chess-spectral v1.19.0 — srmech profile pattern (Task #211, ADR-0001 §7 Step 1)

[lemonforest]

Summary

ADR-0001 §7 Step 1 — chess-spectral simple-profile POC, the first downstream consumer of the srmech v0.3.x profile pattern. Validates ADR-0001's simple tier against a real research package.

This is the rc/production rc cycle for chess-spectral v1.19.0. Per your TestPyPI-config note, the publish workflow is now updated to match srmech and ephemerides-spectral.

What ships

Profile-pattern integration

chess-spectral is now discoverable as the `"chess"` srmech profile:

```python
import srmech
chess = srmech.profile("chess")
enc = chess.encode_2d(chess.fen_to_pos(fen))
```

Eight bridge surfaces declared:

• `encode_2d`, `encode_4d` (canonical 640-dim / 45 056-dim encoders)
• `fen_to_pos` (FEN parser)
• `channel_energies` (per-channel L² introspection)
• `encode_2d_pure_phase` (integer-arithmetic encoder)
• `phase_only_pseudo_legal_moves` (ALU-native move enumeration)
• `encode_2d_bip_hybrid` / `decode_2d_bip_hybrid` (sign × magnitude compression)

Same 8 functions also register as `srmech.amsc.tool_schema` entries (owner=`chess`).

Tier: Simple profile. chess-spectral's own ctypes binding to its C library stays internal (`_native_pure_phase_2d.py`, `_native_bitboard4d.py`); srmech only exposes the Python bridge surfaces. This is the canonical simple-tier pattern from ADR-0001 §5.

chess-spectral-publish.yml — TestPyPI auto-routing

Matches the pattern srmech and ephemerides-spectral use:

Tag	Route
`chess-spectral-vX.Y.ZrcN`	TestPyPI (auto)
`chess-spectral-vX.Y.Z`	PyPI (auto)
`workflow_dispatch` w/ target	Manual override

Also added:

• Description-match guard between `pyproject.toml` ↔ `pyproject-pure.toml` (mirrors srmech's rc8 lesson).
• PyPI Summary 512-char limit guard with 480-char soft warning.

Thanks for setting up the TestPyPI config on test.pypi.org for chess-spectral!

End-to-end verification

Built chess-spectral 1.19.0 wheel locally + clean-venv test against production-PyPI srmech 0.3.1:

```
srmech: 0.3.1 | chess-spectral: 1.19.0

Profile: chess 1.19.0
chess tools registered: 8
encode_2d shape: (640,)
channel_energies: 10 channels
```

Why srmech>=0.3.1 (not 0.3.0)

The chess-POC migration is what surfaced both v0.3.1 fixes (Form-1 entry-point support + `[profile.tool_schema]` extension loading). v0.3.0 would silently enumerate the chess profile as `status='invalid'`. v0.3.1 makes it work end-to-end. Hence the floor.

After merge

Per the autonomous-rc-cycle discipline:

1. You merge with `gh pr merge --merge` (no squash).
2. I'll tag `chess-spectral-v1.19.0rc1` → publish workflow auto-routes to TestPyPI.
3. Verify clean-venv install from TestPyPI works end-to-end against PyPI srmech 0.3.1.
4. If green → open the v1.19.0 production PR for your review.

Test plan

• CI matrix on this PR passes (`chess-spectral-ci`)
• Review the changes — 7 files: 3 new profile artifacts + 2 pyproject bumps + workflow + CHANGELOG
• You merge with `--merge` (not squash)
• I manually tag `chess-spectral-v1.19.0rc1` → TestPyPI publish
• Clean-venv install from TestPyPI; verify `srmech.profile("chess")` works
• On green: open `chess-spectral-v1.19.0` production PR for your review

What's NOT in this PR

• Task ephemerides-spectral 0.13.6 -> 0.13.7 (JPL Rule 10 pedantic-build matrix) #212 (ephemerides-spectral plugin-profile POC) — next phase after this lands.
• Task ephemerides-spectral 0.13.7 -> 0.13.8 (README accuracy patch — two-stage architecture) #213 (antikythera-spectral profile) — Step 3.
• Task ephemerides-spectral 0.13.8 -> 0.13.9 (JPL Rules 6+7 manual audits — all 10 rules satisfied) #214 (third-party authoring guide) — Step 4, written after all three POCs.

\U0001F916 Generated with Claude Code