The iota Gateway

The iota gateway is a model access layer in which neural weights are represented as addressable evidence streams rather than only as completed files. A model artifact is split into a scale-bearing lane and progressively finer refinement lanes. A receiver may open the artifact from local disk, remote object storage, a network drive, or any range-readable host; fetch selected lanes; cache evidence locally; reconstruct a valid model state; and run inference without sending the user's prompt to the storage server.

This paper gives a ground-up formal specification of the iota gateway: floating point evidence, the Rouyea shave, lane decomposition, the crisp byte, prefix-stable topcut, zero-decode observability, evidence pricing, bit-packed and deflated lane storage, tensor-aware manifests, cached gateway retrieval, active evidence scheduling, native bounded-memory lane computation, and the Bourgeois orbience evaluator.

The central claim is deliberately precise: model loading can be transformed from a binary file transfer into a continuous evidence negotiation, and model execution can be organized around bounded evidence streams rather than fully resident tensors. Measured on real Qwen-family artifacts and running Rust code, a bounded-memory native stream sweep executed lossless matrix-vector operations over every measured weight matrix of an 8B-class model with peak resident memory below one-third of raw model size, and removing an unnecessary bit-plane expansion detour cut a single-tensor open path from about 3000ms to about 47ms.

1. 1. The Finding1
2. The old loading model1
3. 3. Floating Point as Evidence2
4. 4. The Rouyea Shave Algorithm3
5. 5. Lane Decomposition3
6. 6. The Crisp Byte4
7. 7. Artifact Format4
8. 8. Lane Codecs5
9. 9. Prefix-Stable Topcut5
10. 10. Zero-decode Observability6
11. 11. Error Envelope6
12. 12. Tensor-Aware Artifact Folders6
13. 13. Storage Backend Interface7
14. 14. The iota Gateway7
15. Gateway Roles7
16. 16. Caching7
17. 17. Value of Evidence8
18. 18. The Bourgeois Orbience Algorithm8
19. 19. Certified Collapse8
20. 20. Active Gateway Scheduling9
21. 21. Native Bounded-Memory Execution9
22. 22. Conformance levels10
23. 23. Empirical Status10
24. Gateway Storage and Scheduling Results11
25. Bounded-Memory Native Compute Results11
26. Decode-Wall Removal11
27. Interpretation11
28. 28. Established Properties12
29. 29. The Metric12
30. 30. Conclusion12