Workflow Evolution Method: V12 v1

Status: METHOD_PRESERVED_GENERIC_CONTROLLER_NOT_PROVEN

Stable invocation: $workflow-evolution <subject> or /workflow-evolution <subject>

Purpose

V12 is the outcome-to-learning method that was used to develop and evaluate workflows. Its value is
that completion does not stop at theory, code, merge, or deployment. It asks whether the result was
independently reviewed, reusable by a fresh AI, effective in the intended environment, converted
into a bounded lesson, and used to establish a better successor baseline.

The name V12 came from the original twelve barriers, B0-B11. The later B12-B15 extension closes
the compound-learning loop. The stable user-facing name is workflow-evolution; users do not need
to remember the historical version label.

Barrier contract

Step	Barrier	Required outcome
1	`B0 OBJECTIVE_FREEZE`	Freeze intended outcome, beneficiary, acceptance, boundaries, and authority.
2	`B1 BASELINE_MEASURED`	Measure current truth before intervention.
3	`B2 DELTA_GRAPH_VALID`	Describe the evidenced difference between baseline and outcome.
4	`B3 ROAD_OPTIONS_REVIEWED`	Produce and independently challenge materially different roads.
5	`B4 OPERATOR_STEP_SELECTION`	Preserve operator selection of the next bounded road.
6	`B5 SUPPLY_RECONCILED`	Reconcile demand with existing services, contracts, tools, and evidence.
7	`B6 STEP_ADMISSION`	Admit one bounded step with authority, controls, recovery, and acceptance.
8	`B7 STEP_RESULT`	Perform only the missing delta and retain exact result evidence.
9	`B8 STEP_RESIDUE`	Package reusable contracts, evidence, procedures, findings, and limits.
10	`B9 INDEPENDENT_REVIEW`	Independently review the exact subject and evidence.
11	`B10 READBACK`	Give the residue to a fresh AI and observe whether it can discover and reuse it.
12	`B11 EFFECT_MEASURED`	Measure actual behavior in the required environment.
13	`B12 LEARNING_CANDIDATE`	Convert success, failure, and rejected theories into a candidate lesson.
14	`B13 NEW_BASELINE`	Measure the post-intervention state using a comparable method.
15	`B14 NEXT_ROADS`	Generate successor opportunities from new evidence.
16	`B15 OPERATOR_DECISION`	Let the operator choose, defer, revise, or close the next road.

Invocation behavior

1. Read METHOD-REGISTRY.v1.json and report the resolved method status.
2. Pin the subject, direct inputs, governing documents, and non-goals.
3. Start at B0; do not jump to a later barrier because an old artifact appears to satisfy it.
4. Treat each barrier as a claim requiring its own evidence and explicit transition.
5. Use Scoped Delivery as a compiler within B0-B9 where its exact instrument is healthy and its
output fits the problem. Scoped Delivery is not the V12 parent lifecycle.
6. Use a fresh agent or session for B10 readback when available; give it the durable residue, not
the authoring conversation.
7. Keep SOURCE, BUILT, DEPLOYED, and EFFECTIVE evidence independent.
8. At B12, store a candidate lesson with evidence, limitations, failed alternatives, and source
identity. Do not call storage learning.
9. Promote a lesson only after transfer/readback and successor evidence show useful reuse.
10. Finish at B15 with an operator decision or an explicit open decision, never an implied next
cycle.

The current default is a governed manual run. As last recorded on 2026-08-21, Role Fabric and
Temporal could coordinate bounded fixed-topology leaves and a no-effect shadow pilot, but could not
execute an arbitrary end-to-end B0-B15 parent workflow. Do not represent that gap as closed
without fresh exact-subject evidence.

Authority stops

B4 requires operator road selection.
B6 requires explicit work/effect admission appropriate to the step.
B9 requires an independent reviewer; the author cannot provide it.
B11 deployment or external effects require separate environment/effect authority and rollback.
B15 remains an operator decision.

Stopping at one boundary does not invalidate evidence completed before it. Record the run as
partial, identify the exact unmet transition, and preserve a durable resume reference.

Durable run residue

A durable run should identify:

programme_run_id, method version, source commit/tree, and input digests;
barrier and node IDs, parent/child references, dependency/join state, and transition receipts;
role bindings, claims, fences, worktree identity, and authority references where applicable;
authored artifacts, source observations, runtime observations, reviews, decisions, and errors as

distinct record classes;

evidence digests and content-addressed locations for large raw artifacts;
current barrier state, blockers, next check, and durable resume instructions;
candidate lessons and their promotion/expiry state;
predecessor and successor method/workflow versions.

Persist stable contracts and accepted evidence in Git. Use Jira for human work projection, Role
Fabric for role-bound work custody, Temporal for orchestration history, the durable channel for
communication provenance, and Cortex for advisory semantic recall with canonical backlinks. None
of those stores replaces the others or grants authority merely by containing a record.

Last recorded position

The method and its evidence were not lost. The repository contains:

the June compound-learning master and outcome-learning spine;
the June validation protocol;
the August V12 end-to-end roadmap and deployment/effectiveness notes;
the August Swarm/Evolution/Auto-Runner/Temporal integration map;
the August parallel-role workflow readiness assessment.

The last recorded recommendation was to ratify a closed V12WorkflowSpec, run
V12_PARALLEL_SHADOW_V0, use its manual join ledger as the acceptance corpus for a generic reducer,
add an observer-only V12 parent, and repeat with automatic dependency release while effects remain
disabled. The recorded state explicitly rejected a claim of automated end-to-end B0-B15
execution.

Governing references

delivery/SWARM-COMPOUND-LEARNING-MASTER-2026-06-07.md
delivery/reviews/2026-06-05-swarm-compounding-learning-from-outcomes.md
delivery/COMPOUND-LEARNING-VALIDATION-PROTOCOL-2026-06-12.md
delivery/tracker/V12-E2E-ROADMAP-2026-08-14.md
delivery/tracker/V12-DEPLOY-RUNBOOK-2026-08-15.md
delivery/tracker/V12-CLONE-EFFECTIVENESS-PROOF-2026-08-15.md
delivery/team-role-queue-temporal-execution/SWARM-EVOLUTION-AUTORUN-V12-INTEGRATION-2026-08-21.md
delivery/team-role-queue-temporal-execution/V12-PARALLEL-ROLE-WORKFLOW-READINESS-2026-08-21.md
delivery/team-role-queue-temporal-execution/META-AGENT-RECURSIVE-PLATFORM-EVOLUTION-2026-08-21.md
delivery/TEAM-OPERATING-CHARTER-2026-08-18.md

Human presentation:

https://fifth9.atlassian.net/wiki/spaces/MD/pages/1483046914/V12+and+Scoped+Delivery+-+Human+Guide+-+2026-09-03

The Confluence page and local recovered preview are explanatory projections. The repository files
above govern the method.

Workflow Evolution Method: V12 v1