OSTOE — open-source theory of everything, claims registry
Established Phase B — d(1500) certified to [38, 48]: the 48 incumbent witnessed on disk after one vectorless death, the dual bound driven 26 → 38
id: m1500-phaseb-48-certified · filed: 2026-10-20 · depends: m1500-sparse-milp-bracket · falsifier: survived — the banked 48 was re-verified in pure Python with no solver in the loop (uncovered = 0 over all 1,124,250 ticks); live falsifiers now a reproducible ≤ 47 (47-hunt in flight) or a dual bound ≥ 48
Mechanism
Same object as m1500-sparse-milp-bracket — the rotation-priced covering
of layer 1500 — attacked from both ends by a single uninterrupted lane:
- [F] Feasibility end: a 48-phase schedule exists, and the witness (the
phase vector itself) now lives on disk, not in a solver's memory.
- [F] Bound end: the LP-dual floor reached 38.0, so no schedule with 37
or fewer engaged primes exists — an impossibility claim over all
untried configurations, not a search failure.
- [G] What remains open: the exact integer. The bracket width is 10;
every notch from either side is permanent.
What happened (the connector, not just the endpoints)
- [E] Run 1 (2026-10-18, 24 h budget) held BestSol = 48 for ~10 h with
dual bound 37.5564, then took an external SIGKILL 69 min before its
own time limit. The runner only wrote vectors after the solve
returned, so the 48 died unwitnessed — the campaign's second
vectorless artifact (after the rig-era 47).
- [E] The rerun ran a checkpoint-on-improvement callback: every improving
solution banks to disk the instant it exists. The rerun's bound ladder
replayed run 1's bit-for-bit through six rungs (13.86 → 32.03),
including one replay executed while the host was starved by a foreign
fleet — determinism held numerically across independent processes.
- [E] Clean exit at ~17.7 h: objective 48, dual bound 38.0.
verify_best48.py — pure Python, no solver — rebuilt the model from
scratch (177 rings, 1,124,250 ticks), counted 48 nonzero phases, and
walked every tick: uncovered = 0. PASS.
- [S, in flight] The 47-hunt (launched 2026-10-20, warm-started from the
banked 48, fresh seed) attacks the incumbent end; the bound end keeps
grinding in every run. [F] If the ends ever meet, d(1500) is a proven
exact integer.
Downstream
- The d(m) curve claim — that the wheel's minimal covering size grows
with m in a characterizable way — gains its second certified anchor
with both ends fenced (witness above, proof below). Single-m values
are data points; the curve is the claim others can engage.
- House law established for all rented-box solves: no claimed solution
may exist only in process memory. Two vectorless artifacts earned it.
- If the 47-hunt lands, this entry keeps its status (the [38, 48]
bracket was true when certified) and a new entry supersedes the
bracket, per registry rules.
- Superseded 2026-10-27 — the bracket closed to [42, 45]
(
m1500-bracket-42-45): a witnessed 45-ring cover demoted this
entry's 48 incumbent, and a pure-integer unsat-tree certificate
raised the floor to 42. This entry keeps its status per registry
rules — [38, 48] was true when certified.