/reports/overnight-2026-10-10.md · raw

bb6-hunt: report for 2026-10-10 to 10-13

Kills: 31, each with tools/check_kill.sh PASS three times

state	machines
Accepted upstream (13)	#2 (PR #14) · #3's table, 4 starts (PR #15) · #11's table, 4 starts (PR #16) · W4-3's table: 341, 512, 588 (PR #17) · C2-552, halts (PR #18); #17/#18 merged 2026-10-14 06:20 UTC
Claimed, not yet submitted (18)	#30's table: #30, #7, W4-1 · #4, #4b · #12, #12b · #8 · W4-2 · #14 · W5-1 (349) · W5-3 (134) · W5-4 (525) · W5-5 (29, halts) · W5-7 (700, halts) · W5-8's table: W5-8 (214), W5-8b, W5-8c

Halting kills: C2-552 (≈1.77e44 steps), W5-5 (1,816,824,067,232,869,945 steps), W5-7. In each case the step count is computed and cross-checked; the Coq proofs show only that the machine halts.

Skelet 1-like Class 2: 12 of the 14 on the 740 list are upstream. Left: R4 and #32 (one table), stopped as cryptid candidates.

Other machines

CRYPTID_CANDIDATE (11): W5-2, R4, #32, #5, #26, #10, #28, #1, #6, #27, and 362 (with 433), added by the halting hunt.
STUCK / parked: stop rules or your answers. Includes #16, W4-6 (+612), W4-7 (+707), W5-6 (+280), W4-4's family, R5 and B-4 (239).

Halting hunt (10-13): no halt

Each pick ran on an exact sped-up model, checked against direct simulation. - 686 (347's table): 3.5e26 steps. The closest approach to the margin was 0.989. - 362/433: past 10^5000 steps. The step counts follow a Collatz-like lottery, so it is now flagged as a cryptid candidate. - 244/216/701: about 10^(10^35) steps. The ranking assumed the one transition that differs from an upstream halter fires only at startup; it fires at every right-end event, so they do not run that halter's table. - 269 (R5): 1.5e77 steps (capped run ended). Its next chance to halt is near 10^(1.4e6) steps.

Astra's audit of the 22 unmerged kills

Formal: 22/22 pass check_kill.
coqchk: 17 complete; 5 hit the 600 s cap, a warning under policy A.
Public text: 9 PASS, 13 CONCERN. Astra's fixes for those 13 are written but not applied: coord/patches/astra-audit-public/README.md.

10-14 so far

Astra's audit patches are applied. The W5-5 header change gave it a new hash, and check_kill --halt passed three times on it.
693: stage 1 (catch-up law) and stage 2 (closed tape language) both parked under their stop rules. No busycoq decider settles it.
Site: the "How we proved it" text is rewritten for all 31 kills, with no fold.
Astra ran its three funded Bouncer studies (6/348, 282, 593). None found a proof, and none of these machines looks like a cryptid. Its new cards recommend a 60-minute stage on 593 and deferring 6/348 and 282.

What needs Dennis

Strategy directions 1-4 finished with 0 kills. - The 8 twin ports were refuted. - The halting search found no halt. - K42 is a cryptid family: park (rec), rules file, or family stage. See reports/k42/decision.md. - The CB3 core is bounded, not a cryptid: closure explorer (rec), hand it to Astra, or park. See reports/cb3/decision.md.
W6-1 (Decisions page): spike (rec), full formalization (about 14 x 2 h), or park.
Submit the 18 pending kills. Dennis will say when.
Commit: the last commit is 65ce973.