Supplement, 12 November 2026. Not a source paper.
Seed-packet reconstruction
A shared recipe turns a 4-byte seed packet, or an 8-byte rank packet, back into the same file bytes. This page does not change Paper IX or the unpadded 399 ledger. The power form is a separate construction: its seed 399 selects 399²³ and a 68-byte print source, not one of these 30 recipes.
The criterion
Let be the file tree for version , recipe , and seed . If the parser returns the pair that the encoder sent, the receiver applies the shared recipe and obtains . The package states the general criterion: one packet determines one artifact exactly when equal packets require equal artifacts. Necessity is one result per packet. Sufficiency chooses any input that produced the packet. The Lean file names this reconstructible_iff_constant_on_fibers. Its sufficiency proof uses classical choice. That file was not compiled here.
The rank packet
and . The predecessor interval is the same one on the transport page. Seeds 398, 399, and 400 all step to 135, with ranks 0, 1, and 2. A packet that carries only 135 cannot separate the three files.
From 399, the first six ranks are 1, 2, 0, 0, 2, 1. Packed, that is . Version 1 and WebAssembly recipe 10 give
01 0a 09 06 00 00 00 00
The recipe emits a 37-byte module. Its i32.const 399 is 41 8f 03, because . This archive instantiated that module for seeds 0, 399, and 32767. Each run() returned that seed. SeedPacket.lean states the 37 bytes in wasm_399. The supplied results say Lean checked them. Lean was not run here. witness_399 is a different recipe: [1, 2, 9, 6, 0, 0, 0, 0]. The rank payload is the same six bytes.
What was run here
python3 verify.py --bytes-only fails on Python 3.10.21 because tomllib is absent. /usr/bin/python3.11 verify.py --bytes-only and the same command under -O both exited 0. Stderr was empty. The two stdout files are the same 547 bytes, SHA-256 e0e32eee6df167b4d975484ef8d3868ffaf22d8e7b885f87a4c5af06a4784483.
Python 3.11.2 and Node v22.23.2 agreed on 2,520 reconstructions: 30 recipes, 21 seeds, both packet formats, and both directions. There were 378 exact data checks and 7 rejected malformed packets per receiver. The packet and canonical-tree corpus is SHA-256 c44a60d5bb90176964985b51992c995238c01e7c97267199f3ddfe93888af85e. That digest is the one in the supplied RESULTS.txt. Those results name Python 3.12.14 and Node v24.22.1.
Separately, this archive ran 33 generated programs: recipes 2 through 8 and 10 through 13, on seeds 0, 399, and 32767. The compilers were OpenJDK 17.0.20, rustc 1.97.1, Julia 1.11.6, and Debian cc 12.2.0, plus the Python sqlite3 module. That is not the package command python3 verify.py. Recipe 9, the Lean proof, was not run. The supplied results record 36 program executions and 21 WebAssembly modules matched by the Lean receiver. Those two counts were not repeated.
Seed 399, selected recipes
The sizes are the file contents in the package. The data checks cover all 21 seeds. The program runs cover three seeds, and they skip the Lean proof.
| Artifact | Bytes | Checked here |
|---|---|---|
| Python project | 182 | Program output on three seeds |
| Java, Rust, Julia | 251, 142, 190 | Program output on three seeds |
| Lean proof file | 128 | Hosted. Not type-checked here |
| Web project | 675 | Program output on three seeds |
| WebAssembly | 37 | run() on three seeds. Not the Lean comparison of all 21 |
| SAT instance and witness | 386 | Assignment check, 21 seeds |
| Polynomial samples | 53 | Six rational coefficients, 21 seeds |
| Generated integer record | 119 | Eight exact readouts, 21 seeds. Not the frozen SRSD audit |
| Paper IX network history | 4,404 | Ten floor steps on 120 nodes, 21 seeds |
| PCM and MIDI | 588, 90 | Sample and note checks, 21 seeds |
| OBJ mesh | 409 | 25 vertices and 16 faces, 21 seeds |
For the generated record at seed 399, row 1 is scale 20, quotient −22, residual 17, and . That identity is the checker's, on this generated table. It is not a new SRSD theorem.
Recipe 19 reads graph_fixture.json, 8,242 bytes, SHA-256 40bc6a1d52efab3d75cf343e65ac06b314f9c9c375fc02bf6d45bf6bb21f9529. Those bytes match the Paper IX specification fixture and /dynamics/600cell/graph-fixture.json. Using the fixture does not change Paper IX or the interval 306 ≤ ℓ* ≤ 360.
SeedPacket/Codec.lean is the same file as the transport codec. A text scan of the Lean sources finds no sorry, admit, native_decide, or declared axiom. That scan is not a compiler run. The supplied results record Lean 4.19.0, commit 6caaee842e94, a completed build, and axiom lines for eight theorems. seed_399, witness_399, and wasm_399 are reported as depending on no axioms. reconstructible_iff_constant_on_fibers is reported as depending on propext, Classical.choice, and Quot.sound.
What the packet does not do
For a fixed decoder, packet bits select at most outputs. The smaller transmission is the cost of sending the recipe once and then sending packets, compared with sending the artifacts. The receiver spends the computation. That comparison is not a claim about a packet whose recipe is not already shared.
- No Lean compilation on this archive.
- No written-proof review.
- The 36 program executions and 21 Lean WebAssembly matches were not repeated.
- No tenth Zenodo deposit, and no change to Paper IX.
The package is OMEGA_Seed_Packet_Proof.zip, 56,303 bytes, SHA-256 42a9a96e2bc4d311fd4363fa32a027860ee6d1402f088d8162c2b4e33bd5e868. Fifty-two checksums matched. This archive's log is stdout.txt.