Four reasons this is not 1.0 yet.

The four boundaries

Quoted from README_V1.md. The package is RC1 rather than stable 1.0 because:

1. official MMLC Runtime execution still requires a network-enabled replay environmentclosed 2026-08-01 — the replay was run, and all four ledgers now pass
2. machine-profile transfer has format and sensitivity tests, but no independent physical-machine dataset yet
3. route-conditional conformal bounds are not simultaneous adaptive-selection guarantees
4. the benchmark corpus is still small

What is not claimed

• A conformal upper bound≠a simultaneous guarantee over all adaptively compared routes
• A machine profile≠transferable to another physical machine on this evidence
• Calibration matching measurement once≠calibration being correct in general
• A valid certificate≠a claim that the selected route was optimal
• A conformance executor bundled in this package≠the official MMLC Runtime — results from the two must be stated separately

The last one is stated in the project's own README, in the same paragraph that describes its bundled executor. A local approximation of a runtime is useful for development and is not evidence about the runtime.

The corpus

Thirty cases, five routes, one machine. That is the fourth boundary, and it is the one no amount of engineering closes — it needs more measurement.