frozen pr body - kimjooyoon/meta-ontology-go GitHub Wiki

Change

Expose the existing typed input-only composition API through gooo body-compose --inputs. Exactly one of --inputs, --cases, --case-series is required. Generation, retained repeats, saved replay and policy continuation reuse the existing native implementation.

The CLI envelope records input_schema; input-only output retains the original bytes as inputs.json. Expected-case output still uses cases.json. Unscored execution reports zero expectations, zero passes and UNKNOWN / NO_RUNTIME_EXPECTATIONS. Source-owned construction cases remain authoritative. A later --cases replay scores the same generated graph without new model calls.

Adds runnable examples, help and a dated research note based on original articles encountered through Hacker News, GeekNews and GitHub Trending. No new dependency or model download. The public v0.6.22-dev binary is explicitly distinguished from this development addition.

Evidence

  • Predeclared PLAN and failing CLI tests preceded implementation.
  • New native CLI tests: PASS, including mixed joins/fan-out, exact integer 9007199254740993, native producer-to-consumer values, repeat executable reuse, saved replay and later 49/49 named expectations.
  • Own record-field model: one construction prediction retained; native execution and saved replay use zero new predictions. Input-only values remain unscored.
  • Nested-helper partial continuation: source cases and retained candidate order preserved, zero new model calls.
  • Invalid schemas, mixed modes, expected fields, unknown/missing/ill-typed roots and caller replacement of a bound producer are rejected.
  • Focused race tests, vet, Go format, Go modernizers and billing semantic check: PASS.
  • Full local unit suite was started and has an independently reproduced failure in unchanged cmd/callback-preview/TestPreviewOutputRejectsSymlinkBoundary: it correctly rejects a symlink but the test expects a different error message. Both package source and dependencies are unchanged from the base. The original full PR CI is authoritative; no full-local-pass claim is made.

Scope and limits

This exposes native observation; workbench saved multi-activity feedback is still a separate extension. Actual outputs do not become generated expectations. No global accuracy or completeness claim is inferred from zero-oracle runs. Models and historical observations are unchanged.

Review-cap exception: 422 changed lines (415 added, 7 removed), including 214 native CLI test lines, the predeclared plan, three tiny inputs and runnable/research documentation. The implementation reuses the existing decoder and typed execution engine. Existing usage and parser slots retain their prior line-length/slot exceptions; native fixture JSON rows preserve exact values on one line.

⚠️ **GitHub.com Fallback** ⚠️