docs: lobes embedder as the reference deployment; scores are model-relative (#10, #11)#12
Conversation
…al:... Documentation-only change in this repo (coherence-cli). Goal: name the lobes embedder role as the reference deployment for the embedding endpoint, worded to fit the upcoming five-domain positioning (issue #11: quality, meaning, signal, investiture, frames — meaning is Layer 2) and the model-relative framing of issue #10 (scores are model-relative, anchor-defined semantic measurements — never universal meaning). Changes: (1) In README.md, find the section documenting COHERENCE_EMBED_URL / COHERENCE_EMBED_MODEL (defaults http://localhost:8002/v1 and Qwen/Qwen3-Embedding-0.6B — do NOT change the defaults) and add a short 'Reference deployment' note: the lobes gateway serves an 'embedder' role discoverable via GET /capabilities (lobes-cli >= 0.38 advertises a client-reachable endpoint per role); consumers like colleague resolve the endpoint from lobes and inject COHERENCE_EMBED_URL/COHERENCE_EMBED_MODEL when armed; also one sentence noting each meaning score is a model-relative, anchor-defined measurement — the embedding model + anchors are the measurement's reference frame (gauge), per issues #10/#11. (2) In coherence/meaning/embed.py, extend the module docstring with the same two facts (lobes embedder as reference deployment; endpoint+model = the score's frame provenance). Keep markdownlint-clean lines (this repo lints all markdown). Do NOT change any code behavior, defaults, tests, version, or CHANGELOG.
#10, #11) Record the CHANGELOG entry for the README + embed.py docstring update naming the lobes embedder role as the reference COHERENCE_EMBED_URL deployment, and re-sync uv.lock's stale recorded version (0.4.1 -> 0.5.0). Authored by the colleague backend (local 27B) and graded 5/5 through colleague's feedback loop.
PR Summary by QodoDocs: clarify lobes embedder reference deployment & model-relative meaning scores
AI Description
Diagram
High-Level Assessment
Files changed (4)
|
main advanced to 0.6.0 (five-domain engine, #14) while this docs-only branch sat at 0.5.1, colliding on the version-coordination files: - pyproject.toml / uv.lock: 0.5.1 vs 0.6.0 -> re-bumped to 0.6.1 (docs-only patch on top of main's 0.6.0). - CHANGELOG.md: main's #13 already took the [0.5.1] slot with different content; relocated this branch's note to a fresh [0.6.1] section, dropped the now-moot uv.lock-resync "Fixed" line (main's lock was never stale in the 0.6.x line), kept main's [0.6.0] and [0.5.1] intact. - Semantic conflict (textually clean): main added tests/test_docs_language.py banning the phrase "universal meaning" in README.md; this branch's note used it ("never universal meaning"). Reworded README, CHANGELOG, and the embed.py docstring to "no model-independent semantics" to match main's house language. 672 tests pass.
|



Summary
Documentation-only change (colleague#291 slice S2, relates to issues #10/#11 — colleague#293 is the consumer side). Adds a README "Reference deployment" note (Meaning Gradient section) and a matching
coherence/meaning/embed.pydocstring addition naming the lobes gateway'sembedderrole (lobes-cli >= 0.38, discoverable viaGET /capabilities) as the referenceCOHERENCE_EMBED_URL/COHERENCE_EMBED_MODELdeployment — consumers such as colleague resolve the endpoint from lobes and inject those two env vars when armed. Both spots also note, worded per issues #10/#11, that each meaning score is a model-relative, anchor-defined measurement (the embedding model + anchor set are the score's reference frame/gauge) — never universal meaning. No defaults, code behavior, or tests changed.Also re-syncs
uv.lock's recordedcoherence-cliversion, which was stale at0.4.1whilepyproject.toml/CHANGELOG.mdhad already moved to0.5.0— a self-heal, not a new behavior change.Honest note on provenance: the documentation change itself (README + embed.py docstring) was authored by the colleague backend (the locally served 27B model, via
colleague work), not by Claude, and was graded 5/5 through colleague's own feedback loop. This PR's version bump + CHANGELOG entry (this commit) were prepared by Claude following the standard release mechanics for the colleague#291 integration arc.Test plan
uv run pytest -n auto -q— 146 passedNo known deviations or honest limits — this is a docs-only change.
🤖 Generated with Claude Code