Both pins take core 0.19.0, and an experience shows its part

[robertblust]

Two releases arrive together because the sites were held back for them rather than re-pinned twice.

0.18.0 fixed a comparison. A date takes the bound its field names — a start the interval's first instant, an end its last — so a period ending 2002 no longer sorts as though it closed on the first of January, before one that ended that March. Nothing on the timeline moves: the ledger sorts by start, where the rule is what it always was.

0.19.0 gave an experience a role, the part played where the H1 names the period and not the part. Eighteen of this model's thirty-six carry one; the other eighteen split between positions whose H1 is already the part and entries where the model does not say, and none was invented.

The model page draws both. It also draws the surface entity, which has grown from eight rules to twenty-one across four productions of the profile it describes — each one finding what reading the file could not.

Verified: npm run verify passes against a served copy; model:check, pages:check, design:check, pin:check, og:check after npm run og, test:og, test:build, test:dupes, conventions-check and conventions-sync check all pass.

🤖 Generated with Claude Code

[robertblust]

Superseded. The model-pin half landed in #170, which moved source.json past this branch's 70496b0 to 5473271 and rebuilt model.json, the principles page and the three og hashes from it — so this branch now conflicts with main on every file except one line.

That line, the companygraph-meta-model pin, is still worth moving and is being taken straight to v0.20.0 rather than landing a v0.19.0 that is already behind. It changes nothing observable: lib/instance.mjs is the same blob at v0.17.0, v0.19.0 and v0.20.0, which is why #170 could rebuild the model locally against 0.19.0 and still match CI's build against the locked 0.17.0 byte for byte.