# LockedIn FDE Production Project Review Pack

`lockedin.training.production-commission/v1` · version 1.0.0 · reviewed 2026-08-29

Use this pack to frame, build, operate, and review one production-bearing forward-deployed engineering mission. Keep the completed copy in the approved project repository. Pair it with `lockedin-production-commission-manifest.schema.json` and bind every submitted record to one exact 40-character commit SHA. The filename and schema identifier retain `production-commission` only for compatibility.

## Noncompensating production rule

No LockedIn-qualified forward-deployed engineer reaches the field without candidate-owned work that was deployed to production, operated against real use, independently reviewed, and transferred with an inspectable evidence chain.

At least one exact candidate-owned release must:

- reach an authorized production environment;
- serve real users or operators;
- remain observable long enough to produce operating evidence;
- bind the reviewed commit to the deployed artifact;
- include a verified rollback or recovery path; and
- transfer to a named owner who can operate it without hidden candidate intervention.

A local demo, screenshot, prototype, preview, simulation, production-equivalent decision, teammate-only deployment, attendance record, or AI-generated score cannot satisfy this rule.

## Information boundary

Do not place passwords, access tokens, private keys, credentials, production dumps, protected health information, payment data, direct personal identifiers, client names, restricted client material, proprietary source, or model/provider secrets in this pack.

Use approved opaque references and SHA-256 digests. Client-derived work requires written authorization covering source custody, data, intellectual property, models, reviewer access, retention, incident ownership, and evidence return. Raw client evidence stays inside that boundary.

---

## 01 · Project review identity

- **Project review ID** (`commissionId` in the compatibility schema):
- **Mission class** (`cohort-venture`, `sponsor-clean-room`, or `supervised-client`):
- **Candidate ID or approved private handle**:
- **Pod ID**:
- **Business owner**:
- **Technical owner**:
- **Production owner**:
- **Requirements version**:
- **Requirements SHA-256**:
- **Approved data classification**:
- **Authorization reference**:
- **Retention-until date**:

### Mission statement

- **Observed workflow and user problem**:
- **Outcome contract**:
- **Acceptance criteria**:
- **Guardrails and critical-failure criteria**:
- **In scope**:
- **Out of scope**:
- **No-build alternative**:
- **Kill condition**:
- **Known assumptions and unknowns**:

---

## 02 · Repository and authorship

- **GitHub repository**:
- **Repository visibility**:
- **Default branch**:
- **Exact assessed commit SHA**:
- **Base commit SHA**:
- **Pull-request references**:
- **Required-check references**:
- **Branch/ruleset evidence**:
- **CODEOWNERS or owner-map evidence**:
- **Candidate contribution summary**:
- **AI assistance disclosure** (tools, scope, material generated work, and what a human verified):
- **Reproduction instructions**:

The current practice workbench accepts an authorized public repository at one exact commit and uses fixed-host, unauthenticated reads; it accepts no credential. Private or continuously connected repositories must use a selected-repository, read-only GitHub App installation. Never add a shared human or bot account as a collaborator and never paste a personal access token into the training platform.

---

## 03 · Requirements-to-evidence matrix

Add one row per mandatory requirement. Every row must cite exact evidence. A requirement without evidence remains open.

| Requirement ID | Requirement and acceptance rule | Implementation reference | Test/eval reference | Production observation | Owner | State |
|---|---|---|---|---|---|---|
| REQ-001 |  |  |  |  |  | OPEN |

Allowed states: `OPEN`, `EVIDENCED`, `REJECTED`, `SUPERSEDED`.

---

## 04 · Architecture and decision evidence

- **System context**:
- **Trust and authority boundaries**:
- **Identity and authorization path**:
- **Data classification and lineage**:
- **Model/provider adapter boundary**:
- **External tools and write authority**:
- **Storage, retention, and deletion path**:
- **Failure, retry, idempotency, and reconciliation behavior**:
- **Architecture decision records**:
- **Threat model and abuse cases**:
- **Residual-risk decisions and owners**:

### Optional Archify evidence adapter

LockedIn may use a pinned copy of Archify `2.15.0` at commit `e1ac748f19cf805e44bf74fb93c796662152e273`, stable archive SHA-256 `156f0d3cd6ba11706344683e3d94e6b6b44b5732bd3c8c81b75c267c3f5d04c4`, under the MIT license.

Retain the typed JSON source, validation receipt, exact repository commit, and PNG preview. Treat self-contained HTML as a downloadable artifact, never trusted application markup. A valid or polished diagram does not prove that its architectural, security, or runtime claims are true.

- **Diagram type**:
- **IR SHA-256**:
- **Validation receipt reference**:
- **PNG SHA-256**:
- **Linked finding IDs**:
- **Human visual-review state**:

---

## 05 · Verification and AI evaluation

- **Unit checks**:
- **Integration checks**:
- **End-to-end critical-path checks**:
- **Negative authorization and tenant-isolation checks**:
- **Accessibility checks for critical tasks**:
- **Performance and capacity checks**:
- **Dependency, secret, static-analysis, and vulnerability checks**:
- **AI task, trajectory, outcome, safety, and operational evaluations**:
- **Held-out boundary and grader versions**:
- **Failure clusters and remediation**:
- **Invalidated checks and exact-candidate rerun**:
- **Open test gaps**:

Do not weaken a case, threshold, grader, expected result, or policy merely to turn a failure green.

---

## 06 · Release and production proof

- **CI run**:
- **Build artifact or image digest**:
- **Software bill of materials reference**:
- **Provenance reference**:
- **Release approval**:
- **Production environment ID**:
- **Production deployment ID**:
- **Production URL or approved service reference**:
- **Deployed-at time**:
- **Deployed commit and artifact reconciliation evidence**:
- **Production smoke result**:
- **Rollback trigger**:
- **Rollback or recovery exercise**:
- **Candidate release responsibility**:

The reviewed commit, checks, build artifact, deployment, and live service must form one traceable chain.

---

## 07 · Operate against real use

- **Intended non-author users or operators**:
- **Observation window or complete business-use cycle**:
- **Service-level objectives and indicators**:
- **Health evidence**:
- **Accepted-outcome evidence**:
- **Adoption or workflow-use evidence**:
- **Critical-failure and guardrail evidence**:
- **Cost and bounded-consumption evidence**:
- **Alerting and support ownership**:
- **Incident or game-day timeline**:
- **Containment decision**:
- **Recovery decision**:
- **What changed in tests, controls, runbooks, or ownership**:
- **Unresolved operating uncertainty**:

Infrastructure health alone is not proof of a useful or adopted production outcome.

---

## 08 · Ownership transfer

- **Named receiving owner**:
- **Runbook reference**:
- **Known-limits register**:
- **Open risk and debt with owners and dates**:
- **Escalation and kill-switch path**:
- **Owner-executed routine change**:
- **Owner-executed rollback or recovery**:
- **Owner teach-back evidence**:
- **Candidate access removed or reduced**:
- **Permission-bounded field-learning memo**:
- **After-action review reference**:

Transfer is incomplete when the system depends on undocumented knowledge, personal credentials, or continued hidden intervention by the candidate.

---

## 09 · Ten-gate review

Score each gate `0` through `4` and cite the exact evidence used.

- `0` — unrateable or unsafe
- `1` — asserted
- `2` — partially evidenced
- `3` — defensible and independently reproducible
- `4` — production-proven and transferable

Every gate must score at least `3`. An unsafe or unrateable critical gate is an immediate `HOLD`; strength elsewhere cannot average it away.

| Gate | Standard | Score | Evidence references | Finding IDs | Reviewer |
|---|---|---:|---|---|---|
| C01 | Requirements traceability |  |  |  |  |
| C02 | Repository craftsmanship |  |  |  |  |
| C03 | Architecture and interoperability |  |  |  |  |
| C04 | Testing and AI evaluation |  |  |  |  |
| C05 | Security, privacy, and authority |  |  |  |  |
| C06 | Release integrity |  |  |  |  |
| C07 | Operability and resilience |  |  |  |  |
| C08 | Business outcome and adoption |  |  |  |  |
| C09 | Ownership transfer |  |  |  |  |
| C10 | Integrated defense |  |  |  |  |

---

## 10 · Assessment and decision record

### Deterministic collector

- **Collector version**:
- **Collected-at time**:
- **Evidence-manifest SHA-256**:
- **Collection gaps**:

### Evidence-citing assessment agent

- **Model/provider/version**:
- **Prompt and rubric version**:
- **Proposed findings**:
- **Contradictions and missing evidence**:
- **Adversarial probes attempted**:
- **Candidate questions**:

The assessment agent is advisory. It may not issue qualification, waive a gate, resolve a reviewer conflict, deploy code, or turn missing evidence into a pass.

### Independent human review

- **Reviewer 1 decision digest**:
- **Reviewer 2 decision digest**:
- **Conflicts**:
- **Adjudication reference**:
- **Conflict-of-interest disclosures**:

### Live integrated defense

- **Requirement change injected**:
- **Failure injected**:
- **Candidate diagnosis and decision**:
- **Architecture/security/operations defense record**:
- **Panel decision digest**:

### Project review disposition

Choose exactly one:

- `PASS` — every gate is at least 3, the production minimum is verified, no critical hold remains, and the required human decision is complete.
- `REVISION REQUIRED` — safe to continue, but evidence or performance remains below the bar. A content-distinct successor must retain lineage.
- `HOLD` — an unsafe boundary, unverifiable production claim, source-custody breach, integrity concern, or unrateable critical gate stops the project review.

- **Disposition**:
- **Decision time**:
- **Decision authority**:
- **Decision digest**:
- **Required remediation**:
- **Appeal path**:

---

## Claim boundary

Completing this pack is not a credential, license, government certification, ISO accreditation, SOC audit, SOC attestation, employment authorization, client-access authorization, or guarantee of performance in another environment.

The truthful eventual claim is narrow: a named individual owned a specific, authorized capability that reached production, served real users or operators, and was independently reviewed against a named version of the LockedIn evidence standard. Professional qualification remains a separate governed decision.