# F399 Cottontail-Inspired Alpha-Normalized Constraint Selection Evidence

This directory seals the 2026-08-13 mechanism evidence for F399. It proves
structured alpha-renaming, semantic and alias separation, joint prefix-target
context, transactional duplicate ranks, structural scheduling with bounded
duplicate penalty, production defaults, lifecycle-scoped provider discovery,
and complete local regression status. It contains no target campaign and makes
no coverage, solver-throughput, bug-yield, or end-to-end speedup claim.

| Artifact | Meaning |
| --- | --- |
| `targeted-python.txt` | Nine F399 normalization, context, scheduler, and default tests |
| `related-python.txt` | QueryStore, self-config, QF_BV, and semantic regressions |
| `full-python-gate.json` | Capability-closed 989-node exact-identity Python gate |
| `full-lit-llvm17.txt` | Complete LLVM 17 lit gate |
| `randomized-property.json` | 1,000 alpha-renaming, constant, and alias checks each |
| `mechanism-benchmark.json` | 1,000-run shape-construction cost benchmark |
| `coordinator-provider.json` | Executable coordinator parameter contract |
| `pytest-inventory.txt` | Canonical node-ID inventory result |
| `source-contract.txt` | SHA-256 identities for implementation and tests |
| `static-checks.txt` | Ruff, compilation, whitespace, diagram, and documentation gates |
| `upstream-revision.txt` | Exact Cottontail paper/artifact provenance and audited distinction |
| `environment.txt` | Host and toolchain context |
| `SHA256SUMS.txt` | Exact evidence-directory manifest |

For 64 reads and 192 nodes, the isolated final timing run reports a median of
970.792 us for a cold 64-root shape, 514.565 us for the actual target plus
eight-prefix-root admission pipeline, and 73.100 us when reusing the validated
DAG for both projections. These are synthetic Python mechanism costs only.