AI Unit-Economics & Token-Shock Exposure Kit

Three things. Per AI workflow, it computes the fully-loaded cost per successful outcome — accounting for retries and expected growth — and bands it PREDICTABLE, DRIFTING, or TOKEN SHOCK. It grades the contract behind each workflow on four controls, so you know whether an unpriced growth month can turn into an unbounded bill. And its Model Swap tab tells you whether a cheaper model would cost less per finished task once its misses are paid for.

Yes, if you bring the pass rates. On the Model Swap tab you enter each model's price per 1M tokens, the tokens in and out per task, the share of your own tasks each model got right in your own test, and what it costs to fix one miss. It prices both models per finished task, shows the pass rate the cheaper model needs to break even, and reads ROUTE DOWN, KEEP, or UNMEASURED. It doesn't test models for you: with no measured pass rate it reads UNMEASURED, however large the saving looks. The Prompt Evaluation & Versioning System is one way to measure those rates.

No. Every figure comes from the numbers you enter — your token counts, your run volumes, the price your vendor actually charges. There is no hidden per-token rate or growth multiplier. The verdict is yours, not ours.

If a workflow has no usage cap AND its true-up terms are uncapped, it reads TOKEN SHOCK even when today's cost per outcome is a penny — because the thing that protects you is not today's price, it's the ceiling on a bad month. A workflow that is cheap now and uncapped is one growth spike from a runaway bill. Add either a cap or a capped true-up and the gate clears.

The Token Economics Workbook is a developer tool for routing, caching, and model-mix decisions — how to make each call cheaper. This kit is a pre-scale exposure check for founders and operators: what does each workflow cost per outcome, and does the contract protect you as volume grows. Its Model Swap tab answers one narrower question the workbook's routing matrix leaves to you: for a workflow you run today, is a specific cheaper model cheaper per finished task once misses are counted. Optimize spend there; price exposure here.

An incident kit is about a leaked key or a security event — containment after something goes wrong. This is about economics before anything goes wrong: whether the unit cost holds and whether the contract caps your downside. Different problem, different moment.

No. It is a plain .xlsx with live formulas — no API call, no upload, no login. It opens in Excel, Google Sheets, or Numbers and recomputes as you change your numbers. It prices workflows and grades contract terms, never people.