1RB1RF_1LC0RB_1RD0LC_1LE1RA_0RC1LB_---1RE
Status: CHECKPOINT_A (confidence low; proof work waits for Dennis) Class guess (agent / Dennis): nested binary counters made of 2-state defect walls in a period-5 sea, with epochs in which the engine dissolves (multi-region) / — Spreadsheet notes / Discord thread: Results tab "Unknown" (no status, no links, not struck through). Classifications tab "Unknown". Direct Simulation tab: listed as not simulated to 1e14 and not to 1e15. Discarded-machines tab: "Peacemaker shared its low cells visited" (that tab says "All of these machines are unanalysed"). It appears in neither Discord list (data/discord/), and nobody has claimed it. In all 26 holdout snapshots, including 2026-10-07_778 (line 661). Not SKIPPED. Viewer: https://bbchallenge.org/1RB1RF_1LC0RB_1RD0LC_1LE1RA_0RC1LB_---1RE Pool row: reports/targeting.md §4.4 R9 (rank 49 here).
Observation
- Direct simulation (tools/sim.py, C core): no halt to 1e10. Span -549..988 (1,538 cells, 949 nonzeros) at 1e9, and -729..1291 (2,021 cells, 1,224 nonzeros) at 1e10. Every edge visit adds 3 cells. Edge visits slow down: the left edge was visited about every 4e7 steps near 1.5e9 and about every 1.5e9 near 8e9 (work/records_1e10.jsonl).
- qsim (Quick_Sim, 3e6 loops = 5.4e6 steps): block size 1, 0 rules, no acceleration (work/qsim.jsonl).
- compress: whole tapes do not compress (median 54 tokens; reports/compress-sweep.tsv). The snapshots at 1e9 are long runs of (11100)/(00111)/(01110) with dozens of defects. rulefit-sweep: none (skeletons do not recur). rulefit was not run again.
- Sea. The tape is mostly the period-5 sea 11100. The two sweeps are exact (validated):
- S_R
B> (11100)^n -> (00111)^n B>, 7n steps; - S_L
(00111)^n <C -> <C (11100)^n, 7n steps.
So the sea is invariant: it shifts +2 under a right sweep and -2 under a left sweep. The head reverses only at defects.
- Turning points (work/scripts/turns.c, a C turn tracker with 8-cell hysteresis): 5,057,910 turns to 1e9. In the last 1e7 steps before 1e9, the turn counts per position halve exactly.
- Right side: R682 5,580, R735 2,790, R780 1,395, R808 698, R821 349, R864 174, R867 88, R875 44.
- Left side: L434 2,790, L340 1,395, L331 697, L317 349, ...
- Right-end region: a second pair (L929 344, L920 172, L911 86, L907 43, and R946 344, R954 172, ..., R988).
- The inner cycle is the 6-turn unit L482 R682 L443 R468 Lx Ry.
- Walls are bits (the windows at each turn; scratch tools work/scripts/.py):
- Right-wall bit, in the pre-right-sweep phase: insert 000 (bit 0) or 100 (bit 1) between 11100 blocks. 000 reflects and becomes 100 (RB0); 100 passes the sweep (RB1: 100 -> 111), and the returning left sweep turns 111 into 000 (RB1r).
- Left-wall bit, in the post-right-sweep phase: a run-2 defect (11 00 111, bit 1) or a gap-1 defect 3 cells to its right (111 0 111, bit 0). Bit 0 reflects a left sweep and becomes bit 1 (LB0, constructed context). The pass-through of bit 1 depends on the phase of the sea on its right, so it is not yet an exact rule.
- Each engine cycle increments both inner counters. Checked on the orbit after 1e9: right counter (8 bits, cells 680..875) 33,140/33,140 turns at the first 0 bit and 33,139/33,139 exact +1. Left counter (4 low bits, cells 315..435) 5,545/5,545 and 5,371/5,371.
- Above the inner counters (not understood).
- Inner right-counter overflows (≈ every 256 increments) send the head into a second engine/counter pair at the right end.
- Left overflows go to an edge structure where a wall drifts left 5 cells per visit.
- Engine-dissolving epochs: the inner engine dissolves; a wall walks left one block per bounce, eating a 4/5-density sea 11110 (made by D sweeps, D> (11100)^n -> (11101)^n D>) back into 11100; then the engine re-forms. There are 43 such epochs between 1e9 and 2.73e9, 2.9e7 to 5.6e7 steps apart, and none in the inner region after 2.73e9. 5 wall-walks happen near the right end between 3.7e9 and 4.6e9. At 9.9e9 the layout is engine (L824/L945, R787/R1050) + counters on both sides + right-end pair, all shifted from 1e9. This regime change is the main unknown.
- Halting.* F0 is the only undefined cell, and the entries into F, A and D are unique (A1, D1, C0). So a halt happens exactly when C reads 0 just before 110 (rule H, 4 steps). Concrete routes: B> 1^n 0 1 0 (a right sweep meets gap 1 + a lone 1), and a D sweep meeting 110. Neither is seen to 1e10.
Figures (img/): - records_1e10.png: edge records to 1e10. - counters_1e9_every8.png: every 8 steps from step 1e9, cells 250..1000. - turnmap_log_1e9.png: custom; turning points in 800 log-time bins to 1e9. - blaze_1e9.png - default_site.png
Candidate rules
All exact rules were checked by work/scripts/validate.py: random n in 1..40 and random 12-cell contexts on both sides, compared cell for cell including the step count.
| rule | from | to | steps | check |
|---|---|---|---|---|
| S_R | B> (11100)^n |
(00111)^n B> |
7n | 200/200 |
| S_L | (00111)^n <C |
<C (11100)^n |
7n | 200/200 |
| RB0 | (00111)^n B> 000 |
<C (11100)^n 100 |
7n+1 | 200/200 |
| RB1 | B> 100 |
111 B> |
5 | 200/200 |
| RB1r | (00111)^n 111 <C |
<C (11100)^n 000 |
7n+3 | 200/200 |
| LB0 | 00011 10111 (00111)^n <C |
00011 00111 (00111)^n B> |
14n+13 | 200/200 |
| D_R | D> (11100)^n |
(11101)^n D> |
5n | 200/200 |
| H | 0 <C 110 or C> 0110 |
halt | 4 | 400/400 |
| CR | inner right counter | binary +1, turn at the first 0 bit | – | 33,140/33,140 on the orbit |
| CL | inner left counter (4 low bits) | binary +1 | – | 5,545/5,545 on the orbit |
| EP | engine dissolves, wall walks left 5 per bounce | – | – | open (no exact rule) |
Template check:
- The nearest proved machines by table distance are 2 cells away: 1RB0RF_..._---0LE (CPS_LRU_v3.v:2133), ..._1RB1LD_---1RE (RWL_mod_v1.v) and ..._---1LB_---0RC (CPS_LRU_v4.v). All three are decider certificates; the first stops growing at 124 cells by 1e8.
- verify/CounterCmp_v1.v TM1 (the targeting row's template) compares unary counters, which is a different mechanism.
- Kill #12 is a recursive run-length counter over runs of 1s, which is a different tape language.
- No template applies.
Cryptid check (ground rule 6): no pseudo-random arithmetic seen; everything observed is binary counting or linear drift. The unknown epoch/bit-creation law could still hide parity branching, so it is not excluded, but it is not flagged either.
Checkpoint A (Dennis)
Open questions are on the card (machines/1RB1RF_1LC0RB_1RD0LC_1LE1RA_0RC1LB_---1RE/card.html). Priority call: spend one more stage on the epoch/bit-creation law, or park as STUCK at observe?
Informal proof
Not started (waits for Checkpoint A; confidence low).
Formalization log
—
Outcome
Open. Observe stage done 2026-10-10, about 1.5 h of the 2 h budget.
Decision (2026-10-10)
Question: Spend one more stage on the epoch/bit-creation law (wall catalogue plus an abstract wall simulator), or park R9 as STUCK at observe? Answer: park_sim (Park as STUCK at observe, plus one niced background run to 1e12 as a cheap halt check). Dennis: go with all recommendations (2026-10-10). Parked by Dennis 2026-10-10 (park_sim): niced background sim to 1e12 as a cheap halt check, work/sim_1e12.jsonl.
Background halt check (2026-10-10)
Direct simulation (C core) to 1e12 steps: no halt. At 1e12: state D moving R, head 1474, span -918..1666 (2,585 cells), 1,667 nonzeros (work/sim_1e12.jsonl). Extends the halting evidence from 1e10 to 1e12; status stays STUCK (parked).