criterion #3. Determine compatibility without eagerly materializing every full map, or load frames incrementally/bounded in parallel.

AGENTS.md reference: AGENTS.md:L265-L267

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve the selected range when opening an endpoint

When playback is stopped on an intermediate frame—or while it is running—the header still displays the original start/end range, but clicking either endpoint calls cancel(), whose default clears that range. The editor then derives its fields from current, which is only the current step comparison, so merely opening a displayed endpoint silently replaces the user's chosen range with the adjacent step pair and makes the original endpoints impossible to edit as shown. Stop playback without clearing the range before opening the field, or restore the full-range comparison first.

AGENTS.md reference: AGENTS.md:L265-L267

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep the task open until its criteria are verified

The task is moved to Done while every acceptance criterion and Definition of Done items #1 and #2 remain unchecked, and its own notes state that browser playback and the required fifty-commit behavior are unverified. This makes Backlog report completion despite missing objective evidence and confirmed functional findings in the supported flow; return it to a non-terminal status until those criteria are satisfied.

AGENTS.md reference: AGENTS.md:L281-L282

Useful? React with 👍 / 👎.