Mission Control · API bay · §4.6
Output Length Linter
Paste the prompt, plug in your measured output-length percentiles, and we'll flag the verbosity bait, missing length caps, and misconfigured max_tokens ceilings that are quietly inflating your output bill.
Commander tier required
Locked on the recon tier.
Output Length Linter is part of the API bay. Commander unlocks all 6 API-bay tiles + all 8 Code-bay tiles, $25 AI budget, public REST API, and CLI plugin for $24/mo.
See tiersWhat this catches
Verbosity bait. Words like "comprehensive", "thorough", "in depth" push the model toward longer answers without adding accuracy. Replace with "concise" or an explicit cap.
Missing length cap. Without an explicit "≤ 80 words" or "single paragraph", outputs drift toward the max_tokens ceiling on long-context inputs.
Missing schema. Free prose averages 30–60% more tokens than JSON / Markdown tables / bullet lists for the same information. If the answer is enumerable, structure it.
max_tokens over-provisioned. If your p99 actual output is 200 tokens and max_tokens = 4096, the prompt is asking for far more than it ever delivers — usually a sign of verbosity bait you can excise.
max_tokens under-provisioned. If p99 is within 10% of the cap, you're truncating the long-tail. Raise the cap or rewrite the prompt to tighten output.