Record sealed ECP-7 parallel-sender batch

[FatosLocos]

Summary

• preregister ECP7-B7-I as one sender-architecture change on the Batch 3 loss set
• add a bounded parallel sender whose four slot heads read one shared joint meaning context
• preserve local channel capacities without factor-specific paths, binding matrices, or semantic codebooks
• add isolated checkpoint-worker coverage after a mechanical smoke integration failure
• run the unchanged ECP-6 positive control and B7 intervention on the sealed development split
• preserve compact integrity evidence and select algebraic context invariance as the next isolated question

Development result

The positive control again achieved 100% on every registered measure. ECP7-B7-I produced the strongest optimization and code-use gains in ECP-7 so far, but did not compose across the held-out color-shape matching:

• train exactness: 9.5965%, up from Batch 3's 1.1911%
• unique messages per sender: 3,118–3,415, up from 585–972
• hard-message entropy: 11.11–11.25 bits, up from 8.20–8.73
• color validation accuracy: 45.64%
• shape validation accuracy: 4.84%, below the 6.25% chance level
• population validation mean: 0.5066%
• universal-translator validation: 2.7588%

The architecture is recorded as a negative development result with a meaningful directional gain. The confirmatory split remains sealed.

Validation

• 68 pytest tests pass
• B7 and positive-control configs validate at 14336/1024/1024
• the repeated intervention smoke and control smoke complete through isolated workers
• all 65 captured artifact hashes match for each full run
• all 16,384 episode rows per arm are validation-only
• no confirmatory-test metrics exist
• every message respects the exact 14-bit local channel