Seam badge, and a pasteable cosmetics.json entry

[timbogdanov]

Phase 5 of the v0.2.0 plan — OpenFront fidelity, continued.

The preview now says whether a pattern shows a seam once the game tiles it: Seamless, Slight seam, or Visible seam.

Why this is a port, not a re-derivation

The metric is subtler than it looks. A correctly-tiling periodic pattern legitimately has a hard transition at the wrap — period-8 stripes on a 32-wide canvas put a stripe boundary exactly there. What makes a seam visible is the wrap being an outlier: more discontinuous than the pattern's own interior ever gets.

So the wrap is ranked against the 90th percentile of interior transitions, not their mean. The mean is dragged down by the many identical column pairs inside a periodic pattern, which is precisely what made an earlier version of this metric flag 17 of 28 correctly-tiling patterns.

It also carries the rhythm check, which local adjacency cannot do: period-7 stripes on a 32-wide canvas make a perfectly ordinary-looking edge while the cadence breaks on tiling. Driven in the running app:

Pattern	Verdict
Wordmark	Seamless
Checker	Seamless
Period-7 stripes on 32 wide	Visible seam

Cost

Scoring walks the bitmap several times looking for its intrinsic period, so it settles 250 ms after a stroke rather than running on every pointermove.

Proof

The DSL fixture now carries both scores and the label for all 62 programs, and the suite compares them to 1e-9 — pinning the numbers rather than only the verdict, so a port that lands on the right label by luck still fails. All three verdicts appear in the corpus.

cosmetics.json entry

Export gains a pasteable entry. The file keys patterns by their encoded data, so the entry is that key plus a slugified name. role_group is entitlement metadata the game assigns — a new pattern does not claim it.

"APF4h8PhcIfD4XCHw-Fwh8...": { "name": "my_cool_pattern" }

🤖 Generated with Claude Code

https://claude.ai/code/session_01UCrKKTD2mwHx8Fm1EnJiQ3