Repository navigation
decision(quotient): rebind project_level after burn-in — pipeline_stage vs orchestrator #146
Description
Context
Since #144 the k=1 quotient graph (domains as nodes) is scored observe-only (enforce: false) against pipeline_stage. It stably measures drift 0.700 — exact arithmetic, not noise: the cross-domain census is 021D-dominated because cgis.pipeline imports/calls every other domain (orchestrator reality), while pipeline_stage ideal expects a 021C chain (extraction→resolution→storage).
Decision needed (after ~2-3 weeks of burn-in, i.e. around 2026-07-01)
Pick one:
- Rebind to
orchestrator— accept reality: cgis at domain level is a hub-and-spoke around the pipeline. Drift would drop to ~0; quotient could then be enforced. - Keep
pipeline_stageas aspiration — treat 0.700 as a measured architectural debt: the pipeline module should shrink to thin wiring and domains should hand off in a chain. Keep observe-only until refactored. - Hybrid — rebind to
orchestratorwith a tight tolerance now (regression guard), open a separate refactoring issue for the chain aspiration.
Data to collect during burn-in
- Does quotient drift move at all as code changes land? (If it's static, observe-only adds no signal and we should just decide.)
tests/self_parsing/test_drift.py::test_py_quotient_drift_observed_not_enforcedprints the report on every run — check CI logs.
Blocks: enforcing project_level (and the unresolved_ratio plumbing (#149) only matters once enforced).
🤖 Generated with Claude Code
Activity
docs(spec): language packs and a two-plane topology
docs(spec): revision 2 — forbidden edges as the gate, scope per langu…
docs(plan): language packs — step 0 in full, steps 1–7 outlined
docs(spec): language packs and a two-plane topology (#593)
chore(ontology): step 6 — gates (#605)
zaebee commentedon Oct 14, 2026
Closing: the question this issue asked — what topology cgis should drift toward, and how to keep it there as languages are added — is answered by the language-packs migration (spec
docs/specs/2026-10-14-language-packs-design.md, steps 0–6, every step graph-identical on cgissrc@715cd9a, owner-api@8de2245 andui/).The answer is not a quotient pattern but enforced boundaries. The
cgis-project→pipeline_stagebinding never fit (the measured quotient was an orchestrator star) and stayed observe-only; it is retired (#605). What enforces the topology now is the forbidden-edge gate intests/self_parsing/test_architecture.py, with exact shrinking allowlists:cross-pack,pack-isolation,surface-packs(refactor(languages): step 1 — lazy registry, python/typescript packs #597),engine-packs(refactor(languages): step 3 — per-pack index facts and classification #600 → emptied refactor(languages): step 4 — resolution rules per pack #601),resolver-storage(emptied refactor(pipeline): step 5 — build-plane passes #602),core-isolation— all empty;surface-storage— the remaining allowlisted debt (cli/api/guardian reading storage directly).How it got there
cgis.languages, python/typescript packscgis/pipeline/); the resolver no longer touches storagelayered_dag0.42),coreaslayered_dag,project_levelretired; build/read plane quotients reported observe-only (build 0.351, read 0.676)Follow-ups: #603 (
IngestionPipeline.runcomplexity), #604 (language residue in the resolver's text). Step 7 — the Go pilot — is next and does not block this.