Move 37 answered a Go position with one coordinate no human expected. Move42 asks the same question in a larger action space: when the board is a requirement, can the move be a complete executable algorithm?
A requirement is a board. The contract defines legal play. The referee decides what worked.
01 · Same abstraction, different action space
State becomes a move; consequence becomes experience.
A requirement is a board. The contract defines legal play. The referee decides what worked.
state → legal action → external consequence → preserved experience.
The proposed output changes from an answer or score to an independently runnable method. The learning target is the distribution of later executable proposals across requirements.
02 · Different output
An independently runnable method.
The proposed output changes from an answer or score to an independently runnable method. The learning target is the distribution of later executable proposals across requirements.
The claim is a shift in output type and learning target—not merely a better prediction, benchmark score, or task-time loop.
03 · AutoResearch contrast
Where does search occur, and what persists?
Where search occurs
Search now. Learn for later. Or combine both.
Current requirement
AutoResearch improves through a proposal–execution–inspection–retry loop on the current requirement.
Across requirements
Move42 aims to improve the proposal distribution across requirements, so the first executable move itself becomes learnable.
Hybrids remain possible; the distinction is where search occurs and what persists between tasks.
Evidence boundary. The current record does not measure superiority over research agents, task-time speedup, or an admitted prospective strict first move.
04 · Five-part blueprint
Board. Move. Rules. Referee. Memory.
01
Board
The requirement plus every permitted input, state transition, and resource boundary.
Encode the requirement and permitted state before any proposal is made.
02
Move
One complete executable algorithm whose outputs can be independently reproduced.
Define the runnable artifact, interface, and terminal outputs that count as a move.
03
Rules
The legality contract that rejects shortcuts, leakage, hidden state, and invalid resources.
Close shortcut and leakage paths before play, then make every violation an explicit failure.
04
Referee
An external consequence-based evaluator that executes the move and records success or a named failure.
Build the referee outside model self-assessment and bind its identity to every outcome.
05
Memory
Canonical success and failure receipts that can supervise later proposals across requirements.
Preserve receipts, train later proposals, and separately freeze a true one-proposal prospective evaluation.
05 · Current evidence
353 standalone algorithms beat TabFM Ensemble.
They came from 1,649 eligible historical cases; 228 beat all four stored foundation configurations. The released 600-case atlas contains all 353 wins plus 247 outcome-ranked losses.
353 recorded wins · 247 recorded losses · all 600 rows visibleSource: formal atlas and aggregate-selection receipts. Historical, outcome-ranked, not a prospective win-rate study.
228 winning workflows completed within twenty seconds.
The winning median fit-and-predict time was 0.706549 seconds.
239 wins fit and predicted within one second; 320 within two seconds.
39 wins were at least two times lower RMSE, 18 at least five times, and 12 at least ten times versus TabFM Ensemble.
Boundary: historical, outcome-ranked, not a prospective win-rate study. Foundation comparator wall-clock was not receipted.
06 · Concrete algorithmic moves
CPU, greenhouse, and spectrometer.
Cpu
19.527s workflow · 17.578s search · 0.480s fit+predict · 5.13× lower recorded RMSE vs strongest stored foundation.
Greenhouse
19.614s workflow · 15.955s search · 1.224s fit+predict · 3.40× lower recorded RMSE vs strongest stored foundation.
Spectrometer
100.180s workflow · 20.440s search · 0.563s fit+predict · 4373.89× lower recorded RMSE vs strongest stored foundation.
Spectrometer is the quality outlier: 4373.89× lower recorded RMSE, 0.563-second fit-and-predict, and a 100.180-second complete workflow. Exact programs remain in marked verbatim-evidence disclosures on the atlas.
07 · Receipts become memory
Successes and failures train later proposals.
The external referee preserves the program, requirement, split, resources, outcome, clocks, and named failures. Stored supervision can update a later challenger; gradients do not pass through arbitrary executed code.
08 · What does this mean?
Four design horizons, all labeled vision.
Vision · Compiler game
Board
A source program, target semantics, architecture, and optimization contract.
Move
A complete transformation or optimization algorithm that produces executable output.
Referee
Correctness suites, resource limits, and consequence-based performance measurements.
A design horizon, not a finding of the current regression study.
Vision · Control game
Board
A control requirement, observable plant state, hard constraints, and permitted actuators.
Move
A complete control algorithm that can be executed against the declared interface.
Referee
Constraint violations, stability, resource use, and measured physical consequence.
A design horizon, not evidence of autonomous control performance.
Vision · Experimental game
Board
An apparatus, measurement protocol, budget, safety envelope, and experimental objective.
Move
A complete executable experiment schedule and analysis method.
Referee
Predeclared measurements, explicit failure outcomes, and independently preserved receipts.
A design horizon, not an experimental result reported here.
Vision · Scientific-design game
Board
A scientific question, admissible evidence, interventions, and falsification criteria.
Move
A complete executable design for collecting and analyzing the next evidence.
Referee
Preregistered consequence tests that can reject as well as support the proposed method.
A design horizon, not a claim of causal or scientific discovery.
09 · Scale, estimate, limits, next study
375,522 processed campaign rows were observed.
The recorded campaign spans 36 generations and 7 retained promotions. Those are work records, not evaluated-program receipts.
Forward-looking company estimate · unobserved
Approximately 1,000,000,000 evaluated program receipts.
Not a finding of this study. Generated or merely compiled programs do not count.
The prospective first-move study is still required.
Historical top-one results did not establish strict one-proposal performance. A future study must predeclare the cohort and proposal budget, retain complete failures, match hardware and timing, and freeze comparators.
Limitations: no expected future win rate, focal-case causality, matched comparator speed, universal superiority, or safe autonomous use is established.