Record sealed ECP-7 position-aware decoder batch

[FatosLocos]

Summary

• add a generic position-aware MLP receiver and use it for both the ECP-7 population and isolated translator
• preregister and execute sealed ECP7-B9-I against the unchanged ECP-6 positive control
• record the negative gate decision and complete artifact hashes without accessing the confirmatory test split
• update the fresh-agent handoff with the evidence-based ECP7-B10 recommendation

Result

The positive control remained at 100%. ECP7-B9-I did not pass the registered joint gate, but improved population train exactness to 71.27%, validation exactness to 52.39%, and translator validation exactness to 54.81%, with roughly 11,000 hard messages per sender.

The confirmatory ECP-7 test split remains sealed.

Verification

• 73 passed
• ECP7-B9 configuration validates with split sizes 14336/1024/1024
• both sealed analyzers passed with 65/65 artifact hashes and no test metrics
• all eight hashes recorded in batch9-manifest.json match the local run artifacts