Why a receipt at all
AI coding bills arrive as one number at the end of the month. When someone asks what that money bought, or whether a tool actually saved anything, the honest answer is usually a guess.
A Mikie receipt replaces the guess with a record you can check, turn by turn.
What a receipt shows
Each receipt is split into sections you can open one at a time:
- Usage: the tokens the turn used.
- Costs: what that usage cost, and the price basis it was calculated from.
- Savings: what Mikie saved compared with the agent working alone.
- Tools and failures: which tool calls ran and which failed.
- Outcomes: whether the work was accepted.
- Privacy and timeline: what was recorded, and when.
Every number says how sure it is
Not every figure can be measured exactly, so every value on a receipt carries a label: exact, calculated, estimated or unavailable.
Unpriced usage stays unpriced instead of quietly becoming zero. A subscription fee is not spread into a made-up per-turn cost. If a number is missing, the receipt says so.
Your code is never on it
Receipts are built from content-free evidence. They never contain source code, prompts, model or tool output, commands, secrets, credentials or raw file paths.
That evidence is encrypted on your machine before upload and signed by your device, so the record your finance team reads is the record your machine made.
Read a result in the right order
Start with the outcome: was the task accepted, and did its tests pass? Then check the usage unit and evidence label. Finish with the cost basis and any savings comparison. This order keeps an attractive number from masking an incomplete fix.
- Match the task and acceptance rule before comparing two sessions.
- Check whether usage is provider-reported, calculated, projected, or unavailable.
- Treat a list-price estimate separately from an invoice or subscription payment.
- Keep retries and failed calls visible when deciding whether the workflow improved.
Turn a receipt into a team decision
Suppose two developers solve different tasks on different models. Their receipts can describe each session, but they cannot establish that one tool saved money relative to the other. Use the same task, model, acceptance check, and complete provider usage when you need a comparison.
Accepted outcome: pass / fail / unknown
Usage unit: input, cached input, output, or plan allowance
Evidence label: measured / calculated / projected / unavailable
Cost basis: provider invoice / list-price estimate / unpriced
Comparable baseline: same task and acceptance rule, or unavailableThis is a review template, not a sample receipt or a measured result. Keep private credentials and source content out of shared notes.
Sources: Receipt sections, value labels and privacy rules are from Mikie's documentation and the shipped Meter receipt view.