Crafter Research · Frontier map
The frontier, classified.
A problem is on this map only if it has an oracle. The axes below say how much each future result can be trusted, what it costs, and why we would be the ones to do it. Rows with no edge stay visible on purpose.
A is machine-checked with no human judgment. B is statistical, held-out, or prospective. C needs a person or fieldwork.
Tractable now, hard, or a long bet. Long bets publish their stepping stones.
Laptop, modest cloud, or out of reach. We do not pretend to have a cluster.
Place, language, our own instruments, or none.
Empty, open community, or a funded race.
Catalogued, belief stated, oracle built, survived, killed, published. Killed beliefs stay listed.
81 / 81 shown
Mathematics
-
M1Erdős problems in LeanLean kernel
-
M2Ramsey number R(5,5), now 43 to 46Exhaustive checker
-
M3Small off-diagonal Ramsey cellsExplicit graph and checker
-
M4Constant-weight code bounds behind kissing numbersThe code itself
-
M5Kissing numbers in low dimensionsExplicit configuration
-
M6Small matrix-multiplication rankTensor identity check
-
M7Record tables: covering designs, packings, Golomb rulersThe object itself
-
M8Kernel-checked certificates for existing table entriesLean or Rocq kernel
-
M9Counterexamples to last month's arXiv conjecturesExhaustive checker
-
M10Unknown unknotting numbers and slice statusExplicit sequence or invariant
-
M11OEIS: new terms, b-file errors, conjecturesRecompute, Lean for proofs
Theoretical CS
-
C1Busy Beaver BB(6) holdouts, one machine at a timeRocq or Lean decider
-
C2Optimal sorting networks, n = 17Exhaustive 0-1 check
-
C3Minimal circuits for S-boxesExhaustive equivalence
-
C4Schur and van der Waerden lower boundsThe colouring
-
C5Agent-evolved SAT heuristicsSAT Competition set and DRAT proofs
-
C6Weakly solving small gamesProof-tree checker
Systems and developer tools
-
A1Verified superoptimization of hot loopsSMT equivalence, differential fuzz, benchmark
-
A2Cross-engine differential testing of the web platformEngine disagreement and a reproducer
-
A3Terminal emulator conformanceCross-emulator disagreement
-
A4Compiler miscompiles and missed optimizationsAlive2 and crash reproducer
-
A5Database logic bugs by metamorphic testingDifferential result
-
A6Byte-matching decompilationByte-identical rebuild
-
A7Fixing non-reproducible buildsIdentical hash
-
A8Rust standard library verificationKani or Verus
-
A9Hutter Prize text compressionPublic leaderboard, CPU rules
-
A10Which CLI designs agents operate reliablyHeld-out task success across models
-
A11C to Rust translation with equivalenceTests and differential fuzz
Security
-
S1Autonomous vulnerability discovery in open sourceSanitizer crash reproducer
-
S2The unaudited LatAm long tail: e-invoicing, tax, paymentsReproducer and coordinated disclosure
-
S3Verified patchesPoC dies, tests and fuzz pass
-
S4Prompt injection and exfiltration in browser agentsCanary exfiltration reproducer
-
S5Protocol state machines with model checkingTLA+ or Tamarin counterexample
-
S6Predicting npm supply-chain malwareLater advisory or takedown
-
S7Under-constrained zero-knowledge circuitsTwo valid witnesses
Cryptanalysis and chains
-
K2Mid-dimension lattice records via better algorithmsSVP Challenge leaderboard
-
K3Automated trail search across lightweight ciphersSolver-verified trail, practical attack
-
K4Keccak reduced-round challengesThe collision or preimage
-
K5Encrypted dispatches of the Viceroyalty of PeruRecovered key decrypts held-out letters
-
K6Constant-time checks of post-quantum implementationsdudect and ctgrind
-
B1Spanish-language crypto Ponzi schemes before the regulator's alertLater regulator alerts
-
B2Wash trading and MEV measurementLabeled sets
Physics
-
P1New periodic three-body orbitsNumerical integrator
-
P3Singularity candidates in Euler and Boussinesq flowsInterval-arithmetic proof
-
P4Audit AI-discovered quantum codesA lower-weight logical operator
-
P5Quantum circuit T-count recordsEquivalence check
-
P6Variational ground-state energiesLower energy wins
-
P7Best-known Max-Cut, spin-glass, and packing instancesThe configuration
Earth science
-
E1Peru's deep seismic catalogueRe-detection at independent stations
-
E2Slow slip and tremor on the Peruvian flat slabTemplate stacking
-
E3Coastal El Niño forecastingHindcast and prospective skill
-
E4AI weather models against Andean stationsHeld-out station data
-
E5Glacial lake outburst risk in the Cordillera BlancaHeld-out inventories
-
E6Anchoveta biomass forecastingProspective surveys and quotas
-
E7Dengue forecasting by departmentProspective weekly counts
-
E8Illegal mining and deforestation detectionHeld-out labels and later imagery
Astronomy
-
R1Anomalies in the Rubin alert streamCross-match and follow-up
-
R2Asteroid discovery from archival detectionsMinor Planet Center designation
Decipherment and language
-
D1Khipu: predict held-out cordsHeld-out khipus against a null
-
D2Khipu digitization from photographsAgreement with expert records
-
D3Linking khipus to colonial recordsNumeric match against a null
-
D4Reading colonial notarial handwritingCharacter error on held-out pages
-
D5Reconstructing proto-Quechua formsLeave-one-language-out prediction
-
D6Quechua and Aymara language modelsAmericasNLP leaderboards
-
D7Accounting scripts with sum checks: Linear A, Proto-ElamiteSums as held-out checks
-
D8Vesuvius Challenge scrollsPrize review
-
D9Archaeological site detection in PeruHeld-out registry sites and field checks
Science of AI
-
I1False negatives in agent verdictsPlanted ground truth
-
I2Claim-to-artifact audit of AI-written papersRe-execution
-
I3Broken or leaky tasks in agent benchmarksReproducer per task
-
I4Same model, different harnessFixed task suite
-
I5Interpretability on tiny models with planted circuitsCausal intervention against known truth
-
I6AI scientists on messy Spanish-language dataReproduction on public microdata
-
I7Arithmetic-impossibility checks in regional journalsArithmetic
-
I8Verified results per token, published as a curveThe lab's own oracles
Where we do not compete
Wet labs, materials and DFT, model training, drug discovery, and headline races on formal math already funded by large labs. We contribute pieces there only when an open community makes it useful.
Social science
X1Later addenda, terminations, sanctions
X2The release itself
X3Re-execution on public microdata
X4Re-execution
X5Held-out votes
X6Held-out survey marginals