fix: re-source Chapter Approved layout geometry from Battlemaster

[wn-mitch]

Re-sources the geometry of all 45 11e Chapter Approved terrain layouts from Battlemaster's public API. Data-only change: no schema, no API, no conformance movement.

Why

The layouts came from tools/extract-terrain-layouts.py, which recovers placements by calibrating per-xref image-stamp CTMs out of the GW Event Companion PDF. That got positions roughly right but rotations wrong, and it predates nubbed footprints being in the template catalog, so nub-aware centroids were never accounted for. Measured against Battlemaster:

	p50	p90	max	affected
Position	0.255″	2.156″	6.605″	520 of 718 areas moved >0.15″
Rotation	94.04°	180.00°	180.00°	498 of 718 changed >1°

Enough of the corpus was out that correcting it piecemeal was not worth it.

What is taken, and what is not

Geometry only. Card identity (id, name, mission_matchup_id, variant, deployment_pattern_id, game_version) and authoring intent (keystones, link_group, objective_role) are preserved from the committed records and asserted to agree rather than overwritten — a disagreement stops the run instead of silently relabelling a card. All 45 ids mapped 1:1, and the deployment-key → pattern mapping was unanimous across all 45.

Also fills the two short-line areas Battlemaster has that disruption-vs-reconnaissance-1 was missing.

Two judgement calls worth a reviewer's eye

EF has no template that models it. Constraint propagation returned an empty domain for CO and EF, which proves the committed data labelled those two ruins inconsistently — so their corpus majority is noise, not evidence. CO was resolved by dimensional impossibility (it is 6″ on its long axis; corner-ruin-balanced-right is 5×4.5″, so the piece cannot be that footprint), deliberately overruling the majority. That leaves EF (4.5×6″) assigned to corner-ruin-balanced-right by elimination, where its 6″ axis does not fit in either orientation. Authoring a proper EF footprint is open follow-up work.

One upstream slip was corrected rather than propagated. Battlemaster places two cards' paired centre trapezoids 0.707″ off 180° symmetry. 43 of its 45 layouts are exactly symmetric, the two exceptions carry byte-identical coordinates, their rotations differ by exactly 180°, and the error is exactly (−0.5, −0.5) — an editor nudge. Propagating it broke the existing keystone-pairing test, which requires a keystone's reflected vertex to land within 0.25″ of its twin so both halves of a printed card measure alike. The pair is snapped to its symmetric mean.

Both are recorded in data/core/_reports/battlemaster-layout-intake.md, along with the per-layout deltas and the part mapping.

Note on the source

The Layout Embed API (/v1/public/embed/*) is documented but not deployed — every route 404s. This used the live TTS Map API's chapter-approved-layout-lite payload, which is better suited anyway: inch geometry and template references rather than rendered images.

Scope

The import tooling was written, used once, and removed in the second commit — the geometry was worth replacing wholesale, and a scraper for a source we do not intend to track again is maintenance without a payer. The net diff is 5 files: the layouts, the three embedded bundles, and the report.

Verification

• TypeScript 1502 tests / 85 files, Rust 22 suites, Python 1168 + ruff + mypy, Go — all green
• npm run validate clean (AJV + referential integrity, 3576 items)
• All four language bundles regenerated; artifact drift is only the three embedded data bundles, the correct signature for a data-only change
• rustfmt / gofmt clean; version lockstep in sync at 1.2.0 (not bumped — release decision)
• Conformance corpora unchanged, so SPEC_VERSION stays at 86. The terrain-resolver and terrain-keystones goldens are built from synthetic cases in gen-conformance.ts, not from terrain-layouts.json
• Geometry invariants gated the write: every layout resolves, all geometry sits on the 60×44 board, and every card holds 180° symmetry

just was unavailable in this environment, so the preflight recipe was run step by step rather than via just preflight.

Not done

• A proper EF footprint (see above)
• Terrain still has no referential-integrity coverage — integrity.ts does not check terrain, so a dangling template or parent_area_id is caught only at resolve time. Pre-existing, not introduced here
• tools/extract-terrain-layouts.py is left in place

---

Generated by Claude Code