Program
Publicus-Cybersecurity Reference Plan Snapshot
Status: reference. Scope: this repository only - registry corpus, evidence API, public trust docs, provenance posture, movement feed, source contract, crosswalks, and contribution model. Active operating truth lives in the docs router, current-state file, decision log, and lineage index.
Governing principle: Build at AI pace. Verify cybersecurity at AI pace in public.
Executive summary
publicus-cybersecurity is the open trust substrate for AI software assurance. It makes trust decisions inspectable at the same speed that AI-generated software is built, modified, and deployed.
The durable goal is to move from a strong live prototype into canonical open infrastructure that is narrow enough to audit, useful without private vendor access, primary-source-cited, machine-readable, forkable, and defensible to government, enterprise, and open-source reviewers.
Core thesis
The agentic economy has made code generation abundant. Trust remains scarce. Teams can now generate application code, change dependency graphs, wire agents to more tools, and deploy before human reviewers fully understand what changed.
Publicus answers the hard trust questions in public: which components are acceptable, which versions are constrained, what public evidence supports the decision, what changed upstream, what provenance exists, how fresh each evidence lane is, and how technical facts map to institutional assurance language.
Manifesto
Trust must be public.Opaque trust does not scale across governments, suppliers, developers, and independent auditors.
Trust must be machine-readable.AI-built systems move too quickly for manual review alone.
Trust must be primary-source-cited.Cybersecurity claims decay when they depend on summaries, sales language, or untraceable model output.
Curated truth and live enrichment must stay separate.EPSS, Scorecard, Rekor, source observations, and model explanations can change urgency, but they cannot silently rewrite the registry-of-record.
Freshness must be multiple clocks.Registry review, projection refresh, source sync, and published movement are different claims.
Vision
Publicus-Cybersecurity should function as the canonical open trust substrate for AI software assurance.
Publish signed component truthStatus, version constraints, advisory evidence, provenance expectations, and rationale remain inspectable.
Expose read-only evidence APIsHumans, CI, agents, downstream mirrors, and public auditors consume the same public evidence.
Surface movement in publicAdvisory drift, source/context movement, provenance changes, and freshness lanes are visible instead of implied.
Publish crosswalk artifactsRegistry decisions become easier to interpret in governance and assurance language without claiming certification.
Stay forkable and independently usefulCommercial workflow layers live outside this repo. The open substrate remains useful on its own.
Product boundaries
This repo is the open registry corpus, read-only trust API, public docs, source contract, live observations, movement feed, crosswalk layer, evidence-first browse surface, and explicit Ask Publicus workspace.
It is not a broader commercial workflow platform, not a generic scanner, not a private procurement workflow application, and not an AI assistant whose answer is the source of truth.
Current baseline
Live trust API and shellFastAPI service with public routes for overview, dependency scan, registry, exploited risks, Ask, human docs, and API docs.
Evidence corpus100 tracked components, advisory and exploited-risk surfaces, expanded source contracts including NVD, GHSA, CVE List V5, OSV, EPSS, deps.dev, PyPI JSON, checked source observations, movement feed, and first NIST AI RMF crosswalk.
Freshness and enrichmentFour freshness lanes, EPSS, Scorecard, registry-native provenance expectations, and Rekor corroboration where entries publish a log index.
Ask PublicusLocal evidence-first synthesis with optional bounded recommendation support. Ask remains subordinate to signed evidence and citations.
OperationsHosted public app, durable projection store, seed command, and distinct web/worker behavior.
Architecture model
Layer 0: raw mirrorsFetch public sources, preserve raw payloads, hashes, provenance URLs, fetch time, source date, and license posture.
Layer 1: normalized evidenceConvert source material into common advisory, exploit-likelihood, project-health, provenance, policy, procurement, and standards schemas.
Layer 2: curated registry truthReviewed component status, allowed or blocked versions, rationale, provenance expectations, and decision lineage.
Layer 3: contextual decision supportAsk Publicus and explain lanes combine curated truth with live evidence while keeping citations and synthesis posture inspectable.
CrosswalksFirst-class artifacts mapping registry facts to frameworks and institutional assurance language.
Outcome gates
Undeniable substrateKeep security headers, docs split, Publicus shell, evidence-first browsing, explicit Ask workspace, mobile layout, route smoke tests, and canonical docs discipline healthy.
Fresh and dependable evidenceKeep web and worker behavior distinct, make projection refresh reproducible, broaden movement beyond advisory-only drift, and keep the freshness clocks honest.
Deep evidence and crosswalksAdd only high-signal evidence sources with fixtures, expand AI-specific risk where generic vulnerability data is insufficient, and publish precise machine-readable crosswalks.
Institutional resilienceMature governance, reviewer-council rules, transparency reporting, bad-actor PR drills, and possible foundation transition without slowing current delivery.
Workstreams
Trust corpusReview component entries, source citations, version constraints, provenance expectations, and category boundaries.
Source ingestion and freshnessMaintain stdlib-only clients where required, fixtures, live observations, and honest source limitations.
Public product surfaceKeep routes clear, mobile-safe, accessible, and evidence-first.
Ask Publicus synthesisKeep recommendation support bounded, transparent, citation-first, and non-authoritative.
Standards and Canada contextKeep Canada-context source families visible and translate registry decisions into assurance language without overclaiming compliance.
Governance and contributionKeep contribution rules public, docs consolidated, reviewer posture visible, and promotion paths explicit.
Public interfaces
The core public API includes /meta, /status, /components, /advisories, /kev, /drift, /scan/manifest, /evidence/events, /evidence/package/{ecosystem}/{package}, /review/proposals, /sources, /sources/live, /crosswalks, /context/*, /assist/query, /assist/query/stream, and /explain.
The trusted refresh path records registry_evidence_checked_at separately from the curated registry baseline date, so the public currentness lane can be fresh without pretending every status decision changed that day.
The primary human routes are /, /scan, /registry, /known-exploited, /ask, /docs, /docs/help/, /docs/source-contract/, /docs/weekly-brief/, /docs/canada-mapping/, /docs/governance-model/, /docs/crosswalks/nist-ai-rmf/, and this page.
Verifiable outcomes
Trust substrateLive trust docs, evidence-first registry and exploited-risk views, source contract, live observations, movement feed, provenance expectations, freshness lanes, and current crosswalks.
OperationsDistinct web and worker roles, reproducible projection refresh, timestamps that move when data moves, published public app, and public smoke routes returning 200.
Institutional proofGovernance model, contribution model, open trust doctrine, reviewer-council posture, transparency page, Canada mapping, and public plan of work are visible without private access.
Evidence qualityPrimary-source discipline, fixture-backed source clients, no invented identifiers, scoped crosswalk language, and citation-backed Ask or explain outputs.
Validation gates
Every material release should keep tests, lint, formatting, canonical docs audit, public-page audit, and dogfood checks green.
UI smoke covers overview, registry, exploited risks, Ask, docs, plan, desktop layout, 390px mobile layout, no unintended horizontal overflow, one visible header logo, and evidence-first drill-down behavior.
API smoke covers meta, status, components, exploited-risk API, drift, sources, live sources, crosswalks, assist query, and streaming assist.
Risk controls
Source sprawlControlled through one source contract, fixtures, source-family explanation, and no promotion for sources that cannot be refreshed or cited.
Stale documentationControlled through current-state, decision-log, lineage, docs audit, and one canonical plan.
AI overreachControlled by evidence-first browsing, explicit Ask workspace, citations, synthesis posture, and fail-closed explain behavior.
Freshness overclaimControlled by separate check time, upstream publication time, projection refresh, and movement timestamps.
Outcome sequence
Clean product journeyKeep visual QA clean, review stale component prose, strengthen Publicus-on-Publicus evidence, improve source refresh visibility, and keep synthesis scoped and fast.
Broader live evidenceDeepen OSV, EPSS, Scorecard, and Rekor where useful; improve weekly brief generation; expand smoke automation; prepare reviewer-council operating rules.
Crosswalk and AI-specific depthPublish the next precise crosswalk, expand AI-specific evidence, add provenance/VEX-style signals where safe, and publish Publicus-on-Publicus evidence.
Governance maturityMature reviewer council, transparency cadence, foundation path, private mirror support, and verified/badge semantics only if core evidence remains stable.
Canonical source
The long-form reference source lives in docs/publicus-cybersecurity-2026-plan-of-work.md. Current operating truth lives in docs/current-state.md, docs/decision-log.md, and tracking/lineage.yaml.
The public docs surface intentionally links to curated routes only, so stale generated archive pages do not become public canon.