dry explain

Assemble an offline LLM-explanation bundle (trace + forensics + verify + a curated prompt)

Sample: VerifyCheck a resolved design against machine-safety contracts and inspect findings.typescript: docs/site/examples/verify.ts · python: docs/site/examples/verify.py

Help

Assemble an offline LLM-explanation bundle (trace + forensics + verify + a curated prompt).

The engine never calls an LLM — `explain` produces a facts-plus-prompt briefing you paste into Claude (or hand to an agent). Markdown by default; `--json` emits the structured ExplainBundle.

Usage: dry explain [OPTIONS] <FILE>

Arguments:
  <FILE>
          

Options:
      --profile <PROFILE>
          Machine/material profile JSON to supply import defaults and verifier contracts

      --filament-diameter <FILAMENT_DIAMETER>
          Filament diameter in mm, used to recover deposited volume from E motion

      --line-width <LINE_WIDTH>
          Assumed line width in mm for structural bead and flow checks

      --layer-height <LAYER_HEIGHT>
          Assumed layer height in mm for structural bead and flow checks

      --max-flow <MAX_FLOW>
          Max volumetric flow (mm³/s)

      --bounds <BOUNDS>
          Build volume as `x0,x1,y0,y1,z0,z1` (mm)

      --monotonic-z
          Require Z to be non-decreasing

      --min-temp <MIN_TEMP>
          Minimum nozzle temperature (°C) required to extrude

      --speed-range <SPEED_RANGE>
          Allowed feedrate range `min,max` (mm/min) for extruding moves

      --window-s <WINDOW_S>
          Fixed trace window duration in seconds
          
          [default: 5]

      --json
          Emit the structured ExplainBundle as JSON instead of Markdown

      --out <OUT>
          Write the bundle to this file instead of stdout

      --llm
          Call Claude directly: build the bundle, get recommendations, apply the executable ones, and report measured before/after results. Requires --model and ANTHROPIC_API_KEY

      --model <MODEL>
          Claude model id for --llm (e.g. claude-sonnet-4-6, claude-opus-4-8). Required with --llm

      --max-applies <MAX_APPLIES>
          Cap on how many executable recommendations --llm actually applies (highest priority first)
          
          [default: 4]

  -h, --help
          Print help (see a summary with '-h')