Project analytics is the statistics of spend, that is, what production costs: tokens, cache-read, the number of passes and the price of a cycle. In Forma the price is not set in advance; it is recorded as work goes on and read as a trend.
What is counted
Each call of a node leaves a trace: how many tokens went, how many of them were read from the cache, how many seconds the work lasted and by which call it was done. From these traces the price of a task is built, from tasks the price of a cycle, and from cycles the Economics of the project.
- tokens — the amount of model work in one call.
- Cache-read — usually a constant part: loading the rules and the role, not the work itself.
- Time — the duration of a call in whole seconds.
- Passes — how many times a task went the path from kitting to check.
Who sets the goal
The analytics goal is set by Core by default, at installation: Economics as it is. A person does not name a budget in dollars, and Forma does not ask for one. Token spend and time are statistics, not thresholds; the protocol has two thresholds: attempts and the volume of a cycle.
How spend is recorded
Spend is written by the calling node, right after the call returns, as a line in the card’s History section. The node does not write its own spend: from inside it does not know it; that knowledge lies in the call’s metadata.
`Run`, 2026-09-26: attempt, 48210 tokens (31000 cache-read), 212 s, `a1b2c3` — claude-code: page template assembled.If the engine did not return a number, the line is still written, with a marker in place of the field and the reason after the dash. Unknown is counted separately and is never passed off as zero.
Goal route
A card’s route is named by its route-N label; for tasks on the short route it is Intent → the human approves five fields → Kit equips and executes → Intent checks against the criterion. The remark on spend is written by Core, when it closes the cycle.
Good, fast, cheap
At the center is “good”, and it does not give way to the other two. Intent holds “cheap”, Kit — “fast”, Core — “good”. Spec and Run outside this trio: their job is execution, not watching beyond the limit.
Where to look at the numbers
The spend lines are gathered by the script tally.cjs, and the dashboard shows them by node, route and cycle. A separate article covers the dialogue “Intent ↔ human”, the shared conversation log. Routes are compared by card lines, not by adding two logs together.
Who confirms
The verdict on a cycle is issued by Core: whether the cycle reached the image, whether a threshold was breached, and where the numbers go from cycle to cycle. The goal is closed by the human; only the human can recognize it as reached.
Leave a Reply