This packet covers primes 359, 373, 379, and 383. It is a proposed proof packet; generation and source reproduction do not establish any complete prime theorem. The accepted import, theorem, and kernel registries are unchanged.
| Prime | Normalized minimum pairs | Numerical frames | Shards |
|---|---|---|---|
| 359 | 178 | 31,750 | 16 |
| 373 | 185 | 31,043 | 16 |
| 379 | 188 | 35,430 | 18 |
| 383 | 190 | 30,048 | 17 |
Every root covers folded supports of size at most six. A root remains entirely within one shard. Each shard has at most 2,000 numerical frames, 24 roots, and 160 project modules. Numerical checks use 32-frame source files. At most three shards run concurrently.
The unchanged producer determines the certificate data and theorem statements. generate_six_speed_late_packet.py checks large compatibility tables in subtrees with at most four entries, then composes those proofs with Boolean conjunction. The original table definitions and public compatible_checked statements remain unchanged. Numerical frame proofs use the structural population count in FuelCounts and FuelCuts; their conclusions remain the original Cuts.boundedCheck propositions. Removing these two proof transformations recovers the exact original source bytes.
The first hosted attempt, run 44555828907, failed while checking the original large compatibility tables for prime 359: the kernel exceeded its 4,500 MiB memory allowance. The smaller arithmetic proofs address that reached failure. Source reproduction covers all 741 roots and 128,271 frames, but acceptance still requires every fresh shard gate and the four complete joins. A new candidate commit supersedes the preceding workflow attempt; receipts from different attempts cannot be combined.
The workflow freshly compiles every shard's complete project closure with -DwarningAsError=true, audits all owned declarations including private ones, and runs the unchanged pinned stock checker on exactly the isolated module set. Every numerical frame is also a required theorem endpoint. Only propext, Classical.choice, and Quot.sound are permitted as transitive axioms.
Each final join requires all its prime's shard receipts from the same Git commit, workflow run, and run attempt. It checks exact root coverage, regenerated sources, source commands, audit inventories, kernel replay logs, and every preserved evidence hash. Copies of a shared proof object must agree byte for byte. The join then compiles, audits, and stock-replays the full classification aggregate.
The aggregate statement concerns actual positive primitive integer speeds v : Fin 6 → ℕ with loneliness at most 1/7, and concludes permutation of the literal list [1, 2, 3, 4, 5, 6]. It still assumes both that the selected prime divides no speed and that LonelyRunnerBound 5 holds. The audit expands the gcd, positivity, lower-runner, prime-avoidance, and actual-value hypotheses in explicit type checks. Neither hypothesis is silently discharged by a prime receipt.
The frozen source model is docs/six-speed-late-batches-sources.json. verify_six_speed_late_batches.py --preflight checks its inventories and controls; --reproduce reruns the complete finite searches and source rendering. Neither command invokes Lean or reports mathematical proof acceptance.
The existing four-prime development layout and the original failed or interrupted local probes remain separate from this packet. No partial root, benchmark, or row-only checkpoint supplies a complete-prime receipt.