Changelog¶
All notable changes to LeafSleuth are documented in this file.
[Unreleased]¶
[0.1.0] - Unreleased¶
Added¶
- Pure-Java
LagAttributionEnginecombiningWorkAttributionLedger,EwmaEstimator, and sliding-window marginal covariance regression (REQ-SLEUTH-LAG-001) with bystander immunity. LagPatternRecognizercomputing bounded lag-$k$ autocorrelation ($k \in [1, 40]$ ticks) and inter-arrival coefficient of variation to classifyPERIODIC_CLOCK,ENTITY_CRAM_BURST,CHUNK_THRASH, andSTEADY_LOAD(REQ-SLEUTH-LAG-002).SpatialAttributionGridexportingHeatmap2Drasters (meanMsptHeatmap,upperCi95Heatmap),SpatialCellMetadatahit-test overlays, andConfigRemediationBridgeconfig recommendations formaps-apiandLeafConfig(REQ-SLEUTH-MAP-001,leafsleuth-ADR-003).- 4-pillar Bayesian behavioral & stylometric identity profiler (
LinguisticStylometryProfiler,MotorKinematicsProfiler,SimultaneousActivityCorrelator,GoalMarkovProfiler,BayesianIdentityLinker,ModeratorInvestigationCase) reporting $P(\text{SamePerson} \mid E) \pm 95\%\text{ CI}$ and rankedExplainableTellevidence markers (REQ-SLEUTH-BIO-001..005). RelativeCoordinateSpaceandPlayerRelativeEmbedding(leafsleuth-ADR-004) projecting intra-player relative feature proportions onto population-centered unit hypersphere coordinates and 64-bit CharikarSimHashLSH bands for sub-quadratic player graph scaling.- Exact inverse-cubic Minecraft mouse sensitivity slider inversion (
0–200%), multi-sample GCD harmonic de-aliasing, quantization residual verification, compose-time linear regression (EwmaLinearFit), IDF-weighted character trigram Cosine Delta, decaying chunk builder provenance, and JVM GC pause subtraction. - Unified causal server forensics engines (
leafsleuth-ADR-005):ConservationOfMassAuditorandItemProvenanceLedger(economy conservation-of-mass and duplication auditing),SupplyChainRingForensics(alt-ring and resource laundering graph forensics),ResourceExhaustionForensicsandResourceCulpritReport(heap/ticket/chunk exhaustion attribution), andInteractionAndProbeForensics(action-entropy macro and sub-tick packet probing). SyntheticSleuthSimulatordeterministic trace and persona simulation harness for statistical calibration and false-positive verification.SleuthMenuBuilderdark-on-parchment Book/Chat interactive reports (REQ-SLEUTH-UI-001),SleuthMapService,SleuthExtensionProvider/SleuthExtensionRegistryServiceLoaderSPI (leafsleuth-ADR-006),SleuthBStats(PaperSleuthBStatsHost,SleuthBStatsCatalogue) telemetry,[LeafSleuth bit]self-test suite, and Paper/Folia non-blocking runtime collectors (S-005).- Contraption-safe Pareto $L_{95}$ closed-loop governor (
CausalAutoTuner,SleuthMitigationActuator,MechanismCausalLinker) with specialist hopper protection, 40-tick piston relocation immunity, stationary redstone edge queue and replay, andGovernorOperatingMode(OBSERVE,GOVERN_SAFE,GOVERN_BLACKBOX). EmpiricalFarmBlackBoxforward-tracked causal provenance DAG (MOB_NATURAL_SPAWN,CROP_NATURAL_GROWTH,BLOCK_GENERATOR_FORM-> observer/piston -> entity fall/death -> item spawn -> hopper -> dropper -> water stream -> terminal sink) with bidirectionalProvenanceTaintpropagation, dual-radius Schmitt-trigger proximity unpause (wakeRadiusBlocks: 28.0,sleepRadiusBlocks: 40.0,rehibernateHysteresisTicks: 200), and optional conservative $L_{95}$ chunk-unload catch-up (unloadCatchUpEnabled: false) with 400-tick anti-thrash reload cooldown.- Permission gating (
sleuth.admin.benchmark) for internal benchmark/harness subcommands (/sleuth governor sim|pareto|restore|paste|randomtick) and storefront documentation (docs/publishing/FRONT_PAGE.md). - Active counterfactual micro-dithering (
TunerTrialState.DIAGNOSTIC_MICRO_DITHER,LagAttributionEngine#isLowVarianceSteadyCandidate) for zero-variance steady-state culprits, connected-component contraption cluster attribution (Chebyshev <= 1and cross-chunk causal links), and bounded taint-debt provenance flushing (taintedItemDebt,taintCleanWindowTicks = 35) inEmpiricalFarmBlackBox. - Sensitivity-invariant 2D Fitts's Law container cursor reconstruction (
inventoryFittsInterceptMs,inventoryFittsThroughputBitsPerSec), per-axis pitch/yaw hardware GCD aspect-ratio canopy blocking (pitchToYawGcdScaleRatio,exactHardwareGcdCanopyKey), first-60s action-sequence bigrams, two-tier lead surfacing (EMERGING_WATCHLIST_LEAD,MODERATE_INVESTIGATE_LEADwith sample-deficit diagnostics), and Wald's Sequential Probability Ratio Test (BayesianIdentityLinker#evaluateWaldSprt) across 5 independent behavioral domains. - Information-theoretic modal base-rate discounting (
COMMON_MINECRAFT_LEXICONpopulation-scaled virtual priors, default100%sensitivity discounting, andhasHighEntropyForensicAnchor/disjointHouseholdBasescorroboration gates) verified across an $N = 80$ (3,160-pair) modal and near-identical doppelganger population (100.0%evader recall,0.000%falseLIKELY_SAME_PERSONconvictions). - Database-grade $N = 10,000+$ (
49,995,000-pair) scaling acrossRelativeCoordinateSpace,BayesianIdentityLinker,GoalMarkovProfiler, andLinguisticStylometryProfiler: $O(N)$ inverted equivalence blocking indexes, 8-byteSimHashcanopy sybil clustering, $O(D)$ algebraic Incremental View Maintenance (8 mssingle-player updates at $N = 10,000$), bounded 256-entry LRU + 64-hashMinHashspatial sketches, Metwally Space-Saving lexical/trigram sketches, bitemporal forward epoch decay (advanceBehavioralEpoch) with frozen point-in-time ban snapshots, and statistical calibration documentation (docs/dev/STATISTICAL_CALIBRATION.md).