fix(ts): bind exported generic type and factory declarations across modules

Follow-up under #487 and tracker #416, building on source-owned generic definitions from #524.

The current Program exported-value collector scans .d.ts lines and retains simple type names; source .ts factories and generic signatures are not carried to consumers. Generic interfaces/type aliases are also absent from the corresponding imported-type environment. The original graph gates still fail 0/2.

Replace this projection with bound declaration identities and shared source-owned annotation graphs. Retain generic function parameters, constraints/defaults, required/rest/receiver metadata and parameterized return types. Resolve aliases/re-exports by owner, not display names. Exercise inferred factory results as well as explicit annotations, invalid arguments/arity, missing/readonly properties, name collisions and root-order independence. Keep unsupported inferred bodies and other unimplemented semantics explicit, never add permissive fallback to manufacture admission. Preserve all existing graph rejection controls unchanged.

Calibrate paired positive/appended-negative controls against TS6.0.3 and native TS7.0.2, verify full regressions, document exact remaining failures and only readmit graph timings after real semantic checks pass. Link implementation and documentation commits here.

Baseline audit dd11247 is pushed: 240 controls across local, named, namespace and two-hop barrel placements, both root orders, with positive/appended-only-negative pairs. TS 6.0.3 and native TS 7.0.2 pass 240/240; frozen Home 3b7f0ea passes 130/240 (local42/48; each imported placement22/48). Six local failures cover dependent generic defaults and factory return rebinding under a same-named nested interface; all failures are retained. Evidence graph-types.4EsbvV/audit-before.log; harness69/69. The implementation is in progress: one bound shared annotation graph for type/value export bindings, parameterized factory signatures and owner-resolved interfaces, replacing line-based .d.ts value collection. Original graph controls remain unchanged and ineligible until verified.

Debug candidate evidence: the unchanged original import_graph and reexport_graph correctness gates now both pass for TS 6.0.3, native TS 7.0.2 and Home. The 240-case factory audit improves from Home 130/240 to 220/240; all 20 remaining failures are dependent-default controls, retained unchanged. Static values remain 84/84; nominal origins remain 44/52 (the same eight inherited private/protected failures). A source-owned default substitution fix is now under verification; 10/10 focused checker tests pass. Full regressions and optimized-build correctness/performance are still pending, so this is not a measured-release claim. Logs: graph-types.4EsbvV/{audit-v2,graphs-v2,static_values-v2,nominal_origins-v2,focus-v3}.log.

The dependent-default fix now passes all 240/240 factory controls in Debug (TS6 and TS7 also 240/240). Frozen ReleaseFast candidate SHA-256 9f5e333beaf9828a9720ce260ae05a872ebcea3bb51bf5785ffa3ccd7b72f1bc passes all 19/19 workload admission gates, including both original graph rejection checks unchanged. Evidence: graph-types.4EsbvV/audit-v3.log and admission-release-v3.log. Program133/133, driver187/187, CLI69/69, focused checker10/10 and harness69/69 pass. Full checker/conformance and expanded Release audits are still running; no new timing claims yet. Profiling follow-up #536 tracks duplicate source compilation during CommonJS export queries; any optimization baseline will use this semantically corrected candidate, not the formerly ineligible binary.

Landed implementation de8fe28 and documentation/README f3740df, pushed as Chris without coauthors. Release factory controls240/240, unchanged graph gates2/2 and full workload admission19/19 all pass. Checker4286, conformance1417, Program133, driver187, CLI69, focus10 and harness69 tests all pass. Complete run20260828T224220Z validates570 rounds/1710 samples; fresh generation matches513 measured corpus files. Graph means: Home33.7ms vs TS7 50.8ms; barrel25.1ms vs41.1ms. Home wins means18/19 overall; large predicates lose1.42x (0/30 paired wins), explicitly published and tracked in #537. Profiled redundant CommonJS queries are tracked separately in #536. All remaining audit failures and historical results are retained in docs/docs/TS_PERFORMANCE.md. This resolves the measured exported-factory contracts, not full TypeScript parity.

Fresh pinned-Zod checkpoint under #548 (Home 4fda2da): the exact Zod 4.5.2 src/v4 corpus still exits 1 with 600 primary diagnostics. The dominant codes are TS2345 212, TS2339 165, TS7006 100 and TS2344 43. A representative general failure is a local interface with multiple heritage clauses—one local base plus one namespace-qualified imported generic base—whose imported inherited members are absent downstream ($ZodStringFormatDef extends $ZodStringDef, checks.$ZodCheckStringFormatDef<Format> loses pattern/abort). The existing two-hop single-base controls pass, so I am reducing this specific multiple-heritage/source-owned projection path with TS 6.0.3 and native TS 7.0.2 positive/negative oracles before changing the checker. No Zod-specific names or permissive fallback will be used; the full corpus stays untimed until it reaches zero diagnostics.

Correction to the immediately preceding checkpoint: the exact primary diagnostic-line count is 602, not 600. rg -c " - error TS[0-9]+:" reports 602, and the per-code totals also sum to 602. The four dominant counts (212 + 165 + 100 + 43) were stated correctly; two less-common diagnostics were omitted from the earlier aggregate. I will use 602 as the immutable before-count.

ReleaseFast correctness checkpoint for the cross-module generic/factory work:

• Exact recursive indexed-alias member projection reduction: 0 diagnostics.
• Recursive generic interface heritage reduction: Home 0; TypeScript 6.0.3 0; TypeScript 7.0.2 0.
• Program suite: 187/187 passed.
• Checker suite: 4,353/4,353 passed.
• Pinned Zod 4.5.2 corpus: 602 -> 523 diagnostics. Exact location/code audit: 79 unique baseline keys removed, 0 added.

This removes the newly exposed TS2352, the two to-json-schema.ts:453 TS2345 diagnostics, and the $ZodEnum TS2430 without source-name exceptions or whole-object approximations. Issue remains open because the corpus is not yet at zero.

Pushed the cross-module/generic-owner implementation to main as 197ee07 (fix(ts): resolve inherited generic members (#534)). The companion flow correction is 2e749f7 (#654). The validated corpus checkpoint remains 602 -> 523 diagnostics, 79 unique keys removed, 0 added.

Completed and pushed. The original 240/240 exported-factory controls were already green; 197ee07 now closes the remaining inherited generic/default/member projection gaps exercised by the pinned graph, with checker 4,353/4,353 and Program 187/187. The exact Zod audit is 602 -> 523 diagnostics with 79 unique keys removed and 0 added. Documentation and the README table are on main in 4ef52ad. Further zero-diagnostic work continues under #548 and #416.