The runs
Every report here is a real run, published as the shape of the work rather than as whose work it was. The numbers are recomputed from the run store, and the run that ended one checkpoint short is in the list on purpose.
- Report 1The fleet round
A four-site web fleet, one shared component library, one round of coordinated changes
- Report 2The long build
A large interactive feature built inside an existing site, and the checkpoint that never closed
- Report 3The engine run
A static-analysis engine for a compiled language, with an evaluation suite as the release gate
The whole corpus
Three of these runs have a report above. The rest are here because a corpus that shows only the runs worth writing up is not a corpus, it is a portfolio. Every row is recomputed from the same store the reports are: what the run closed against what it planned, how many sessions it took, what it cost, and how its gates went.
Read down and the fortnight happens in order. It opens with a plan tried twice and dropped, a run aborted inside its first minute, and a prototype left where it stood when the work moved elsewhere. Those are the honest part of the list, and they are why the last row adds up to more than the three reports do.
The store still marks three of them running. They are not. Their engine exited without ever closing the record, and nobody went back. Whether a stopped run was abandoned or is merely waiting is the one thing about this corpus no query can answer, so it is written down by hand beside the name each run is published under — and the table says abandoned here rather than leaving a reader to assume the work is still in flight.
| the run | checkpoints closed | sessions | of engine time | tokens | spent | gates green |
|---|---|---|---|---|---|---|
| A first attempt at an orchestrator plan, one session long, stopped before it had cost anythingabandoned | 0/0 | 1 | <1m | 0 | $0.00 | 0/0 |
| The same plan tried again three sessions later, still with no checkpoint declared and no gate greenabandoned | 0/0 | 3 | 20m | 7.3M | $0.17 | 0/4 |
| A run aborted inside its first minute, kept because a corpus that hides its dead starts is not a corpusaborted | 0/0 | 1 | 14m | 0 | $0.00 | 0/0 |
| A prototype run that reached six of eleven checkpoints and was left there when the work moved elsewhereabandoned | 6/11 | 11 | 11h 19m | 109.7M | $139.68 | 27/27 |
| A static-analysis engine for a compiled language, with an evaluation suite as the release gate | 22/22 | 26 | 14h 47m | 523.7M | $358.42 | 72/81 |
| Shared CI and a 404 page added to a site template in a single session | 2/2 | 1 | 11m | 3.2M | $2.91 | 3/3 |
| A web fleet's accumulated backlog worked to the end of the list across four sites | 29/29 | 24 | 7h 53m | 219.2M | $224.28 | 20/20 |
| A four-site web fleet, one shared component library, one round of coordinated changes | 25/25 | 28 | 15h 1m | 621.4M | $425.12 | 32/32 |
| The orchestrator's own core built by the orchestrator: sessions, gates and the event store | 26/26 | 28 | 14h 12m | 504.5M | $360.14 | 94/96 |
| The round after the round: the same web fleet, a shorter list, and a run that closed every checkpoint it opened | 17/17 | 17 | 8h 13m | 317.6M | $222.63 | 38/38 |
| A large interactive feature built inside an existing site - the run that ended at 45 of 46, and why the last one did not close | 45/46 | 69 | 1d 1h | 494.4M | $421.46 | 59/59 |
| A terminal interface for a long-running engine, written in a second language against the first one's store | 24/24 | 41 | 14h 40m | 397.2M | $297.24 | 104/110 |
| A second interactive feature on the same site, planned after the first one had taught the plan what to say | 17/17 | 15 | 5h 11m | 118.2M | $97.48 | 24/24 |
| Continuous integration made honest across a set of related repos, the cheapest complete run in the corpus | 20/20 | 12 | 3h 10m | 44.4M | $37.99 | 23/28 |
| A third feature on the same site, including the way back out of it, closed at twenty of twenty | 20/20 | 15 | 5h 46m | 141.8M | $110.93 | 43/43 |
| The engine taught to account for itself - what it did and what it cost - and closed at 25 of 32, seven checkpoints short | 25/32 | 32 | 14h 1m | 403.9M | $317.84 | 97/100 |
| The engine demonstrating itself against a throwaway repo, six sessions, no model spend | 3/3 | 6 | <1m | 1.6K | $0.00 | 6/6 |
| A rehearsal of one engine feature end to end before it was let near real work | 6/6 | 10 | <1m | 2.9K | $0.00 | 6/6 |
| The orchestrator's own release era - seven stages of hardening and an open door for other people's plans - ended by hand at the ship stage once the release it was for was outaborted | 30/32 | 24 | 6h 53m | 198.6M | $147.55 | 66/68 |
| The era after that release - gates an agent cannot game, and a courier that carries a run to a phone - parked one checkpoint short at a ship step only a person can perform, and released by hand from that park rather than by resuming the runabandoned | 23/24 | 23 | 14h 39m | 414.3M | $324.01 | 67/68 |
| 20 runs across 7 repositories | 340/356 | 387 | 6d 17h | 4.5B | $3,487.85 | 781/813 |
Recomputed from conductor history --json --limit 0 and run.db, opened read-only. Nothing in this table is typed in.