Examples

Every document on this page is a file in the specification’s examples folder, and every one passes loopmath ocp validate with no findings except where a rule file is meant to fail. The values are made up. The shapes are the ones in loopmath’s workflow catalog, inlined, so each configuration.id is the canonical id of that shape and its settings.

Workflow shapes

In each diagram the boxes are pieces (units of agent work, with their role), the rounded labels are artifacts they read and write, and a dotted edge is the repair a gate triggers when it rejects.

solo

One agent implements. The smallest workflow, with a verdict from CI and a runtime score. configuration.id cfg_261aaa36000b; source usual; budget 1 round.

Attempt	Setting	Status	Cost
`implement.a1` round 1	claude-code, claude-opus-5-5, high	done	344,000 tokens $0.99
Run			344,000 tokens $0.99

Acceptance rule tests: tests pass. Signals: tests verdict pass (verified); runtime_s score 960.0 (verified). Full document on GitHub (1 attempt, OCP 0.3).

best_of_n

Three agents implement the same task in parallel (width 3) and a referee keeps one result. configuration.id cfg_3f65648c26b6; source alternative; budget 1 round; gates select.

Attempt	Setting	Status	Cost
`implement.a1` round 1	codex, gpt-6-sol, high	done	344,000 tokens $0.39
`implement.a2` round 1	codex, gpt-6-sol, high	done	344,000 tokens $0.42
`implement.a3` round 1	codex, gpt-6-sol, high	done	344,000 tokens $0.45
`select.a1` round 1	claude-code, claude-opus-5-5, high	done	344,000 tokens $0.99
Run			1,376,000 tokens $2.25

Acceptance rule tests+referee: tests pass and the referee accepts. Signals: referee verdict accept (reported); tests verdict pass (verified); runtime_s score 1860.0 (verified). Full document on GitHub (4 attempts, OCP 0.3).

plan_implement

A planner writes a plan and an implementer follows it. configuration.id cfg_9c77371c53df; source habit; budget 1 round.

Attempt	Setting	Status	Cost
`plan.a1` round 1	claude-code, claude-opus-5-5, xhigh	done	344,000 tokens $0.99
`implement.a1` round 1	claude-code, claude-fable-5-1, medium	done	172,000 tokens $0.13
Run			516,000 tokens $1.12

Acceptance rule tests: tests pass. Signals: tests verdict pass (verified); runtime_s score 1740.0 (verified). Full document on GitHub (2 attempts, OCP 0.3).

implement_review

An implementer, then a reviewer whose rejection sends the change back once (control.repair). configuration.id cfg_ab15440a2d9c; source usual; budget 3 rounds; gates review.

Attempt	Setting	Status	Cost
`implement.a1` round 1	claude-code, claude-opus-5-5, high	done	344,000 tokens $0.99
`review.a1` round 1	codex, gpt-6-astra, xhigh	rejected	344,000 tokens $0.77
`implement.a2` round 2	claude-code, claude-opus-5-5, high	done	344,000 tokens $0.99
`review.a2` round 2	codex, gpt-6-astra, xhigh	done	344,000 tokens $0.77
Run			1,376,000 tokens $3.52

Acceptance rule tests+review: tests pass and the reviewer approves. Signals: review verdict reject (reported); review verdict accept (reported); tests verdict pass (verified); runtime_s score 3300.0 (verified). Full document on GitHub (4 attempts, OCP 0.3).

plan_implement_review

Plan, implement, then review, with a repair loop the run did not need. configuration.id cfg_9f823d1b1d85; source exploration; budget 3 rounds; gates review.

Attempt	Setting	Status	Cost
`plan.a1` round 1	claude-code, claude-opus-5-5, xhigh	done	344,000 tokens $0.99
`implement.a1` round 1	claude-code, claude-opus-5-5, high	done	344,000 tokens $0.99
`review.a1` round 1	codex, gpt-6-astra, xhigh	done	344,000 tokens $0.77
Run			1,032,000 tokens $2.75

Acceptance rule tests+review: tests pass and the reviewer approves. Signals: review verdict accept (reported); tests verdict pass (verified); runtime_s score 2520.0 (verified). Full document on GitHub (3 attempts, OCP 0.3).

swarm

A planner, three workers in parallel, and a reviewer that sends the work back once. configuration.id cfg_3c53469b7f4a; source user_edit; budget 3 rounds; gates review.

Attempt	Setting	Status	Cost
`plan.a1` round 1	claude-code, claude-opus-5-5, xhigh	done	344,000 tokens $0.99
`work.a1` round 1	claude-code, claude-fable-5-1, medium	done	172,000 tokens $0.13
`work.a2` round 1	claude-code, claude-fable-5-1, medium	done	172,000 tokens $0.14
`work.a3` round 1	claude-code, claude-fable-5-1, medium	done	172,000 tokens $0.16
`review.a1` round 1	codex, gpt-6-astra, xhigh	rejected	344,000 tokens $0.77
`work.a4` round 2	claude-code, claude-fable-5-1, medium	done	172,000 tokens $0.13
`work.a5` round 2	claude-code, claude-fable-5-1, medium	done	172,000 tokens $0.14
`work.a6` round 2	claude-code, claude-fable-5-1, medium	done	172,000 tokens $0.16
`review.a2` round 2	codex, gpt-6-astra, xhigh	done	344,000 tokens $0.77
Run			2,064,000 tokens $3.39

Acceptance rule tests+review: tests pass and the reviewer approves. Signals: review verdict reject (reported); review verdict accept (reported); tests verdict pass (verified); runtime_s score 4320.0 (verified). Full document on GitHub (9 attempts, OCP 0.3).

full-fields

The implement and review shape with every field v0.3 adds filled in: task features, signals including a late event and an unmeasured score, a slate, a preference and a receipt. configuration.id cfg_ab15440a2d9c; source exploration; budget 3 rounds; gates review.

Attempt	Setting	Status	Cost
`implement.a1` round 1	claude-code, claude-opus-5-5, xhigh	done	344,000 tokens $0.99
`review.a1` round 1	codex, gpt-6-astra, xhigh	rejected	344,000 tokens $0.77
`implement.a2` round 2	claude-code, claude-opus-5-5, high	done	344,000 tokens $0.99
`review.a2` round 2	codex, gpt-6-astra, xhigh	done	344,000 tokens $0.77
Run			1,376,000 tokens $3.52

Acceptance rule tests+review+runtime: tests pass, the reviewer approves and the run finishes under 60 minutes. Signals: review verdict reject (reported); review verdict accept (reported); tests verdict pass (verified); runtime_s score 3300.0 (verified); quality score null (asserted); issue event issue/431 (reported). Full document on GitHub (4 attempts, OCP 0.3).

Rule pairs

Each numbered v0.3 rule has a pass file and a fail file in examples/golden/v0.3. Apart from its id and title, the fail file differs from the pass file only in what breaks the rule, and expected.json lists the codes each file must produce. A checker conforms to a rule when it reports exactly those codes. The tables show that difference; the message is what loopmath ocp validate prints.

E190

Path	Pass file	Fail file
`$.run.configuration.settings.implement.effort`	`"high"`	`"xhigh"`

E190-fail.ocp.json

  error E190 $['run']['configuration']['id']: configuration id 'cfg_261aaa36000b' is not the canonical hash of its workflow and settings ('cfg_ebcc7a1d9137'); see loopmath.ocp.canonical

E191

Path	Pass file	Fail file
`$.run.configuration.id`	`"cfg_ab15440a2d9c"`	`"cfg_1c618686576b"`
`$.run.configuration.settings.review.context_policy`	`"fresh"`	`(absent)`
`$.run.configuration.settings.review.effort`	`"xhigh"`	`(absent)`
`$.run.configuration.settings.review.harness`	`"codex"`	`(absent)`
`$.run.configuration.settings.review.model.id`	`"gpt-6-astra"`	`(absent)`
`$.run.configuration.settings.review.model.raw`	`"gpt-6-astra"`	`(absent)`
`$.run.configuration.settings.review.options`	`{}`	`(absent)`

E191-fail.ocp.json

  error E191 $['run']['configuration']['settings']: piece 'review' has no setting; every piece vertex has one

E192

Path	Pass file	Fail file
`$.run.configuration.id`	`"cfg_9c77371c53df"`	`"cfg_e4e353faecd0"`
`$.run.configuration.workflow.edges[7][0]`	`(absent)`	`"diff"`
`$.run.configuration.workflow.edges[7][1]`	`(absent)`	`"plan"`

E192-fail.ocp.json

  error E192 $['run']['configuration']['workflow']['edges']: workflow edges contain a cycle through: diff, implement, plan, plan_doc; the repair loop belongs in control.repair

E193

Path	Pass file	Fail file
`$.run.configuration.id`	`"cfg_ab15440a2d9c"`	`"cfg_5d8a7e22679c"`
`$.run.configuration.workflow.control.repair.review`	`"implement"`	`"plan"`

E193-fail.ocp.json

  error E193 $['run']['configuration']['workflow']['control']['repair']['review']: repair target 'plan' names no piece of the workflow

E194

Path	Pass file	Fail file
`$.run.preferences[0].members[0]`	`"run_golden_e194_pass"`	`"run_golden_e194_fail"`
`$.run.preferences[0].slate`	`"slt_ex_full"`	`"slt_other"`
`$.run.preferences[0].winner`	`"run_golden_e194_pass"`	`"run_golden_e194_fail"`
`$.run.slate.members[0]`	`"run_golden_e194_pass"`	`"run_golden_e194_fail"`

E194-fail.ocp.json

  error E194 $['run']['preferences'][0]['slate']: preference names slate 'slt_other' but this run is in slate 'slt_ex_full'

E195

Path	Pass file	Fail file
`$.run.signals[0].value`	`"reject"`	`"approved"`

E195-fail.ocp.json

  error E195 $['run']['signals'][0]['value']: verdict 'review' has value 'approved'; a verdict is one of accept, error, fail, pass, reject

W182

Path	Pass file	Fail file
`$.run.ext["dev.dagr.note"]`	`(absent)`	`"extension under the pre-rename namespace"`
`$.run.ext["dev.loopmath.note"]`	`"extension under the loopmath namespace"`	`(absent)`

W182-fail.ocp.json

  warning W182 $['run']['ext']['dev.dagr.note']: extension key 'dev.dagr.note' uses the pre-rename namespace 'dev.dagr.'; it is still read through v0.4, write 'dev.loopmath.' instead

W196

Path	Pass file	Fail file
`$.producer.name`	`"loopmath"`	`"example-orchestrator"`
`$.run.preferences[0].members[0]`	`"run_golden_w196_pass"`	`"run_golden_w196_fail"`
`$.run.preferences[0].winner`	`"run_golden_w196_pass"`	`"run_golden_w196_fail"`
`$.run.slate.members[0]`	`"run_golden_w196_pass"`	`"run_golden_w196_fail"`

W196-fail.ocp.json

  warning W196 $['run']['receipt']: run.receipt is written only by loopmath, but producer.name is 'example-orchestrator'

Get started

Concepts

Implement OCP

Examples

Workflow shapes

solo

best_of_n

plan_implement

implement_review

plan_implement_review

swarm

full-fields

Rule pairs

E190

E191

E192

E193

E194

E195

W182

W196

​Workflow shapes

​solo

​best_of_n

​plan_implement

​implement_review

​plan_implement_review

​swarm

​full-fields

​Rule pairs

Workflow shapes

solo

best_of_n

plan_implement

implement_review

plan_implement_review

swarm

full-fields

Rule pairs