Statistics
Statistics of our own development
Our own development of spec-editor: one project, one pipeline.
spec-editor is developed with its own pipeline of AI agents, and every step of that work is an event in a journal. This page publishes the totals of that journal and nothing else. It is not a benchmark, not a measure of quality, and not a price list: it shows how much work one project has put through one pipeline, so you can judge the scale of what the tool does. Individual tasks and accounts are not published.
Period
2026-09-19 to 2026-10-12
2.11 million
Events
Steps recorded in the journal
70,225
Model calls
Finished calls to a language model
2.76 billion
Prompt tokens
Text sent to models, cache reads included
37.12 million
Completion tokens
Text written by models
2,252
Confirmed requirements
Distinct requirements that reached the status confirmed
1,605
Applied plans
Distinct plans applied to the model and code
1,499
Deployments
Distinct builds recorded as live on a stand
Breakdown
Model calls
| Name | Value (model calls) |
|---|---|
| coder | 59849 |
| analyst | 2489 |
| tester | 2196 |
| analysis-validator | 1713 |
| project-manager | 1469 |
| analyst-manager | 884 |
| scenario_designer | 385 |
| other | 1240 |
| Name | Value (model calls) |
|---|---|
| sonnet | 40238 |
| opus | 29823 |
| other | 164 |
Read this first
Caveats
- Token counts are a lower bound: about 53 percent of model calls report no token counters at all.
- Most prompt tokens are cache reads, that is, text the model had already seen and read again.
- The pipeline ran on subscriptions, so these are not purchased tokens.
- The numbers describe one project and are not a benchmark.
- Cost is not published.
Shared statistics
Shared observations store
0
Shared observations store: 0. It fills as installations opt in to the exchange.