Human-governed decision infrastructure for Frontier Mission Assurance.
Stable Decision Lab reference surface. The public root begins with a single natural-language decision input before any formal comparison.
The public root asks one question: What are you considering?
Type or paste relevant context in ordinary language and choose Continue. For ordinary in-scope material, FDE returns a working Decision Brief immediately rather than beginning a compulsory clarification sequence. The brief separates the decision or focus, what matters, what remains unclear, and the next useful move. Sparse or multi-decision context can still produce a partial brief; unsupported information remains explicit instead of being silently invented.
The original source text remains preserved as working context. Candidate decisions, choices, criteria, and conditions may be broader than the formal model. When the person voluntarily opens Compare options, they select the bounded set for one deterministic comparison: 2–4 objectives, 2–3 strategies, and 2–4 scenarios. This bounds the computation without silently discarding the source context.
When explicit language identifies a requirement, time gate, dependency, conditional, or blocker that may control the decision, FDE can surface a provisional decision hinge with source provenance. It does not affect formal comparison inputs until the person explicitly keeps it or leaves it uncertain during comparison setup.
The browser does not call a remote AI provider, retrieve outside facts, invent evidence, probabilities, scores, thresholds, scenario effects, or recommendations.
The Governed context route accepts Mission Graph Decision Context Packet 0.3.0 for FDE_PREPARATION_ONLY use. FDE verifies payload and envelope SHA-256 integrity locally while keeping integrity, origin, evidence assurance, freshness, and human authority as separate trust properties.
Only PRIVATE or PROTECTED packets with explicit non-release handling are accepted. A self-asserted authenticated origin is rejected unless a future external trust-policy verifier can independently establish it; a valid digest proves integrity, not authorship. Future-dated context fails closed. Review-due, expired, or freshness-not-established context remains inspectable but cannot enter active decision preparation. Legacy Decision Context Packet 0.2.0 remains inspection-only and is never silently upgraded to current trust semantics.
Accepted context stays in JavaScript page memory only and is discarded on refresh; it is not placed in normal browser autosave or silently promoted into the canonical decision model. Source lineage and packet identity remain behind Show me why. The accountable human remains the decision authority.
FDE is the human-governed decision layer within Bridge Node 7's Frontier Mission Assurance architecture. Materials-to-Mission and other assurance profiles can contribute governed evidence and dependency context; Governed Context carries bounded preparation context into FDE; the accountable human records the consequential decision and reassessment. Formal comparison is optional. A single comparison-setup surface prepares the bounded explicit inputs required by the existing deterministic Decision Lab. The Decision Lab preserves the existing ranking semantics, published schemas, saved-work protections, and truthful alternate outcomes including ties, no acceptable option, and insufficient information.
The comparison informs. A person decides. The result surface leads with what held up, why, what could change it, the next required proof, and the next action before exposing detailed calculations. A lightweight Decision Brief can be copied or downloaded before recording; when the working decision later changes, FDE shows a deterministic reassessment delta and preserves prior valid Receipt v2 records in a bounded Local Decision Receipt Archive before a new Receipt is recorded. The archive preserves independently verifiable local receipts; it is not a cryptographic continuity chain.
Technical users can inspect the underlying inputs, assumptions, evidence state, uncertainty, thresholds, scenario effects, calculations, provenance, and decision record without forcing that depth into the first-run experience. When sustainability is relevant, an optional SEER reminder prompts consideration of People, Planet, Profits, and Product as independent criteria. It supplements the decision rather than changing the comparison, and it preserves the same evidence, posture, monitoring, reassessment, and human-authority model.
Technical documentation and exports may still use Decision Map and decision-science terminology where those terms improve precision; first-time users do not need to learn that vocabulary before receiving value.
The repository includes a synthetic critical-material source-qualification case to demonstrate the formal Decision Lab without making claims about a real supplier, material, capacity, compliance status, investment, or forecast.
The application is static and browser-local. It has no backend, account system, analytics, telemetry, cookies, remote AI provider, or default upload endpoint. Session recovery and browser autosave are convenience features for ordinary FDE drafts, not encrypted confidential storage. Governed Mission Graph context uses a separate memory-only path.
FDE provides decision support. It does not approve, authorize, certify, qualify, consent, authenticate unsigned evidence, or make an investment or other consequential decision. Human final decision authority is preserved. Scope policy is re-evaluated before accountable recording and before imported decisions become active work. First-person personal-life decisions are redirected by a neutral scope boundary unless they are clearly framed as technical, organizational, mission, or strategic work; that redirect is not a clinical judgment.
Requirements: Node.js 22+, Python 3.11+, and Chromium or Google Chrome.
macOS / Linux:
node scripts/run-python.mjs -m venv .venv
source .venv/bin/activate
node scripts/run-python.mjs -m pip install -r requirements-dev.txt
node scripts/run-python.mjs -m playwright install chromium
npm ci --ignore-scripts --no-audit --no-fund
npm run checkWindows PowerShell:
node scripts/run-python.mjs -m venv .venv
.\.venv\Scripts\Activate.ps1
node scripts/run-python.mjs -m pip install -r requirements-dev.txt
node scripts/run-python.mjs -m playwright install chromium
npm ci --ignore-scripts --no-audit --no-fund
npm run checkGenerated release facts and current application identity are recorded in project-facts.json. The full verification gate is defined by the repository's npm run check command.
main is the current reviewed source. GitHub Releases are stable distribution milestones and may intentionally lag reviewed maintenance on main. A source version ahead of the latest GitHub Release is unreleased until a release is published. The GitHub Release publication timestamp is the authoritative stable-release date; reviewed source does not claim a stable release date before publication.
Changes must preserve human decision authority, privacy, accessibility, and the complete validation gate. See CONTRIBUTING.md.
Report vulnerabilities through GitHub private vulnerability reporting as described in SECURITY.md.
Apache-2.0 licensed. See LICENSE.