Changelog¶

All notable changes to Turing are documented here. Format follows Keep a Changelog.

v4.8.0: Skills as Command Source¶

2026-05-10: The modern skills/turing/ layout is now the command source, not a mirror.

Made skills/turing/ the editable command source and generate commands/ as a compatibility tree
Added src/sync-commands-layout.js as the primary source-to-compatibility sync script, with sync:skills retained as an alias
Updated installer deployment to read from skill sources while preserving .claude/commands/turing runtime paths
Enabled model invocation for command skills and aligned router/registry tests with direct sub-command handling

2037 tests | 74 commands | 93 scripts | 5 commits

v4.7.0: Modern Skills Layout Mirror¶

2026-05-10: Turing now ships a modern Claude Code skills/turing/ mirror without changing the /turing runtime contract.

Added a generated skills/turing/ mirror for modern SKILL.md package layout conventions
Added src/sync-skills-layout.js with write/check modes to keep the mirror synchronized from commands/
Added mirror drift tests for missing, stale, and divergent skill files
Tightened npm package contents to exclude generated Python cache/build artifacts while including skills/turing/**

2032 tests | 74 commands | 93 scripts | 5 commits

v4.6.0: Command Registry Contract¶

2026-05-10: Command semantics are now explicit instead of implied by scattered manifests.

Added config/commands.yaml as the canonical command and config-file registry
Moved installer and verifier command manifests to registry-derived loading
Clarified /turing router behavior for slash-only commands and model-invocation metadata
Added registry/frontmatter/router/install/verify drift tests, including mutation semantics checks

28 registry/manifest tests | 74 commands | 11 config files | 10 commits

v4.5.0: Release Reliability¶

2026-05-10: The docs got a public home and the installer/doctor path got boring in the best way.

Published the documentation site: homepage, getting started, architecture, command reference, changelog, and philosophy sections
Hardened turing-init template discovery and scaffold verification across installed layouts
Fixed Claude Stop hook generation and added /turing:doctor --fix migration for legacy hook settings
Polished README/docs prose, diagrams, homepage layout, and visual styling

2010 tests | 74 commands | 93 scripts | 68 commits

v4.4.0: Operational Intelligence¶

2026-04-02: The final phase. All 29 phases (84 features) complete.

/turing:postmortem: automated failure diagnosis with 5 root cause categories
/turing:doctor: harness self-diagnosis with 7 health checks and --fix mode
/turing:plan: strategic research campaign planning with ROI-based budget allocation

1986 tests | 74 commands | 93 scripts | 18 commits

v4.3.0: Model Lifecycle¶

2026-04-01

/turing:update: incremental model update with forgetting detection
/turing:registry: 4-stage model registry with promotion gates
Enhanced /turing:card with fairness analysis and registry status

1876 tests | 71 commands | 90 scripts | 16 commits

v4.2.0: What-If Analysis¶

2026-04-01

/turing:whatif: answer hypotheticals from existing data (7 estimators)
/turing:counterfactual: minimum-change prediction explanations
/turing:simulate: predict experiment outcomes with surrogate model

1740 tests | 69 commands | 88 scripts | 18 commits

v4.1.0: Collaboration¶

2026-04-01

/turing:onboard: project walkthrough for new collaborators
/turing:share: experiment packaging into portable archives
/turing:review: peer review simulation with venue calibration

1576 tests | 66 commands | 85 scripts | 15 commits

v4.0.0: Research Communication¶

2026-04-01: The v4.0 milestone. Every result becomes a shareable artifact.

/turing:cite: citation & attribution manager with BibTeX generation
/turing:present: presentation figure generation (5 chart types, 3 styles)
/turing:changelog: model changelog for technical and stakeholder audiences

1566 tests | 63 commands | 82 scripts | 14 commits

v3.5.0: Experiment Archaeology¶

2026-04-01

/turing:trend: long-term improvement velocity and diminishing returns
/turing:flashback: session context restoration
/turing:archive: experiment lifecycle cleanup
/turing:annotate: retrospective annotations
/turing:search: natural language experiment search
/turing:template: reusable config library
/turing:replay: re-run old experiments with current infrastructure

1547 tests | 60 commands | 79 scripts | 26 commits

v3.4.0: Model Surgery¶

2026-04-01

/turing:prune: weight pruning (magnitude/structured/lottery)
/turing:quantize: post-training quantization (FP16/INT8)
/turing:merge: model merging (soup/TIES/DARE)
/turing:surgery: architecture modification

1494 tests | 53 commands | 72 scripts | 18 commits

v3.3.0: Feature & Training Intelligence¶

2026-04-01

/turing:feature: automated feature selection with consensus ranking
/turing:curriculum: training curriculum optimization

1440 tests | 49 commands | 68 scripts | 11 commits

v3.2.0: Model Debugging¶

2026-04-01

/turing:xray: internal model diagnostics
/turing:sensitivity: hyperparameter sensitivity analysis
/turing:calibrate: probability calibration (ECE/MCE, Platt, isotonic)

1398 tests | 47 commands | 66 scripts | 15 commits

v3.1.0: Pre-Training Intelligence¶

2026-04-01

/turing:sanity: pre-training sanity checks (6 fast checks)
/turing:baseline: automatic baseline generation
/turing:leak: targeted data leakage detection

1339 tests | 44 commands | 63 scripts | 15 commits

v3.0.0: Meta-Intelligence¶

2026-04-01: The v3.0 milestone. Turing becomes project-aware.

/turing:transfer: cross-project knowledge transfer
/turing:audit: pre-submission methodology audit (8 checks)

1272 tests | 41 commands | 60 scripts | 15 commits

v2.5.0: Scaling & Efficiency¶

2026-04-01

/turing:scale: scaling law estimator with power-law fit
/turing:budget: compute budget manager with phase shifting
/turing:distill: model compression via distillation

1195 tests | 39 commands | 58 scripts | 18 commits

v2.4.0: Model Composition¶

2026-04-01

/turing:ensemble: automated ensemble (voting, stacking, blending)
/turing:stitch: pipeline composition with stage caching
/turing:warm: warm-start from prior model

1072 tests | 36 commands | 55 scripts | 18 commits

v2.3.0: Deep Analysis¶

2026-04-01

/turing:diff: deep experiment comparison
/turing:watch: live training monitor with alerts
/turing:regress: performance regression gate

931 tests | 33 commands | 52 scripts | 23 commits

v2.2.0: Orchestration¶

2026-04-01

/turing:queue: batch experiment scheduler
/turing:retry: smart failure recovery
/turing:fork: experiment branching

727 tests | 30 commands | 49 scripts | 17 commits

v2.1.0: Research Workflow¶

2026-04-01

/turing:lit: literature search via Semantic Scholar
/turing:paper: draft paper sections from experiment logs

664 tests | 27 commands | 46 scripts | 12 commits

v2.0.0: Deployment Bridge¶

2026-04-01: The v2.0 milestone. Experiments cross into production.

/turing:export: model export to 6 production formats with equivalence checks

611 tests | 25 commands | 44 scripts | 14 commits

v1.5.0: Performance & Resources¶

2026-03-31

/turing:profile: computational profiling with bottleneck detection
/turing:checkpoint: smart checkpoint management (Pareto pruning)

542 tests | 24 commands | 39 scripts | 13 commits

v1.4.0: Experiment Intelligence¶

2026-03-31

/turing:diagnose: error analysis with failure mode clustering
/turing:ablate: systematic ablation studies
/turing:frontier: Pareto frontier visualization

487 tests | 22 commands | 37 scripts | 18 commits

v1.3.0: Statistical Rigor¶

2026-03-31

/turing:seed: multi-seed study (mean/std/CI)
/turing:reproduce: reproducibility verification

407 tests | 19 commands | 34 scripts | 28 commits

v1.2.0: Tree-Search Exploration¶

2026-03-31

/turing:explore: AB-MCTS tree search over hypothesis space via TreeQuest

379 tests | 17 commands | 31 scripts | 11 commits

v1.1.0: Cost-Performance & Model Cards¶

2026-03-31

Cost-performance Pareto frontier
/turing:card: standardized model card generation
Leaderboard, config diffs, trajectory plotting

345 tests | 16 commands | 30 scripts | 19 commits

v1.0.1: Multi-Project Support¶

2026-03-31

Multi-project scaffold support
Model contract and registry templates

311 tests | 15 commands | 25 scripts | 11 commits

v1.0.0: Initial Release¶

2026-03-31: The research assistant that can't fool itself.

Core autoresearch loop (init, train, status, compare)
Hypothesis database with priority queue and detail files
Research briefing with campaign summary and recommendations
Immutable evaluation with 6-layer anti-cheating stack
Novelty guard, experiment families, decision packets
Literature-grounded suggestions, experiment design, mode switching
2 specialized agents (@ml-researcher, @ml-evaluator)
16 Architecture Decision Records

257 tests | 14 commands | 23 scripts | 193 commits