Stage 2 · farming-workshop-v1 · contract 1
Four precommitted directional scenarios probe a narrow frontier: discover an observed workshop, place and configure a legal farm, complete native planting, and produce one requested bed or barrel through typed actions.
One external-model south/barrel episode and one reviewed reference-policy episode are accepted proofs. The other directional attempts remain failure or exploratory evidence. This page does not turn a single accepted scenario into a model comparison or a claim of autonomous fortress control.
trajectory 7095a5b0ad5347de…4c6bcd78d741f8
Four checked scenarios
Action count and final tick are reported from the retained artifact index. A retained exploratory attempt is not promoted to an accepted benchmark result.
| Scenario | Observation evidence | Action / horizon | Status + result | Limitation | Provenance |
|---|---|---|---|---|---|
north-farm-bednorth · bed |
Retained trajectory predates per-turn program-event capture. | 19 actions final tick 4,270 |
NOT ACCEPTEDpreflight-rejected |
Exploratory artifact; insufficient accepted evidence. | trajectory 37eac230…bf17d |
south-farm-barrelsouth · barrel |
One bounded region query; exact typed observation, decision, frame-fence attempts, receipts, and outcomes retained. | 17 actions objective 7,200 final 16,800 |
ACCEPTED · PASS2 × 2 farm; 4 seed jobs; barrel +1; 1,200-tick holdout. | One scenario and one recorded model route; no comparative or cross-seed conclusion. | trajectory 7095a5b0…d741f8 |
east-farm-bedeast · bed |
Retained trajectory and exact rejection: farm plot 11 was incomplete. | 22 actions final tick 4,884 |
NOT ACCEPTEDpreflight-rejected before crop configuration mutated state. |
Farm construction incomplete; no accepted production proof. | trajectory 42c40e27…98fe05 |
west-farm-barrelwest · barrel |
Retained trajectory includes two exact frame-fence mismatch rejections. | 28 actions final tick 1,746 |
NOT ACCEPTEDconstruction and both production objectives missing; budget exhausted. | Farm incomplete, no farm/workshop production, invalid attempts remain counted. | trajectory 6969e523…e2614b |
Proof ledger
The accepted model episode, accepted reference policy, and verified failed replay prove different contracts. They are not ranked against one another.
south-farm-barrel@1omp-completion / gpt-5.62daec423…efdb4c7095a5b0…d741f8reference-farming-policy-v1dwarf-societya36bf6ee…817d0791762243…a3aec5dscenario-unsatisfiedscenario-unsatisfiedunsupported-query0abd3e67…f1082ea97c53…bd93Complete Stage 2 protocol reference
Inference receives only the precommitted scenario, capability manifest, current observation, and bounded transcript. It cannot call DFHack, inspect host files, choose a private tool, or create a task-specific endpoint.
farming-workshop-v11classic-farming-workshop@1classic-linux-farming-workshop-trajectory-v1inspect_fortressCitizens and current jobs; bounded item counts; seed crops and eligibility; buildings, farm settings, active jobs and recipes; exact completed farm and workshop event evidence.
inspect_regionAt most 256 observed-map tiles: type, shape, material, hidden/outside/subterranean state, flow, occupancy, building, pathability, and 1 × 1 farm eligibility. Read-only discovery.
| Capability | Effect | Hard bound |
|---|---|---|
place_building | Place one farm plot. | Causally prior eligible region; reachable 1…5 × 1…5 rectangle; at most 25 tiles. |
configure_building | Set one farm season crop. | Completed observed farm; observed current seed-bearing crop; seeds available. |
enqueue_job | Queue one observed workshop recipe. | Completed observed workshop/furnace; queue below 10. |
set_job_priority | Change do_now. | Active job created by this run only. |
cancel_job | Cancel a job. | Active job created by this run only. |
advance | Advance native simulation. | Exact integer 1…1,200 ticks; returns paused. |
PlantSeeds job on that farm.model-failed, invalid-decision, unsupported-query, unsupported-action, preflight-rejected, apply-rejected, observation-failed, stale-evidence, corrupt-evidence, and turn-budget-exhausted.
Trajectory digest, stable identities, scenario and manifest, causally prior placement query, payloads, retries, receipt/revision/tick chains, program termination, exact outcome or failure, cleanup, and source/save integrity.
Limitations / provenance
classic-linux-v18fa73584a26ba3122f79f956d10ff0bf4b83889a0ac3f289bf56aa530d3576773a63960234adecdbded87a544cbfd85dd5fc77b37e5d679bd0cf23e97bfc153a