Credibility evidence register

Register ID: LL-FDE-CRED-001
Version: 0.1.0
Status: public claims-control register; not a validation, certification, or conformance report
Evidence cutoff: September 1, 2026
Current benchmark line: 0.1.x Public Draft
Official numeric scoring: disabled

1. Purpose

This register answers a narrow question:

What evidence exists today for each credibility signal the LockedIn FDE Benchmark might display, and which public claims does that evidence permit?

It prevents visual polish, citation volume, software completion, roadmap language, or unverified commissioned research from being mistaken for measurement authority.

The controlling sources are the specification, benchmark card, scoring architecture, governance policy, independence and institutional firewall, data governance policy, source ledger, study register, benchmark formation protocol, and Live Versus Planned register.

2. State model

Each register item receives one control state and one evidence state.

2.1 Control state

2.2 Evidence state

source_available does not mean deployed, independently reviewed, legally cleared, validated, or fit for use. synthetic_tested does not establish human reliability, fairness, or validity.

3. Current public release-health snapshot

The following values may be displayed only with the stated context and evidence cutoff. Zero, absent, and blocked values belong in this release-health or transparency record. They MUST NOT be promoted as proof of rigor, framed as achievement metrics, or used as the landing-page hero value proposition.

Targets, proposed sample sizes, planned studies, roadmap milestones, and synthetic examples MUST NOT be added to this snapshot as completed values.

4. Foundation and construct register

5. Research and validity register

6. Reproducibility and publication register

7. Governance, independence, and operations register

8. Controlled public wording

8.1 Permitted now

• “LockedIn FDE Benchmark v0.1 Public Draft.”
• “A proposed evidence-oriented framework for demonstrated Forward Deployed Engineering capability.”
• “The public-draft ontology contains 10 domains and 55 capability families.”
• “The framework is designed to connect claims to observed work, artifacts, outcomes, corroboration, and human review.”
• “Official numeric scoring is disabled, examples are synthetic, and no verified credential is issued.”
• “A dated source ledger records the standards, research, and benchmark precedents that informed design decisions.”
• “Job analysis, calibration, rater, fairness, accessibility, criterion, and longitudinal studies are planned and not yet established.”

8.2 Prohibited until evidence state changes

• “Validated,” “predictive,” “scientifically proven,” “unbiased,” “fair,” “independently validated,” or “independently reproduced.”
• “Certified,” “accredited,” “official FDE qualification,” “industry standard,” or “authorized hiring benchmark.”
• “Based on interviews with 500 practitioners and 50 executives,” “surveyed 550 leaders,” or any paraphrase implying admitted research evidence.
• “Less than 5% of FDEs pass,” any pass-rate claim, percentile, comparison-population result, readiness threshold, or score band.
• “First,” “only,” “best,” “definitive,” or “category creator” without fresh, appropriately scoped substantiation.
• Outside endorsement, partnership, governance, peer review, sponsor, or signatory claims without signed authorization.
• Compliance with a law, standard, or certification scheme without the designated authority’s decision.

9. Evidence receipt schema

Every future status change in this register SHOULD be backed by a receipt containing:

receipt_id: LL-FDE-CR-0000
register_item_id: CR-X00
claim_text_exact: "..."
evidence_state_before: not_established
evidence_state_after: source_available
control_state_after: adopted_now
evidence_cutoff: YYYY-MM-DD
artifact_ids:
  - type: study|source|test|review|decision|archive|incident
    id: "..."
    version: "..."
    digest: "sha256:..."
    uri: "..."
population_and_use_scope: "..."
independent_review:
  status: none|planned|complete
  reviewer_record: "..."
limitations:
  - "..."
claims_enabled:
  - "..."
claims_still_blocked:
  - "..."
approved_by:
  - "..."
effective_at: YYYY-MM-DDTHH:MM:SSZ
next_review_at: YYYY-MM-DD

The public register may omit protected signatures or paths, but MUST retain their existence, authority, digest, and disclosure rationale.

10. Go-live gates by public meaning

“Go live” has four different meanings. The interface and release decision MUST name which one applies.

No later gate may be implied by activating an earlier one.

11. Priority evidence backlog

Publication readiness

1. Approve individual authors, affiliations, contributor roles, and accountable guarantor.
2. Finalize funding/in-kind and conflict/recusal statements.
3. Approve operative code, content, data, and third-party licenses.
4. Create and validate CITATION.cff, BibTeX, canonical metadata, and version history.
5. Establish a public archive and exact-version DOI; verify checksum parity with the approved manifest.
6. Operate monitored correction, security, and private conduct channels.
7. Record the formal publication decision for the exact commit and package.

Measurement credibility

1. Resolve FDE-JA-001 through evidence intake or close it without using the 500/50 claims.
2. Authorize and preregister a prospective job-analysis and adjacent-role study if the legacy package cannot satisfy admission requirements.
3. Publish the source-to-job-activity-to-blueprint traceability package and expert-panel record.
4. Pilot task solvability, response process, burden, accessibility, accommodations, and contamination controls.
5. Train raters and study agreement, reliability, drift, disagreement, adjudication, and decision consistency.
6. Predefine calibration, uncertainty, missingness, critical-floor, and standard-setting studies before any human score proposal.
7. Conduct fairness, subgroup/error, consequence, and later criterion/longitudinal studies before expanding use.
8. Commission non-conflicted independent measurement, accessibility, privacy, security, and disclosure reviews.
9. Seat the external-majority Commission and publish its roster, independence determinations, terms, rules, conflicts, funding, and first valid decision record.
10. Implement and test the training/employment/staffing-to-assessment role, data, compensation, publication, and appeal firewall.
11. Publish the structural-entity decision before any official human output or credential proposal advances to release authorization.

12. Maintenance

• The Research Director owns evidence intake; the Claims Reviewer owns exact public wording; the Release Authority owns status changes. One person MUST NOT approve all three roles for an empirical or high-consequence claim.
• Status changes are append-only. Corrections link prior and replacement states rather than overwriting history.
• Public metrics refresh on every material evidence event and receive a no-change review at least quarterly.
• Living sources are checked at the cadence in the source ledger and before each release.
• A stale or unreachable receipt does not automatically invalidate the underlying work, but it places the affected public claim under review until provenance and availability are restored.
• If this register conflicts with marketing copy, a roadmap, product interface, or unpublished assertion, this register and the controlling source documents determine the permitted current claim.

13. External reference boundary

The publication and evolution controls draw operating lessons from MLCommons submission rules, NIST AI 800-2, METR RE-Bench, Artificial Analysis methodology, OSF registrations, CRediT, and Zenodo versioning.

Those sources support design discipline, transparent methods, packaging, review, archiving, and attribution. They do not provide evidence that the LockedIn benchmark measures FDE capability validly, and no endorsement or conformance is claimed.