/machines/1RB0LF_1LC0LA_---1RD_1RF1RE_1LF1RD_1LB0RE/notes.md · raw

1RB0LF_1LC0LA_---1RD_1RF1RE_1LF1RD_1LB0RE

Status: CHECKPOINT_A (confidence low; proof work waits for Dennis, see decision.json) Class guess (agent / Dennis): Bouncer + Chaotic Edge (sheet agrees). Exact bouncer + 7-bounce junction + chaotic edge; the edge is #13's edge, cell for cell. / Dennis: open Spreadsheet notes / Discord thread: results sheet: "Bouncer + Chaotic Edge", no claim. Discord: only on mxdys's UBRRBA_TODO list (2026-08-24); Dennis: fair game. No claim in data/discord or data/sheet. Byte-exact on data/holdouts/2026-10-07_778.txt (line 567). Viewer: https://bbchallenge.org/1RB0LF_1LC0LA_---1RD_1RF1RE_1LF1RD_1LB0RE Pool: reports/targeting.md section 4.3, #35. Near-twin #36 1RB1RF_1LC0RF_1LE0LD_1RC0LB_---1RA_1LB1RA is observed by another agent.

Observation

Simulated 1e9 steps from blank: no halt, span -14842..6905 (21,748 cells), 15,177 nonzeros, ends in E. Left edge ~ -0.47 sqrt(t), right edge ~ +0.22 sqrt(t). qsim (block 3) does not accelerate it (2 rules in 2e5 loops).

Structure, all in left-edge coordinates (one left record per bounce):

0^inf <A 0 (1101101101)^n H_k Z 0^inf        k = bounce index mod 7, Z = the edge
  1. Bulk and wall. At every left record, the tape starts 0^inf <A 0 (1101101101)^n. The wall: 0^inf <A 0 r -> 0^inf 1 1 E> r (4 steps). Right sweep: E> (1101101101)^n -> (1011110110)^n E> (12n). Left sweep: (0111101101)^n <A -> <A (0110110111)^n (50n), which writes the bulk back shifted one cell left. So one bounce costs 62n + c steps and leaves the bulk unchanged relative to the wall.
  2. Junction (exact, period 7 bounces). After the bulk come seven fixed words: H_0 = 10010010011001001001001101101100111011001001101101110100110 (59 cells) H_1 = 110110010011001001001001101101100111011001001101101110100110 (60 cells) H_2 = 1011101001101100100111011001001001111011001001101101110100110 (61 cells) H_3 = 11010011011001001100111011001001001111011001001101101110100110 (62 cells) H_4 = 110110110010011001100111011001001001111011001001101101110100110 (63 cells) H_5 = 100111001100111011001001001111011001001101101110100110 (54 cells) H_6 = 1101111001100111011001001001111011001001101101110100110 (55 cells) Bounces k = 0..5 map H_k to H_(k+1) and never read Z (60/60 random n, random Z each); k = 4 moves one block into the bulk (n+1). Steps: 62n + c_k, c = 13, 433, 55, 49, 73, 13. In absolute cells the junction moves +3 per 7 bounces, the bulk gains one 10-cell block per 7 bounces. Checked on the real run: at every edge visit from bounce 301 to 14,840 (2,078 visits) the junction sits at absolute cell 6165 + 3k and the 113 cells around it are identical.
  3. Edge visit (k = 6). 0^inf <A 0 (1101101101)^n H_6 Z -> 0^inf <A 0 (1101101101)^n H_0 Phi(Z). Phi does not depend on n: running each real Z at n = 3 reproduces the next real Z at all 2,077 visits of the 1e9 run. Phi walks into Z in many small round trips (the crossing sequence at the H_6|Z cut varies: 28-32 crossings), extends the right end by 30 cells (sometimes 60) when it reaches it, and has no closed form.
  4. Edge statistics. Iterating Phi alone (C, scripts/phi2.c, from the real Z at bounce 14,840) for 5,000,000 visits (about 5e15 machine steps): |Z| stays in 143..563, no exact repeat of Z (hash table), no halt. k-gram counts in Z grow exponentially (12-grams: 607, 24-grams: 22,839), so Z has positive entropy; it is chaotic, not a counter.
  5. Same edge as #13. #13 1RB0LE_1LC1RD_1LA0RB_1RC1RB_1LF0LC_---0RC (parked 2026-10-10 on its edge lemma E') has the same bulk word and a 54-cell junction X0. Its tail at bounce c-6 equals the fixed word 0010010011001001001001101101100111011001001101101110100110 followed by our Z at bounce c, at all 1,952 overlapping edge visits (#13's work/tails_1e9.jsonl). So the chaotic edge is literally the same object, and its |D| range 144-564 matches ours. A closed language for that edge would serve #13, #35 and probably #34, #36 and R6.
  6. Halting. C is entered only by B0 (1LC). The halting routes are 0 0 0 E> 0 (E0, F0, B0, C0) and 0 0 0 1 <A (A1, F0, B0, C0). To 1e9 there are 38,720 near misses 1 0 0 E> 0, all inside the edge, and none of the A route. At left records the tape has no 000 inside the span (zero runs are 1 or 2).
  7. Early regime. Bounces 0..300 use a period-5 junction (gaps of 5 between edge visits). Only plain simulation covers it (bounce 301 is step ~7.3e5).

Diagrams (img/): default_1e5.png (site view), records_1e7.png, zone_every28.png (custom: every 4th edge visit in the junction frame), cycle1_explore.png (one bounce, constructed start), cycle7_every5.png (7 bounces). Viewer: img/viewer.html.

Candidate rules

All validated by direct simulation (scripts/val_basic.py, scripts/validate.py, scripts/phicheck.py; results in work/): - wall 0^inf <A 0 r -> 0^inf 1 1 E> r, 4 steps: 80/80 random r. - right sweep E> (1101101101)^n -> (1011110110)^n E>, 12n: 80/80 (random n <= 200, random contexts). - left sweep (0111101101)^n <A -> <A (0110110111)^n, 50n: 80/80. (The first form tried, (1011110110)^n <F -> <F (1011011011)^n, depends on the right context: 1/80. Do not use it.) - junction bounces k = 0..5: 360/360 (60 per k). - edge visit (k = 6) at n = 3 vs the real run: 2,077/2,077. - 7-bounce cycle on the real run: 2,078/2,078 (frame formula and gadget words). Phi itself is not a rule; it is the open part.

Template check: verify/Bouncer_v3.v TM1 has the same shape (sideRLs_wall at the left, everything else as the right side, BigStep over 128 bounces checked by vsideRLs_es_spec + native_check_eq on RC i -> RC (1+i)). That check needs the right side to be eventually periodic in i; ours is chaotic, so the RInc step has no analogue. verify/InfiniteRect_v1.v needs eventually periodic crossing streams per column; the edge columns never settle. A closed-language certificate (verify/CTLCertv1.v, solve_cert from CTL62) is the only formal route seen; mxdys's standard CTL ranges already failed on the holdout list. Cryptid check (ground rule 6): no arithmetic in the rules (no exponent parities, no Collatz-like branching). The open part is a chaotic symbolic map on a bounded-looking window. Not a cryptid by the definition, but just as hard in practice: the same edge lemma blocked #13.

Checkpoint A (Dennis)

(open) See card.html and decision.json. Recommendation: park with #13.

Informal proof

Not started (confidence low; waits for Dennis). If funded, it would be #13's conditional reduction with this machine's bulk and junction words.

Formalization log

None.

Outcome

Open (CHECKPOINT_A).

Reproduce

Scratch scripts are in scripts/ (paths inside point to the session scratchpad; the 165 MB per-bounce tape dump snaps_all.txt is not kept, scripts/snap.c regenerates it in ~10 s): - gcc -O2 -o cyc scripts/cyc.c && ./cyc 1000000000 gives per-bounce stats (work/cycles_1e9.txt). - scripts/snap.c dumps the tape at every left record; work/iface_1e9.txt has the junction position per bounce. - scripts/phi2.c iterates the edge map (input work/phi_init.txt); work/phi2_5e6.txt is the 5e6-visit log.

Decision (2026-10-10, auto-applied)

Question: #35's chaotic edge is cell for cell #13's parked edge: park #35 with #13, or fund a shared stage on that edge? Answer: park_with_13 (Park with #13: mark STUCK at observe and link the two). auto-applied: Claude's recommendation, under Dennis's standing instruction 'do your recommendations and continuously orchestrate' (2026-10-10, while he was at lunch). Override any time. Parked 2026-10-10 with #13 (auto-applied recommendation): its chaotic edge is cell for cell #13's edge lemma (E').