Record sealed ECP-7 hard-utilization batch

[FatosLocos]

Summary

• preregister ECP7-B3-I as a one-variable follow-up to the Batch 2 relaxation gap
• feed the unchanged utilization formula the same straight-through one-hot messages used by receiver training
• validate that forward messages are discrete while gradients still reach the sender
• run the unchanged ECP-6 positive control and ECP7-B3-I on the sealed development split
• preserve the negative but directionally improved result in compact tracked evidence
• define direct joint collision pressure as the next isolated development question

Development result

The positive control again achieved 100% on every registered measure. ECP7-B3-I remained below the development gate, but materially improved the discrete protocol over Batch 2:

• unique messages per sender: 585–972, up from 85–104
• hard-message entropy: 8.20–8.73 bits, up from 6.13–6.31
• population validation mean: 1.9226%, up from 0.6165%
• universal-translator validation: 2.2705%, up from 0.6592%

Color accuracy improved to 17.67%, while shape stayed near its 6.25% chance level. Marginal balance and pairwise slot independence still permit extensive higher-order full-message collisions, so the confirmatory test remains sealed.

Validation

• 54 pytest tests pass
• B3 and positive-control configs validate at 14336/1024/1024
• both smoke arms 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 compositional_test metrics exist
• every logged message respects the exact 14-bit local channel