# Tangled stats

Crawls every pull request on Tangled and renders the leaderboards as a
self-contained HTML block for a Leaflet post.

## Embedding

`embed.html` is the whole panel as one fragment: a `<style>` block plus one
`<section class="tg-stats">`. Paste it into the post.

`blocks/` holds the same content cut into 14 fragments, each independently
pasteable so you can place them one at a time between paragraphs:

| File | Block |
|---|---|
| `01-headline.html` | hero figures + KPI strip |
| `02-merged-alltime.html` | merged-PR leaderboard, all time |
| `03-merged-august.html` | merged-PR leaderboard, August 2026 |
| `04-commits.html` | commit leaderboard, all time |
| `05-commits-august.html` | commit leaderboard, August 2026 |
| `06-pies.html` | all four pies together |
| `07-pie-merged-alltime.html` | one pie: merged PRs, all time |
| `08-pie-merged-august.html` | one pie: merged PRs, August |
| `09-pie-commits-alltime.html` | one pie: commits, all time |
| `10-pie-commits-august.html` | one pie: commits, August |
| `11-timeline.html` | merged PRs per month, network-wide |
| `12-august-daily.html` | my August, day by day |
| `13-comparison.html` | me versus everyone else |
| `14-method.html` | method + the time-to-merge caveat |

Each block carries its own copy of the stylesheet so it works alone. The rules are
identical and scoped to `.tg-stats`, so repeated blocks in one post are harmless —
but if you paste several, you can delete the `<style>` from all but the first.

`blocks/index.html` is a contact sheet showing every block rendered on black; open
it locally to pick the ones you want. It carries no JavaScript, so it survives HTML sanitizing;
hover tooltips come from SVG `<title>`, and the leaderboard bar charts have a table view
behind a `<details>`.

Colors are pinned to the permadeath.com Leaflet theme — `--bg-page: 0,0,0` and
`--primary: 255,0,0` — and the type inherits Sometype Mono, which the blog already
loads, so the block needs no extra network request. The panel commits to that one
theme: no light mode, every color painted explicitly, nothing inherited from the host.

## Re-running

    ./run_all.sh          # waits for api.tangled.org, then runs everything

Stages, if you need them individually:

| Script | Does |
|---|---|
| `crawl_pulls.py` | every PR for each repo DID in `repo_dids.txt` (resumable) |
| `stage2_backfill.py` | maps repos to owners, finds repos on self-hosted knots |
| `stage3_commits.py` | indexes merged PRs and resolves each author's PDS |
| `stage3b_blobs.py` | fetches patch blobs, counts commits (resumable) |
| `stage4_handles.py` | resolves DIDs to handles via plc.directory |
| `analyze.py` | writes `stats.json` |
| `build_html.py` | writes `embed.html` and `preview.html` |

Seed `repo_dids.txt` from the knot before a fresh run:

    curl -s "https://knot1.tangled.sh/xrpc/sh.tangled.sync.listRepos?limit=1000&cursor=N"

`valpal.py` checks a chart ramp against the black surface — ordinal monotonicity,
adjacent lightness gaps, contrast. Run it before changing any chart color.

## Caveats worth keeping in the post

- **Commits are counted from PR patches**, not from git history — Tangled publishes
  no global commit index (`sh.tangled.git.refUpdate` returns empty). Commits pushed
  straight to a branch are not counted.
- **Time to merge is not measurable.** `createdAt` is asserted by the author's client;
  `stateUpdatedAt` derives from a different record's key, written by a different person
  on a different machine. 18% of other authors' merged PRs report merging before they
  were opened. Month bucketing survives this; durations do not.
- **Merge rate counts only decided PRs.** Counting open PRs as failures compares one
  month of work against the network's two-year backlog.