# Lumyn FDE School Field Capstone

Take one inherited brief through the complete field-to-operation spine using live
field evidence, a sanitized retrospective, or synthetic practice. Keep source
evidence and revisions connected; do not create ten unrelated artifact systems.

## Choose an evidence mode

- **Live field evidence:** use permitted customer sources and record their exact
  authority, confidentiality, and retention limits.
- **Sanitized retrospective:** preserve evidence classes, chronology, and limitations
  while removing customer-confidential context.
- **Synthetic practice:** use the [fictional claims-denial case](https://www.lumyn.cc/resources/claims-denial-practice-case/).
  Label every result as practice evidence; it cannot prove field observation,
  customer acceptance, release readiness, or operating ownership. Compare your
  first pass with the [annotated strong answer](https://www.lumyn.cc/resources/claims-denial-annotated-answer/).

Use the [field conversation kit](https://www.lumyn.cc/resources/field-conversation-kit/) for observation and
reframe meetings. Use the five [optional technical lab briefs](https://www.lumyn.cc/resources/optional-technical-labs/)
when you want to implement and test the engineering boundaries yourself.

## Capstone case

- Customer-safe case name:
- Inherited promise:
- Consequential workflow:
- Decision event and deadline:
- Confidentiality constraints:
- Evidence mode and limitations:

## Living artifact registry

Assign stable IDs. Add a row for every revision instead of overwriting history.

| Artifact ID | Type | Revision or digest | Status | Source dependencies | Supersedes | Owner | Current limitation |
| --- | --- | --- | --- | --- | --- | --- | --- |
|  | observation / reframe / charter / mechanism / design / eval / security / release / service / handoff |  | current / review required / superseded |  |  |  |  |

## Typed dependency register

Use only relationships that can be explained and reviewed. The register routes
change review; it does not grant authority or prove completion.

| From object and revision | Relationship | To object and revision | Provenance | Currentness | Change consequence |
| --- | --- | --- | --- | --- | --- |
|  | supports / contests / constrains / depends on / supersedes / affects |  |  |  |  |

When evidence or a disposition changes, update only dependency-linked objects and
preserve unaffected revisions.

## Current Outcome Contract projection

This is the current accepted view—or an explicitly provisional view while scoped
authority remains open—derived from the registry, not a separately maintained
source of truth. A provisional projection must name the authority gap and cannot be
used as customer acceptance or production authority.

- Decision event and deadline:
- Accepted outcome, eligible workflow, and exclusions:
- Baseline, target, measurement window, and attribution limits:
- Metric owner and independent verifier:
- Economic exposure, full cost basis, and decision threshold:
- Human authority retained and authority basis:
- Safe fallback and stop condition:
- Current evidence maturity and material conflicts:
- Current mechanism and exact candidate, when applicable:
- Operating owner, transfer condition, and next review:
- Projection revision, generated from registry revisions:

## Phase 1 — Correct the story

1. Observe one representative case and classify consequential evidence.
2. Preserve the inherited brief and expose one material contradiction.
3. Qualify the accepted outcome—or preserve a provisional outcome hypothesis with
   its authority gap—plus economics, four constraint gates, authority, and next
   field move.

Evidence to attach:
- Field observation log
- Engagement reframe record
- Workflow charter and value case

Review gate: another reviewer can distinguish what was sold, observed, tested, allowed, asserted, contested, and unknown.

## Phase 2 — Prove the slice

4. Select the smallest sufficient mechanism for each consequential decision.
5. Design one typed, recoverable vertical slice and its dependency graph.
6. Build realistic deterministic, behavioral, safety, cost, latency, and recovery evidence.
7. Secure model-visible reads and real-world effects at trusted boundaries.
8. Decide launch, constrain, or hold for the exact candidate.

Evidence to attach:
- Intelligence-selection record
- Bounded system design packet
- Evaluation set and report
- Tool contract and threat tests
- Production readiness review

Review gate: every consequential route is traceable from source evidence through trusted authorization, effect verification, and rollback.

## Phase 3 — Prove ownership

9. Run a service review using outcome, adoption, reliability, safety, cost, change, and ownership evidence.
10. Exercise support, evaluation, change, incident, rollback, and retirement with the receiving team; preserve sanitized learning.

Evidence to attach:
- Production service review
- Enablement handoff and field-learning record

Review gate: the accountable team can operate, change, recover, and retire the service without hidden delivery-team dependence.

For synthetic practice, replace this gate with: the learner can demonstrate the
required decisions and tests in a simulation while naming every capability that
still requires live operating-team evidence.

## Final review rubric

- The original brief and every superseded revision remain visible.
- Consequential claims are source-linked or explicit unknowns with evidence limitations.
- Human authority is named, scoped, verified, and never inferred from model output.
- Only dependency-linked state changes; unaffected objects retain their revisions and history.
- Evaluation and release evidence bind to the exact candidate.
- The current outcome contract states measurement, economics, operating ownership, limitations, and exit conditions.
- The Outcome Contract is reproducibly projected from current registry revisions;
  it is not an independently edited artifact.
- The operating team has exercised recovery and change, not merely received documents.
- Customer-confidential context is absent from reusable learning.
- Synthetic practice, sanitized retrospective, and live field evidence are visibly
  distinguished.

## Capstone disposition

- Ready for peer review / needs more field evidence / stopped:
- Reviewer:
- Strongest evidence:
- Most material limitation:
- Next field move: