Mechanically checked result

FPRD-T98

Claim FPRD-T98

Exact statement

A finalizable delayed resolver emits append-only online tagged batches and an explicit finish batch at end of word. Its semantic visible-stack evaluator has exactly the resolver state and direct visible run on the online output after every prefix, and after finish it has exactly the direct run on the completed annotation. A lawful fixed finite Complete Cover therefore recognizes exactly the forgotten-action projection, including at epsilon. This theorem does not invoke the strict compiler.

Statusmechanically checked
External reviewNo documented external or specialist review of this FPRD result is recorded.

Context

This row is imported exactly from the governed FPRD claims ledger. The matrix assignment records normalized inventory coverage; it is not a new proof or a promotion of the ledger status.

Proof or evidence

The ledger points to formalizations/frontier-derivative/FrontierDerivative/DelayedResolver.lean. The matrix has not yet rewritten that source into a self-contained public proof summary.

Verification notes

Reviewed on 2026-08-25 for identifier, status, provenance, evidence-link presence, and dependency import only. Mathematical proof audit remains the next action unless separately documented.

Limitations

  • Lean 4.30.0 checks online input-prefix splitting under lawfulness, batch-run composition, online_run_exact, finishConfig_exact, delayed_decides_eq, soundness, complete-cover recognition, and the zero family. The one-lookahead calibration proves its input-balance invariant and lawfulness for all words, not only tests. The semantic evaluator may perform several tagged stack actions per observed input and post-input finish work, so it is deliberately not an FPRD.StackMachine and no FPRD.compile theorem is claimed.

Open work

Expose the formal/checker scope, reproducibility instructions, and the gap to a complete mathematical proof.

Notes

Lean 4.30.0 checks online input-prefix splitting under lawfulness, batch-run composition, online_run_exact, finishConfig_exact, delayed_decides_eq, soundness, complete-cover recognition, and the zero family. The one-lookahead calibration proves its input-balance invariant and lawfulness for all words, not only tests. The semantic evaluator may perform several tagged stack actions per observed input and post-input finish work, so it is deliberately not an FPRD.StackMachine and no FPRD.compile theorem is claimed.