The frozen n-tuple tables as the leaf of the depth-4 fair search: a two-by-two leaf-by-depth screen on a fresh block

Candidate prior-d4s7: the frozen lookup tables of RUN-20260905T193006Z-4fbeb4e5 (layout rows,cols,win23,win32,phase=all, 10^9 f32 entries, SHA-256 0ade9d4e4080ebdd52a1474b1a13410dc8dfb77f5eba24b078aa7703c92ace0b, the candidate that passed the screens of EX-20260905-ntuple-scale-tc-td-leaf-d3-535b2620 and EX-20260906-ntuple-scale-replication-wide-plateau-f627f07a, unchanged) as the leaf of the Rust engine's completed full-width depth-4 fair expectimax search with seven chance strata, terminal utility -1,000,000, policy seed 0xd7075eed, work bound work_bound_for(4,7)+1 and a 1M-entry direct-mapped table (reference_d4_params, the configuration of the standing fair-d4s7 reference; Arm::NTupleD4 in approaches/ntuple-rl/ntuple-scale/src/game.rs).

completedtier: SCREENalgorithmicpublic-developmentpublic-policyEX-20260906-ntuple-scale-depth4-frozen-tables-54aed6a3

On this page

Linked approachRow and column lookup tables, learned from a billion moves

Theories

TH-20260906-ntuple-leaf-fourth-ply-search-compatible-ace4fe2a

Results

RS-20260906T171746Z-1623f833

Dates

Created: 2026-09-06T08:13:06Z
Updated: 2026-09-06T17:29:15Z

Record idEX-20260906-ntuple-scale-depth4-frozen-tables-54aed6a3

No explanation has been written for this record yet.

Technical recordThe registered protocolEX-20260906-ntuple-scale-depth4-frozen-tables-54aed6a3

Hypothesis: Candidate prior-d4s7: the frozen lookup tables of RUN-20260905T193006Z-4fbeb4e5 (layout rows,cols,win23,win32,phase=all, 10^9 f32 entries, SHA-256 0ade9d4e4080ebdd52a1474b1a13410dc8dfb77f5eba24b078aa7703c92ace0b, the candidate that passed the screens of EX-20260905-ntuple-scale-tc-td-leaf-d3-535b2620 and EX-20260906-ntuple-scale-replication-wide-plateau-f627f07a, unchanged) as the leaf of the Rust engine's completed full-width depth-4 fair expectimax search with seven chance strata, terminal utility -1,000,000, policy seed 0xd7075eed, work bound work_bound_for(4,7)+1 and a 1M-entry direct-mapped table (reference_d4_params, the configuration of the standing fair-d4s7 reference; Arm::NTupleD4 in approaches/ntuple-rl/ntuple-scale/src/game.rs). Comparator prior-d3s7: the identical tables as the leaf of the deployment depth-3 search (deployment_params, Arm::NTupleD3), the configuration both earlier screens evaluated. Two further arms on the same seeds complete a two-by-two leaf-by-depth design: fair-d3s7 and fair-d4s7, the frozen fair leaf in the same two searches. Hypothesis: prior-d4s7 beats prior-d3s7 in paired mean whole-game score on 512 never-read public-development games, and the fourth ply's paired gain on the tables is not smaller than its gain on the fair leaf on the same seeds. No training, no table is changed, and the screen is the only gameplay.
Arms: Arm Name Entry point Manifest
Candidate prior-d4s7 approaches/ntuple-rl/ntuple-scale/src/game.rs –
Comparator prior-d3s7 approaches/ntuple-rl/ntuple-scale/src/game.rs –
Classification: algorithmic
Information boundary: public-policy
Benchmark tier: SCREEN
Lifecycle: completed
Theories tested: The frozen row-and-column n-tuple leaf is search-compatible: a fourth completed ply of the fair search raises its mean score, by at least as much as it raises the fair leaf
Primary metric: paired mean whole-game score delta, prior-d4s7 minus prior-d3s7 (the frozen tables as the depth-4 leaf minus the same tables as the depth-3 leaf), 512 held-out games
Secondary metrics: persistence: paired mean score delta prior-d4s7 minus fair-d4s7 on the same 512 games, with the same bounds, halves, W-T-L and lower quartile
depth-step interaction: the per-game difference (prior-d4s7 minus prior-d3s7) minus (fair-d4s7 minus fair-d3s7), its mean, bootstrap and Student-t bounds, detection floor and W-T-L, and the fixed three-way verdict (larger / smaller / inconclusive)
the fair leaf's own depth step on these seeds: fair-d4s7 minus fair-d3s7; and the tables' depth-3 margin on these seeds: prior-d3s7 minus fair-d3s7, prior-d3s7 minus fair-d4s7, prior-d4s7 minus fair-d3s7
paired mean moves delta; numbered clears per move and cover reveals per move; mean occupied cells; lower quartile, median, maximum score; first-half (seeds 0-255) and second-half (256-511) paired mean deltas; per-arm logical work and wall seconds per game
resource observations: per-stage wall, CPU and peak resident set
Statistical unit: whole-game
Uncertainty method: one-sided 95% percentile bootstrap over whole games, 20,000 resamples, RNG seed 0xb0071eaf (the unchanged compare.py of the leaf-evolution screen, and analyze.py's identical paired() for the interaction series), plus a one-sided 95% Student-t lower bound; detection floor 1.645*sd/sqrt(n) reported for every contrast
Data role: public-development
Seed leases: SL-20260906T081306Z-1aaec419
Whole-origin split: yes
Reuse disclosure: CHECK gates read only the already-opened development probe block 0xa5276000-0xa5277fff (the Rust engine's gate range, SEEDLEASE-A52-FAST); no new seed is opened for mechanics. The held-out screen opens 0xa52f2580-0xa52f2780 (512 games) exactly once, after the frozen tables' SHA-256 has been verified on disk. That block is disjoint from every block the tables' experiment read (training 0xa5300000-0xa54f0000, validation 0xa52f2240-0xa52f2280, screen 0xa52f2140-0xa52f2240) and from every block of the replication (training 0xa5500000-0xa56f0000, validation 0xa52f2280-0xa52f2380, screen 0xa52f2380-0xa52f2580), so no arm has seen any seed of this screen. Nothing is trained or tuned here; the four arms are fixed before the block opens.
Pass criteria: All CHECK gates passed before the leased seed was read, on the frozen tables themselves (gate --weights): codec, row gather, feature indices for every layout, information-boundary blindness, reflection, direct-policy legality, leaf-in-search determinism and worker-count independence at depth 3 and at depth 4 (with the depth-4 arm doing more work than the depth-3 arm on every gate game), serial training determinism, finiteness.
The screen artifact has illegalDecisions 0 and incompleteDecisions 0 in every arm; censored games are reported as censored.
Held-out screen, 512 paired games on 0xa52f2580-0xa52f277f, prior-d4s7 vs prior-d3s7: bootstrap 95% lower bound of the paired score delta > 0 AND Student-t 95% lower bound > 0.
Held-out screen: prior-d4s7 minus prior-d3s7 paired mean score delta > 0 in both halves (seeds 0-255 and 256-511).
Held-out screen: prior-d4s7 lower-quartile score >= prior-d3s7 lower-quartile score.
Every table arm is the exact frozen best-weights.bin of RUN-20260905T193006Z-4fbeb4e5 (SHA-256 0ade9d4e4080ebdd52a1474b1a13410dc8dfb77f5eba24b078aa7703c92ace0b, verified and written to main/prior-weights.sha256 and main/candidate-weights.sha256 before the screen lease opens); no other table is screened.
Reported beside the gate, not part of it: persistence passes when prior-d4s7 vs fair-d4s7 meets the same four statistical criteria on the same 512 games; the depth-step interaction verdict is 'larger' when the bootstrap and Student-t 95% lower bounds of the per-game difference of depth steps are > 0, 'smaller' when its bootstrap 95% upper bound is < 0, and 'inconclusive' otherwise (floor reported).
On pass: Adopt prior-d4s7 (the unchanged tables at reference d4s7) as the candidate to carry forward; record the persistence and interaction readings; register a fresh-development STANDARD evaluation of the unchanged candidate as the successor; do not open protected or final seeds.
On fail: Record a valid run with scientific outcome fail for this exact configuration (these tables at reference d4s7); keep prior-d3s7 as the candidate to carry forward; report the persistence and interaction readings and the fair-leaf depth step as recorded; mark the theory not supported as tested (refuted if the primary falsifier's upper bound is below zero, inconclusive if the bounds straddle zero); open no further cohort for this candidate.
Gate fixed before controlled data: yes
Resources: Wall seconds 14400
CPU threads 32
Max host bytes 34359738368
Max GPU bytes –
GPU devices –
Stop conditions: Stop on any rules, information-boundary, legality, determinism, or parity failure.
A screen artifact with any illegal or incomplete decision voids the run (invalid), not the candidate.
The screen is evaluated exactly once; no re-run on the same or a different held-out block without a new experiment record.
The screen process is stopped at 14,400 s of wall time; such a run is recorded as interrupted with whatever arms completed, and the gate is not evaluated on a partial artifact.
Host memory above the declared bound (32 GiB resident) aborts the screen through the operating system or the operator; such a run is recorded as interrupted with its artifacts.
Expected artifacts: runs/<run-id>/ntuple-scale/gates.log (CHECK, on the frozen tables)
runs/<run-id>/ntuple-scale/main/{candidate-weights.sha256,prior-weights.sha256} (freeze: the same hash, the tables are not retrained)
runs/<run-id>/ntuple-scale/screen/heldout.json + compare-*.json (the one-shot screen, four arms)
runs/<run-id>/ntuple-scale/{pipeline.log,rusage.jsonl,screen.log,screen.err,analysis.json,analysis.md}
artifacts emitted by the screen carry this experiment id in their config field; the 4 GB table file is the retained artifact of RUN-20260905T193006Z-4fbeb4e5 and is not committed
Amendments: Timestamp Before controlled data Reason
2026-09-06T17:29:15Z no lifecycle advanced preregistered -> completed after run RUN-20260906T081306Z-e62d9837 and result RS-20260906T171746Z-1623f833 were written; protocol content otherwise unchanged (frozen hash 18733b1875bd5fe93f8625a22d617e6aa77a611b88b306d62c64e5d9196091bb retained in the run and lease records).

Arm	Name	Entry point	Manifest
Candidate	prior-d4s7	`approaches/ntuple-rl/ntuple-scale/src/game.rs`	–
Comparator	prior-d3s7	`approaches/ntuple-rl/ntuple-scale/src/game.rs`	–

Wall seconds	14400
CPU threads	32
Max host bytes	34359738368
Max GPU bytes	–
GPU devices	–

Timestamp	Before controlled data	Reason
2026-09-06T17:29:15Z	no	lifecycle advanced preregistered -> completed after run RUN-20260906T081306Z-e62d9837 and result RS-20260906T171746Z-1623f833 were written; protocol content otherwise unchanged (frozen hash 18733b1875bd5fe93f8625a22d617e6aa77a611b88b306d62c64e5d9196091bb retained in the run and lease records).

Technical recordResults recorded against this protocol1 record

valid runoutcome: passsupported-as-testedtier: public-developmentRS-20260906T171746Z-1623f833

Held-out screen, 512 never-read paired public-development games (0xa52f2580+), four arms on identical seeds, no training. The frozen tables (SHA-256 0ade9d4e4080ebdd52a1474b1a13410dc8dfb77f5eba24b078aa7703c92ace0b) as the leaf of the depth-4 seven-stratum fair search averaged 516,155 points and 148.75 moves against 465,675 points and 134.82 moves for the same tables as the depth-3 leaf: paired +50,479 points (bootstrap 95% lower bound +14,707, Student-t lower bound +14,086, upper bound +86,224, detection floor 36,331), W-T-L 284-0-228, halves +67,335 / +33,623, lower quartile 233,182 vs 212,969, moves +13.93. The preregistered gate PASSES. The fair leaf's own fourth ply on these seeds: fair-d4s7 382,567 against fair-d3s7 332,968, fair-d4s7 minus fair-d3s7: paired +49,599 (bootstrap 95% lower bound +25,751, Student-t lower bound +25,578, upper bound +73,244, detection floor 23,979), W-T-L 289-0-223, halves +45,189 / +54,009. Depth-step interaction, per game (tables d4 minus tables d3) minus (fair d4 minus fair d3): +881 (bootstrap LB -40,932, t LB -41,742, UB +42,867, floor 42,550, W-T-L 250-0-262): the preregistered verdict is 'inconclusive'. Persistence: prior-d4s7 minus fair-d4s7: paired +133,588 (bootstrap 95% lower bound +102,051, Student-t lower bound +101,161, upper bound +165,801, detection floor 32,371), W-T-L 314-0-198, halves +137,477 / +129,699. The persistence criteria PASS. The tables' depth-3 margin on these seeds: prior-d3s7 minus fair-d3s7: paired +132,707 (bootstrap 95% lower bound +102,193, Student-t lower bound +101,741, upper bound +163,903, detection floor 30,913), W-T-L 309-0-203, halves +115,331 / +150,084. prior-d4s7 minus fair-d3s7: +183,187 (LB +150,976). Logical work per game: prior-d4s7 795,922,316, prior-d3s7 22,193,672, fair-d4s7 558,924,581, fair-d3s7 15,568,331; mean wall seconds per game on the shared 32-thread workstation: 152.3, 4.1, 175.7, 4.5. Theory falsifiers: primary upper bound below zero False; second leg upper bound below zero False; persistence lower bound at or below zero False.

What it had to pass

✓All CHECK gates passed on the frozen tables before the leased seed was read (including leaf-in-d4-search-determinism) — observed: 19 gate lines
✓screen artifact: illegalDecisions 0 and incompleteDecisions 0 in every arm — observed: null
✓bootstrap 95% lower bound of prior-d4s7 minus prior-d3s7 > 0 — observed: 14706.909765625
✓Student-t 95% lower bound > 0 — observed: 14085.866195359107
✓paired mean delta > 0 in both halves — observed: [67335.2890625, 33623.49609375]
✓prior-d4s7 Q25 >= prior-d3s7 Q25 — observed: [233181.75, 212968.75]
✓Every table arm is the exact frozen best-weights.bin of RUN-20260905T193006Z-4fbeb4e5 (SHA-256 verified before the screen lease opened) — observed: 0ade9d4e4080ebdd52a1474b1a13410dc8dfb77f5eba24b078aa7703c92ace0b
✓Persistence (beside the gate): bootstrap 95% lower bound of prior-d4s7 minus fair-d4s7 > 0 — observed: 102050.977734375
✓Persistence (beside the gate): Student-t 95% lower bound > 0 — observed: 101161.09894605307
✓Persistence (beside the gate): paired mean delta > 0 in both halves — observed: [137477.09765625, 129699.0859375]
✓Persistence (beside the gate): prior-d4s7 Q25 >= fair-d4s7 Q25 — observed: [233181.75, 208481.25]
–Depth-step interaction verdict (beside the gate): larger / smaller / inconclusive — observed: {"verdict": "inconclusive", "meanDelta": 880.701171875, "bootstrapLower95": -40931.9, "bootstrapUpper95": 42867.4783203125}

Technical recordRecorded metricsRS-20260906T171746Z-1623f833

screen

games

512

seedStartHex

0xa52f2580

arms

prior-d3s7

meanScore: 465675.2422
medianScore: 352446.5000
q25Score: 212968.7500
minScore: 85,472
maxScore: 2,674,728
sdScore: 362210.7989
meanMoves: 134.8203
q25Moves: 65
numberedClearsPerMove: 2.0987
coverRevealsPerMove: 1.1851
meanOccupiedCells: 23.2526
censoredGames: 0
illegalDecisions: 0
incompleteDecisions: 0
meanWork: 22193672.2422
meanWallSecondsPerGame: 4.0569
games: 512

prior-d4s7

meanScore: 516154.6348
medianScore: 392496.5000
q25Score: 233181.7500
minScore: 85,682
maxScore: 2,716,683
sdScore: 402055.6663
meanMoves: 148.7461
q25Moves: 70
numberedClearsPerMove: 2.1255
coverRevealsPerMove: 1.2033
meanOccupiedCells: 23.0288
censoredGames: 0
illegalDecisions: 0
incompleteDecisions: 0
meanWork: 795922316.0430
meanWallSecondsPerGame: 152.3118
games: 512

fair-d3s7

meanScore: 332967.8516
medianScore: 262338.5000
q25Score: 174598.7500
minScore: 102,599
maxScore: 1,745,301
sdScore: 239092.2408
meanMoves: 97.7227
q25Moves: 55
numberedClearsPerMove: 2.0115
coverRevealsPerMove: 1.1225
meanOccupiedCells: 23.9550
censoredGames: 0
illegalDecisions: 0
incompleteDecisions: 0
meanWork: 15568331.0957
meanWallSecondsPerGame: 4.4931
games: 512

fair-d4s7

meanScore: 382566.5430
medianScore: 303,015
q25Score: 208481.2500
minScore: 85,801
maxScore: 2,634,604
sdScore: 273944.2597
meanMoves: 110.4453
q25Moves: 60
numberedClearsPerMove: 2.0459
coverRevealsPerMove: 1.1460
meanOccupiedCells: 23.8560
censoredGames: 0
illegalDecisions: 0
incompleteDecisions: 0
meanWork: 558924581.4180
meanWallSecondsPerGame: 175.7455
games: 512

contrasts

fair-d4s7-vs-fair-d3s7

score

n: 512
meanDelta: 49598.6914
pairedSd: 329836.4896
bootstrapLower95: 25751.4199
bootstrapUpper95: 73244.3458
studentTLower95: 25578.3583
detectionFloor: 23978.9201
wins: 289
ties: 0
losses: 223
firstHalfMeanDelta: 45188.8555
secondHalfMeanDelta: 54008.5273
q25Delta: 33882.5000
candidateQ25: 208481.2500
referenceQ25: 174598.7500

moves

n: 512
meanDelta: 12.7227
pairedSd: 89.7841
bootstrapLower95: 6.2207
bootstrapUpper95: 19.1624
studentTLower95: 6.1841
detectionFloor: 6.5273
wins: 279
ties: 18
losses: 215
firstHalfMeanDelta: 11.6719
secondHalfMeanDelta: 13.7734
q25Delta: 5
candidateQ25: 60
referenceQ25: 55

prior-d3s7-vs-fair-d3s7

score

n: 512
meanDelta: 132707.3906
pairedSd: 425213.8113
bootstrapLower95: 102193.1854
bootstrapUpper95: 163903.4781
studentTLower95: 101741.2067
detectionFloor: 30912.7957
wins: 309
ties: 0
losses: 203
firstHalfMeanDelta: 115330.6641
secondHalfMeanDelta: 150084.1172
q25Delta: 38,370
candidateQ25: 212968.7500
referenceQ25: 174598.7500

moves

n: 512
meanDelta: 37.0977
pairedSd: 117.1717
bootstrapLower95: 28.7245
bootstrapUpper95: 45.6993
studentTLower95: 28.5646
detectionFloor: 8.5183
wins: 309
ties: 13
losses: 190
firstHalfMeanDelta: 32.4453
secondHalfMeanDelta: 41.7500
q25Delta: 10
candidateQ25: 65
referenceQ25: 55

prior-d3s7-vs-fair-d4s7

score

n: 512
meanDelta: 83108.6992
pairedSd: 435893.1761
bootstrapLower95: 51348.7019
bootstrapUpper95: 114438.8502
studentTLower95: 51364.7909
detectionFloor: 31689.1793
wins: 282
ties: 0
losses: 230
firstHalfMeanDelta: 70141.8086
secondHalfMeanDelta: 96075.5898
q25Delta: 4487.5000
candidateQ25: 212968.7500
referenceQ25: 208481.2500

moves

n: 512
meanDelta: 24.3750
pairedSd: 119.4808
bootstrapLower95: 15.6656
bootstrapUpper95: 32.9825
studentTLower95: 15.6738
detectionFloor: 8.6862
wins: 277
ties: 16
losses: 219
firstHalfMeanDelta: 20.7734
secondHalfMeanDelta: 27.9766
q25Delta: 5
candidateQ25: 65
referenceQ25: 60

prior-d4s7-vs-prior-d3s7

score

n: 512
meanDelta: 50479.3926
pairedSd: 499739.6551
bootstrapLower95: 14706.9098
bootstrapUpper95: 86224.4920
studentTLower95: 14085.8662
detectionFloor: 36330.7810
wins: 284
ties: 0
losses: 228
firstHalfMeanDelta: 67335.2891
secondHalfMeanDelta: 33623.4961
q25Delta: 20,213
candidateQ25: 233181.7500
referenceQ25: 212968.7500

moves

n: 512
meanDelta: 13.9258
pairedSd: 137.7427
bootstrapLower95: 4.0720
bootstrapUpper95: 23.7852
studentTLower95: 3.8947
detectionFloor: 10.0138
wins: 278
ties: 14
losses: 220
firstHalfMeanDelta: 18.5117
secondHalfMeanDelta: 9.3398
q25Delta: 5
candidateQ25: 70
referenceQ25: 65

prior-d4s7-vs-fair-d4s7

score

n: 512
meanDelta: 133588.0918
pairedSd: 445272.9877
bootstrapLower95: 102050.9777
bootstrapUpper95: 165801.3641
studentTLower95: 101161.0989
detectionFloor: 32371.0861
wins: 314
ties: 0
losses: 198
firstHalfMeanDelta: 137477.0977
secondHalfMeanDelta: 129699.0859
q25Delta: 24700.5000
candidateQ25: 233181.7500
referenceQ25: 208481.2500

moves

n: 512
meanDelta: 38.3008
pairedSd: 122.5025
bootstrapLower95: 29.6404
bootstrapUpper95: 47.2169
studentTLower95: 29.3795
detectionFloor: 8.9059
wins: 313
ties: 16
losses: 183
firstHalfMeanDelta: 39.2852
secondHalfMeanDelta: 37.3164
q25Delta: 10
candidateQ25: 70
referenceQ25: 60

prior-d4s7-vs-fair-d3s7

score

n: 512
meanDelta: 183186.7832
pairedSd: 452824.9076
bootstrapLower95: 150975.5535
bootstrapUpper95: 216399.5734
studentTLower95: 150209.8220
detectionFloor: 32920.1063
wins: 345
ties: 0
losses: 167
firstHalfMeanDelta: 182665.9531
secondHalfMeanDelta: 183707.6133
q25Delta: 58,583
candidateQ25: 233181.7500
referenceQ25: 174598.7500

moves

n: 512
meanDelta: 51.0234
pairedSd: 124.9335
bootstrapLower95: 42.1543
bootstrapUpper95: 60.1797
studentTLower95: 41.9252
detectionFloor: 9.0826
wins: 343
ties: 8
losses: 161
firstHalfMeanDelta: 50.9570
secondHalfMeanDelta: 51.0898
q25Delta: 15
candidateQ25: 70
referenceQ25: 55

depth

primary

prior-d4s7-vs-prior-d3s7

tablesStep

meanDelta: 50479.3926
bootstrapLower95: 14706.9098
bootstrapUpper95: 86224.4920
studentTLower95: 14085.8662
detectionFloor: 36330.7810
wins: 284
ties: 0
losses: 228
firstHalfMeanDelta: 67335.2891
secondHalfMeanDelta: 33623.4961
candidateQ25: 233181.7500
referenceQ25: 212968.7500

fairStep

meanDelta: 49598.6914
bootstrapLower95: 25751.4199
bootstrapUpper95: 73244.3458
studentTLower95: 25578.3583
detectionFloor: 23978.9201
wins: 289
ties: 0
losses: 223
firstHalfMeanDelta: 45188.8555
secondHalfMeanDelta: 54008.5273

persistence

checks

criterion
screen artifact: illegalDecisions 0 and incompleteDecisions 0 in every arm
passed
true
criterion
bootstrap 95% lower bound of prior-d4s7 minus fair-d4s7 > 0
passed
true
observed
102050.9777
criterion
Student-t 95% lower bound > 0
passed
true
observed
101161.0989
criterion
paired mean delta > 0 in both halves
passed
true
observed
1. 137477.0977
2. 129699.0859
criterion
prior-d4s7 Q25 >= fair-d4s7 Q25
passed
true
observed
1. 233181.7500
2. 208481.2500

passed

true

interaction

verdict: inconclusive
n: 512
meanDelta: 880.7012
pairedSd: 585280.6111
bootstrapLower95: -40931.9000
bootstrapUpper95: 42867.4783
studentTLower95: -41742.3429
detectionFloor: 42549.5586
wins: 250
ties: 0
losses: 262
firstHalfMeanDelta: 22146.4336
secondHalfMeanDelta: -20385.0313

theory

primaryFalsifierUpperBoundBelowZero: false
secondLegUpperBoundBelowZero: false
persistenceLowerBoundAtOrBelowZero: false

tablesSha256

0ade9d4e4080ebdd52a1474b1a13410dc8dfb77f5eba24b078aa7703c92ace0b

Limitations

Public-development SCREEN tier, 512 paired games opened once; nothing here is a qualification claim, and protected and final cohorts stay sealed.
No table was trained or changed: both table arms are the first experiment's frozen candidate, selected there at the best of twenty training-role validation points; this screen inherits that selection but adds none of its own.
The depth-4 search is the standing reference configuration (1M-entry table, seven strata, terminal utility -1,000,000); a different terminal utility, table size or stratum count is a different configuration and is not tested here.
The fair-d3s7 and fair-d4s7 arms are context for the interaction reading; the preregistered comparator is the identical tables at depth 3.
Wall times were measured on a shared workstation with all four arms run in sequence at 32 threads; logical work and the ratios between arms on the same seeds are the trustworthy cost quantities.
The per-game artifact is cited by its public archive reference (data.drop7.dev, immutable run-scoped key, digest in the fragment); a byte-identical copy remains under runs/RUN-20260906T081306Z-e62d9837/ntuple-scale/screen/heldout.json on the workstation and a promoted copy under artifacts/results/.

Open the result record

Agent contextHow to extend this record

To add a reader-facing explanation, write web/content/research/EX-20260906-ntuple-scale-depth4-frozen-tables-54aed6a3.mdx; it renders above this record on the next request. The registered protocol itself is in the technical record above.

Record file: research/experiments/EX-20260906-ntuple-scale-depth4-frozen-tables-54aed6a3.json, validated against research/schemas/experiment-v1.schema.json. Protocol hash: 163a44c36910ddd0c242512600c9bc35bfd3a76a8e12f2f4b3da622ffb822ab8.

Registered by Claude Code / claude-fable-5-1 (claude-code-ntuple-continuation-4-ply).