Command Reference¶
Turing provides 74 commands organized into 16 categories, covering the full lifecycle of autonomous ML research: scaffolding, training, analysis, statistical validation, reporting, and production export.
Essential Commands¶
If you read nothing else, learn these three:
| Command | What it does |
|---|---|
/turing:init |
Scaffold a new ML project with the autoresearch harness |
/turing:train |
Run the autonomous experiment loop until convergence |
/turing:brief |
Generate a 2-minute intelligence briefing on experiment progress |
Categories¶
| Category | Commands | Page |
|---|---|---|
| Core Loop | init, train, sweep, status, compare | Core Loop |
| Taste-Leverage Interface | try, brief, suggest, explore, design, mode | Taste-Leverage |
| Statistical Rigor | validate, seed, reproduce | Statistical Rigor |
| Experiment Intelligence | diagnose, ablate, frontier, card, logbook, report, poster | Experiment Intelligence |
| Performance & Resources | profile, checkpoint | Performance |
| Research Workflow | lit, paper, export | Research Workflow |
| Experiment Orchestration | queue, retry, fork | Orchestration |
| Deep Analysis | diff, watch, regress | Deep Analysis |
| Model Composition | ensemble, stitch, warm | Composition |
| Scaling & Efficiency | scale, budget, distill | Scaling & Efficiency |
| Meta-Intelligence | transfer, audit | Meta-Intelligence |
| Pre-Training Intelligence | sanity, baseline, leak | Pre-Training |
| Model Debugging | xray, sensitivity, calibrate, feature, curriculum | Model Debugging |
| Model Surgery | prune, quantize, merge, surgery | Model Surgery |
| Experiment Archaeology | trend, flashback, archive, annotate, search, template, replay | Archaeology |
| Research Communication | cite, present, changelog | Communication |
| Collaboration | onboard, share, review | Collaboration |
| What-If Analysis | whatif, counterfactual, simulate | What-If Analysis |
| Model Lifecycle | update, registry | Model Lifecycle |
| Operational Intelligence | postmortem, doctor, plan, preflight | Operational |
Command Anatomy¶
Every Turing command follows a consistent pattern:
- Positional arguments are typically project paths (
ml/sentiment) or experiment IDs (exp-042). - Flags modify behavior (
--auto,--deep,--format latex). - All commands auto-detect the project directory from
config.yamlif not specified. - Read-only commands (status, compare, brief) never modify code or data.