Drop7 Research

Type to search every page, record and vocabulary term.

GitHubPlay

Theories

Registered falsifiable claims: each names a mechanism and the criteria that would refute it.

TitleStatusTierDateApproach
A fill-conditioned n-tuple leaf: bucketing every table by how full the board is, warm-started from the frozen tables, beats the frozen tables inside the same depth-3 search
mixed
tier: public-developmentRow and column lookup tables, learned from a billion moves
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
supported-as-tested
tier: public-developmentRow and column lookup tables, learned from a billion moves
Exact Rust leaf and transposition efficiency
mixed
tier: mechanics-onlyRust bitboard engine
A row-and-column n-tuple network trained by temporal-coherence TD on a billion engine moves beats the frozen fair leaf inside the same depth-3 search
supported-as-tested
tier: public-developmentRow and column lookup tables, learned from a billion moves
One bootstrapped refit of the linear fair leaf to the depth-4 search value gains points inside fair D4
not-supported-as-tested
tier: pilotOne-ply Q as a pruning prior and a refit leaf
A linear action value over one-ply features prunes fair expectimax without changing its decisions
mixed
tier: pilotOne-ply Q as a pruning prior and a refit leaf
Direct policy search over the six Klein-Friedmann feature weights beats their temporal-difference fixed point but stays far below fair depth-4
supported-as-tested
tier: pilotKlein-Friedmann linear Q-learning
Klein-Friedmann six-feature linear Q-learning transfers to corrected Hardcore as a survival heuristic, not a scoring policy
mixed
tier: pilotKlein-Friedmann linear Q-learning
A depth-5-distilled, whole-game-evolved NNUE leaf beats the frozen fair leaf inside the same depth-3 search
not-supported-as-tested
tier: public-developmentAn evolved NNUE leaf, distilled from a depth-5 teacher
One Rust search-matrix config reproduces depth, strata, and leaf comparisons locally and on EC2
supported-as-tested
tier: mechanics-onlyRust bitboard engine
A centralized deterministic frontier scheduler keeps deep fair search workers busy across columns
supported-as-tested
tier: mechanics-onlyRust bitboard engine
A nibble-packed column-major Rust engine reproduces fair expectimax bit-for-bit faster than the C++ fast engine
untested
tier: proposalRust bitboard engine
With reveal quadrature fixed at M=6, search depth is second-order for within-root sibling ordering (P-SOL-3)
untested
tier: proposalThe fast engine learns reveal sampling
Complete-sibling long-horizon outcome labels fix the learned leaf's within-root discrimination (P-SOL)
untested
tier: proposalSeparate reveal and next-disc sampling
Depth-3 factored reveal quadrature (N=7, M=6) beats fair depth-4 five-stratum search on the shared cohort
supported-as-tested
tier: public-developmentFair expectimax reference (D3/D4)
Oracle-visited public states carry transferable fair value (stage D0): the fair-relabelled lifetime of oracle-visited states exceeds that of matched fair-D4 states by a transferable fraction tau >= 0.25
not-supported-as-tested
tier: pilotPerfect-information oracle
A leaf-affordable NNUE-class student trained on successor-closed exact D4 sibling values reproduces D4's within-root ordering at the d4q gate thresholds
not-supported-as-tested
tier: pilotA learned survival estimate inside the reference search
Reveal construction: pricing the aligned same-wave double hit on a solid gray, and chains that terminate on a cover, raises reveal flow and lifetime
not-supported-as-tested
tier: public-developmentChain-reveal leaf substrate
A one-entry leaf memo across chance strata is a semantics-preserving speedup of the fast fair-D4 search
supported-as-tested
tier: mechanics-onlyRemembering the last board
Survival instinct: refusing to place a disc above its vertical-clear height extends life
mixed
tier: public-developmentSurvival instinct — never place a disc where it can't clear vertically
Entombed numbered discs are an early, unpriced death predictor for the fair depth-4 search
not-supported-as-tested
tier: mechanics-onlyEntombed discs: the 3 that can never clear
The NNUE-shaped survival leaf is accuracy-saturated at hidden width 64 on the existing corpus
supported-as-tested
tier: mechanics-onlyIs the small survival network as accurate as it can be?
The frozen fair-leaf weights are not locally optimal for the corrected-score depth-4 search they are deployed in
not-supported-as-tested
tier: public-developmentEvolving the leaf weights at the depth they are used
The fair search's depth gradient continues past the fourth ply once the chance estimator is exact
mixed
tier: public-developmentFast engine
Chance-node decorrelation and search depth compound rather than substitute
not-supported-as-tested
tier: public-developmentSeparate reveal and next-disc sampling
Search-guided self-play with every-sibling labels converts compute into long-horizon strength
untested
tier: proposalDistributional afterstate ranker
Successor-closed distributional afterstate values rank siblings as well as fair D4
not-supported-as-tested
tier: pilotDistributional afterstate ranker