meted inspect

Show what Meted would allocate for a prompt.

meted inspect "What is a semaphore?"

definition · simple

Target       38
Maximum      64
Confidence   96%

Runs the Sufficiency Engine on the prompt and prints the decision. The engine makes no provider call, so inspect works before a credential is configured.

Options

Flag
`--model <id>`	Model to size against. Default `gpt-4o-mini`
`--mode <mode>`	Show the allocation this mode would apply
`--system <text>`	Include a system instruction
`--file <path>`	Read the prompt from a file. `-` for stdin
`--verbose`	Explain every rationale code
`--json`	Machine-readable output

Seeing what a mode would do

$ meted inspect "What is a semaphore?" --mode aggressive

definition · simple

Target       38
Maximum      64

Applied      51 (aggressive)
Hint         Aim for roughly 29 words unless the task genuinely needs more.

Understanding a decision

$ meted inspect "Summarize this in 3 bullet points" --verbose

summarization · trivial

Target       54
Maximum      112
Confidence   93%

Rationale
  task:summarization          Task type identified from the request
  complexity:trivial          Complexity band the request fell into
  base_table                  Started from the sufficient-answer budget for this task and complexity
  explicit_length:exact:3:bullets  Caller stated an explicit length

Engine  baseline-heuristic 0.1.0 (baseline)
Model   gpt-4o-mini, max 16384 output tokens
Decided in 0.38ms

Reasoning models

meted inspect "Explain the CAP theorem" --model o3-mini --verbose

The ceiling grows to cover hidden reasoning tokens, with a reasoning_reserve:<n> rationale code giving the amount.

Piping

cat prompt.txt | meted inspect --file -
meted inspect --file ./prompts/onboarding.txt --json | jq .decision

JSON output

meted inspect "What is a semaphore?" --json

{
  "ok": true,
  "prompt": { "model": "gpt-4o-mini", "mode": "observe" },
  "decision": {
    "taskType": "definition",
    "complexity": "simple",
    "targetOutputTokens": 38,
    "maxOutputTokens": 64,
    "confidence": 0.96,
    "rationaleCodes": [
      "task:definition",
      "complexity:simple",
      "base_table",
      "very_short_prompt",
    ],
  },
  "allocation": {
    "applied": false,
    "suggestedMaxOutputTokens": 64,
    "skipReason": "observe_mode",
  },
  "rationale": [/* each code, expanded */],
  "engine": {
    "name": "baseline-heuristic",
    "source": "baseline",
    "durationMs": 0.38,
  },
  "model": {
    "id": "gpt-4o-mini",
    "maxOutputTokens": 16384,
    "reasoning": false,
    "known": true,
  },
  "features": {/* only from the open-source reference engine */},
}