docs: namespace-only resolution — containment tree as single naming authority (operator-signed)

[gunbai-bot]

Records the operator-signed (2026-07-06) committed direction for name resolution. Docs-only (DESIGN.md open-threads entry + the design doc); no code.

What this commits to

Name resolution collapses to one structure — syntactic containment. Nesting induces everything at once: qualified name = nesting position (T{a}→ a is T.a; T{a{a}}→ inner is T.a.a → Rule-2 by construction), reference = lexical lookup up the ancestor chain (sibling module visible, members projected node.Symbol), . = projection down (field/module-member/variant, one op), and . is the std.induction sub-value relation. One content-addressed tree grounds three consumers — resolution walks it, content-addressing hashes it, termination reads its sub-value edges (§2 Realization).

• Supersedes resolver-graph-major §1c's import-name-universe (carries its mechanics forward: binding-edge owner, unbound/collision = error, patterns-via-scrutinee, no expected-type picker).
• Rides on SymbolIndex = the containment tree materialized (type-env-single-authority, lively-raven lane); import-scoped and namespace-only-Y are two walk-rules over one fill-agnostic index. Gated on loyal-heron's scaling receipt before any resolver surgery.
• Terminal step: delete the import grammar → import becomes a parse error, deps derived from container.member references (Rule-1 end-state).

Notes for review

• Docs-only, no resolver code — this is the signed direction + inbound DESIGN.md link (clears the orphan-doc lens). Implementation is separately gated on the SymbolIndex substrate + scaling receipt.
• Clean 2-file branch cut fresh from main (the session branch was contaminated by auto-committer regen sweep divergent from the merged Emitter-seed re-sync to im_rc carriers: prime emit machinery, regen container-dimension seeds (92->30 verify staleness) #6322; deliberately avoided PRing that).
• ci floor is pre-existing-red on main (merge-admission-gate compile-wall P3+P5: infer whole-corpus scans to per-module maps #6239), independent of this docs change; rust_tests is the relevant gate.

🤖 Generated with Claude Code

[gunbai-bot]

Both findings valid — fixed in 4f5f263f94 (docs-only staleness from incremental edits):

1. §9 dropped the 'uniqueness test (X vs Y)' item — §4 records DECIDED: (Y), so it's no longer open (removed the contradiction an implementer could re-litigate).
2. §9 dropped the 'precise census' to-verify item and §8 step 2 no longer says 'run the census' — §5.1 already reports it (declared module paths, collision counts, fork worklist); steps 3–4 now act on §5.1's outputs, no re-run.

Thanks for the catch — both were exactly the kind of same-doc single-authority drift the design itself argues against. — sent from loyal-dove-903