cat / miau-is-true
clone

a browser grid puzzle game where you push words around to rewrite the rules of each level

Record solver routes and expansion baselines across worlds

Dami committed cce73d8
commit
cce73d8e9061566ca82dc7cc23d4ed7cc3d5acfe
parent
54b39f72d7344ac9928bcfc5c9f72f3f07abd15b
author
Dami <dami.etoile@gmail.com>
2 files changed, +16 -1
Mopenspec/changes/scale-solver-state-search/baseline.md15+++++++++++++++
Mopenspec/changes/scale-solver-state-search/tasks.md2+-
M openspec/changes/scale-solver-state-search/baseline.md +15 -0
@@ -11,6 +11,21 @@ Measured on the local development host with Node 26.8.1 after W6 and the restore
1111 | Tidewater / `low-tide` | 2,000,000 | solved, 10 steps | 0.17 s | 75.2 MiB |1212 | Clockwork / `mouse-hole` | 4,000,000 | solved, 23 steps | 9.39 s | 220.6 MiB |1313 14+For comparable search counts, a temporary copy of `tools/solve.mjs` counted each dequeued frontier state and attempted transition without changing its search logic. This follow-up measured RSS after every attempt; the per-proof peaks therefore differ slightly from the 20 ms sampled whole-command table above.15+16+| Pack / proof | Shortest path | Expanded | Attempted | Per-proof peak RSS |17+| --- | --- | ---: | ---: | ---: |18+| Core / `break-the-rule` | `UDDDRRRRRRRRR` (13) | 11,698 | 46,792 | 237.9 MiB |19+| Core / `make-room-to-push` | `RRRRRRRRR` (9) | 133 | 532 | 244.6 MiB |20+| Core / `two-things-can-win` | `RRRRRR` (6) | 80 | 320 | 245.1 MiB |21+| Word Lab / `shared-control` | `RRR` (3) | 16 | 64 | 55.2 MiB |22+| Word Lab / `change-what-you-are` | `RRRRRR` (6) | 12 | 48 | 55.8 MiB |23+| Word Lab / `rules-are-cargo` | `RRRRRRR` (7) | 54 | 216 | 63.7 MiB |24+| Crossword / `pass-it-on` | `URDDDRUUURULDLULURRRDDDDR` (25) | 72,523 | 290,092 | 342.9 MiB |25+| Crate Works / `night-shift` | `UUUURRRLLLDDDDRRRRUUUUUDLLLLUURRRRRRLLDDRRDDDRRR` (48) | 1,159 | 4,636 | 133.2 MiB |26+| Tidewater / `low-tide` | `DDUURRRRRR` (10) | 376 | 1,504 | 73.6 MiB |27+| Clockwork / `mouse-hole` | `URRRRRDRUUUUDDDDRDDDLLL` (23) | 12,265 | 61,323 | 195.2 MiB |28+1429 The W6 content review remains separate: `shove-the-danger` and `carry-the-pair` both reached LIMIT at 300,000 transitions before the worlds team's re-authoring. A LIMIT is inconclusive about solvability. The heavier Tidewater and Clockwork challenge cases have documented multi-million-state search needs and should be measured under a bounded memory budget after the compact codec is available.1530 1631 The first codec slice encodes the three core starting states in 151, 265, and 69 bytes, versus 1,842, 3,131, and 1,153 characters for their full JSON snapshots. These are payload sizes before JavaScript object, queue, and map overhead; the BFS rewrite must measure actual peak RSS separately.
M openspec/changes/scale-solver-state-search/tasks.md +1 -1
@@ -1,6 +1,6 @@
11 ## 1. Baseline and state inventory22 3-- [ ] 1.1 Record current solver route, expansion, time, and memory baselines for core, Word Lab, and representative Crossword, Crate Works, Tidewater, and Clockwork proofs.3+- [x] 1.1 Record current solver route, expansion, time, and memory baselines for core, Word Lab, and representative Crossword, Crate Works, Tidewater, and Clockwork proofs.44 - [x] 1.2 Inventory every field in settled engine snapshots and identify which fields determine future transitions, including automata phases and current transformation nouns.55 - [x] 1.3 Add a regression that restoring a settled state and taking one step matches direct play for noun chains, automata, removal, and rule edits.66