Battam1111/Myco

v0.8.7 - 2026-05-13 - URL-drift fix + 5-batch slimdown sweep

Six-commit cleanup sweep responding to the post-v0.8.6 audit. The
substantive user-visible fix is the docs/ → .docs/ URL drift
(PyPI logo + Documentation link were 404 across v0.8.4–v0.8.6); the
rest is structural slimdown that doesn't change runtime semantics
but makes the substrate's working surface dramatically scannier.

What changed

P0 — URL drift (user-visible PyPI 404s)

The v0.8.4 root-cleanup renamed docs/ → .docs/ but left 24
hardcoded GitHub-raw / blob / tree URLs pointing at the pre-rename
path. PyPI's project page was the most visible casualty:

• pypi.org/project/myco/ logo was a broken <img> for 3 releases.
• The "Documentation" sidebar link returned 404.

curl verification of the fix:

GET .../main/.docs/assets/logo_light_512.png → 200 ✓
GET .../main/docs/assets/...                 → 404 (pre-fix state)

24 sites updated across 9 files (README + pyproject + i18n READMEs +
example MYCO.md / _canon.yaml × 2 + schema README).

P1 — Craft archival (17 LANDED crafts → _landed/<era>/)

.docs/primordia/ had grown to 21 LANDED crafts flat at root,
matching the same accretion pattern that motivated the v0.5.x-era
archival to _landed/v0_5_x/. Restored discipline:

• _landed/v0_6_x/ — 6 crafts (2026-04-28 → 2026-04-29).
• _landed/v0_7_x/ — 8 crafts (2026-04-30 → 2026-05-10).
• _landed/glama/ — 3 crafts (2026-04-24 → 2026-04-28).

21 cross-reference updates across 11 doctrine files repoint at
_landed/<era>/. .docs/primordia/README.md rewritten to surface
the active v0.8.x working set first.

P1 — contract_changelog.md archive split

Main changelog was 1739 lines (1241 of which described v0.7.10 →
v0.6.10). Split at v0.7.10:

• Main file (513 lines): v0.8.7 → v0.8.0 active window.
• Archive at .docs/contract_changelog/_archive/pre_v0_8_0.md
  (1257 lines): full immutable history of v0.7.10 ← v0.6.10.

Format preserved (newest-first, one section per version); content
fully preserved (no deletion); git history retains the unified state.

P2 — Config-layer slimdown

• myco.mcp shim: docstring compressed from 53 lines of
  historical narrative to 16 lines of essential migration pointer.
  Full file 93 → 47 lines (−50%). Behaviour byte-identical:
  re-exports build_server + main, emits stderr deprecation +
  DeprecationWarning. Verified import: 1 warning emitted, exports
  intact.

• .docs/comparisons/ → .docs/comparisons.md: a 1-file
  directory is navigation friction. git mv to single file.

• v0.8.x coverage IOU resolved + archived. The IOU's trigger
  condition ("Y commits") was satisfied at commit 2e538d8 (v0.8.4
  prep); the gate flip was just pending. Now done:

  • .github/workflows/ci.yml::--cov-fail-under=82 → 85.
  • Verified with the exact CI command: total 86.26% (1.26 pts
    of headroom above the gate).
  • IOU moved to .docs/iou/_archive/v0_8_x_coverage_uplift.md
    with RESOLVED header.

• coverage_floors.py prefix bug: FLOORS dict was keyed by
  myco/<pkg>/ but coverage.xml class filenames are rooted at
  <pkg>/__init__.py (no myco/ prefix; pytest-cov strips
  src/myco/). Every per-package check silently exit-0'd since
  v0.6.0. The "Coverage per-package floors" CI step was a no-op
  for ~9 minor versions. Fixed:

  • Keys rewritten to match the actual prefix shape.
  • Floors temporarily calibrated to currently-achieved values
    (core=94, homeostasis=86, circulation=82) with explicit
    target N at v0.9 comments so the v0.9 mock-test pass owns
    the uplift work.
  • Verified: 8/8 floors actually pass now (was 8/8 silent SKIP).

Bumper bug (caught during this release):

• .scripts/bump_version.py::sync_cmd had a literal
  scripts/sync_plugin_mirrors.py (pre-v0.8.4 path); the v0.8.4
  hidden-prefix sweep had missed this script. Any Myco-self bump
  would have failed at the sync step. Fixed in the v0.8.6 release
  commit; verified working at v0.8.7 bump (used the corrected
  .scripts/ path).

Break from v0.8.6

None. The CI gate flip back to 85% is a re-tightening of an
intentionally-loose v0.8.4 hotfix; the IOU documented the trigger
condition and the v0.8.7 measurement confirms it. The
coverage_floors.py revival starts enforcing per-package floors
at currently-met baselines — no run-time behavior changes; only
the silent-no-op CI step becomes a real check. Downstream
substrates that consume myco-install, mcp-server-myco, or
python -m myco see byte-identical behavior; the myco.mcp shim
slimdown preserves the public API + deprecation surface exactly.