Frontier Intelligence Workflows

Trace frontier-technology claims to evidence before the decision moves forward.

Frontier Intelligence Workflows (FIW) is an open, evidence-first method for consequential technology questions. It traces claims to sources, separates observation from inference, exposes contradictions and unknowns, and records the evidence needed before a decision moves forward.

Latest release · Quickstart below · Partner

Core workflow

Decision Question
  → Claims → Evidence → Provenance
  → Contradictions / Challenges
  → Knowns · Unknowns → Assumptions
  → Assessment → Decision Boundary
  → Next Evidence → Accountable Human

Decision artifacts

• FIW Decision Record — portable record for a bounded assessment: context, claims, evidence, provenance, uncertainty, decision boundary, next evidence, accountable owner, and rationale.
• Decision-Ready Brief — human-facing handoff: supportable disposition or next move, strongest countercase, decision hinge, criteria and tradeoffs, minimum sufficient commitment, material unknowns, stop conditions, and reconsideration triggers. Advisory preparation; the accountable human decides.

Real case inputs and generated results remain in the authorized working environment unless explicitly approved for public release.

Explore

Path	What it shows
Frontier Claim Experience	Local-first synthetic example: apparent corroboration that shares one provenance root
Frontier Technology Diligence	Fictional diligence scenario: repeated reporting vs independent corroboration
Perception Integrity	Deterministic checks for declared lineage, observation vs inference, assumptions, stop conditions, and decision authority
Language Integrity	Preserves declared source meaning: forecast vs result, attribution, inference labeling, provenance-root collapse, supersession
Source Genealogy	Collapses declared derivatives, mirrors, translations, syndication, and coordinated publication to provenance roots
Frontier Signal Integrity	Frontier-signal genealogy, surprise hypotheses, prediction integrity, discriminating tests, update receipts, Proof Loop reassessment
FIW-SYN-001 — Synthetic Component Readiness	Complete evidence-to-decision artifact chain using fictional data

Status semantics

• Perception / Language Integrity: NO_FINDINGS / REVIEW_REQUIRED describe configured deterministic findings in the declared assessment.
• Repository controls: PASS / FAIL / NOT_RUN describe whether a named machine control executed and satisfied its assertion.

See Status Semantics.

Use FIW

Use a separate working directory in the environment authorized for assessment inputs and generated results.

macOS, Linux, or Git Bash

mkdir -p ../fiw-work ../fiw-results
cp templates/perception-integrity-starter.json ../fiw-work/assessment.json

python scripts/run_perception_integrity.py \
  --root . \
  --assessment ../fiw-work/assessment.json \
  --evaluated-at "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
  --json-output ../fiw-results/validation.json

Windows PowerShell

New-Item -ItemType Directory -Force ../fiw-work, ../fiw-results | Out-Null
Copy-Item templates/perception-integrity-starter.json ../fiw-work/assessment.json
$evaluatedAt = (Get-Date).ToUniversalTime().ToString("yyyy-MM-ddTHH:mm:ssZ")

python scripts/run_perception_integrity.py `
  --root . `
  --assessment ../fiw-work/assessment.json `
  --evaluated-at $evaluatedAt `
  --json-output ../fiw-results/validation.json

The first run is expected to produce findings and exit with code 1. Improve the declared record and rerun. Preserve prior artifacts when reassessment is needed.

CLI result contract

• Exit 0: READY_FOR_HUMAN_REVIEW
• Exit 1: deterministic findings require action before ordinary review
• Exit 2: input, path, or output handling error
• Exit 3: assessment fails the bounded schema contract

Verification

Use python3 (3.11+). Keep the virtualenv outside the clone.

python3 -m venv ../fiw-validation-venv
source ../fiw-validation-venv/bin/activate  # Windows PowerShell: ..\fiw-validation-venv\Scripts\Activate.ps1
python3 -m pip install -r requirements-validation.txt
mkdir -p ../fiw-validation ../fiw-release
python3 scripts/run_tests.py --root . --json-output ../fiw-validation/source-test-summary.json
python3 scripts/compile_sources.py --root .
python3 scripts/validate_repo.py --root . --json-output ../fiw-validation/validation-report.json
python3 scripts/build_release.py --root . --output-dir ../fiw-release --check --commit "$(git rev-parse HEAD)"
sha256sum -c MANIFEST.sha256
git diff --check

Assurance boundary

Automation checks structure, traceability, and declared evidence conditions. Scientific truth, readiness certification, investment decisions, and authorization to act remain with accountable human processes. See boundary notes.

Security

Private vulnerability reporting via GitHub. See Security.

License

MIT License. Published by Bridge Node 7.

Release

Distribution milestones live on the repository Releases page and in CHANGELOG. main is the current reviewed source.