Journal

Provenance v0.3 is active. The governance field changed from a single object to an array of policy references. Real-world artifacts live under multiple concurrent governance policies — PCI-DSS + internal AI usage, environmental permits + change management, university AI policy + journal disclosure. v0.2 forced a choice between losing traceability or losing structure. v0.3 makes multi-policy governance first-class.

v0.2 was promoted to active, then immediately superseded by v0.3. All domain examples updated to v0.3.

Ninth domain example: open source project governance. Priya Chandrasekaran, a lead maintainer of a Rust permissions library, rejected an AI-recommended unsafe_inner() API and designed a GroupPolicy trait instead — because the unsafe API would have required callers to reason about safety invariants the library should enforce internally. Brings the total to nine domains.

Self-provenance bootstrapping (plan 004). Each spec version — v0.1, v0.2, v0.3 — now has its own .halos.json provenance record. The spec validates against its own schemas. The provenance records validate against the spec they describe. If HALOS defines how to record provenance, its own history should be the first reference implementation.

Schema test suite. Scenario-based tests: enum coverage, schema evolution, semantic validation. CI catches drift between the spec and its schemas. Agent bootstrap files added alongside the tests to make onboarding easier.

Example generation workflow (plan 003). GENERATE-EXAMPLE.md is an agent skill that produces new domain examples — narrative + provenance record — given a domain. Lowers the barrier for contributors.

Terminology page enriched. Added provenance vs lineage comparison and deeper explanations of the three-layer model. The distinction matters: provenance is the complete history; lineage is the chain of derivation within it.

Site updates. FAQ page, homepage storytelling redesign, examples section pulling from halos-spec.

Next: tagged release, software-dev domain profile, Phase 2 adoption tooling.

The name is settled. HALOS is the Human–Agent Lineage and Origin Standard. “Living Operating System” is retired — it read as runtime software and didn’t position well alongside CycloneDX, SLSA, or W3C PROV. “Lineage and Origin” maps to both layers: the Principles define an ethical origin framework; the Provenance Spec captures lineage of work. The full name now appears consistently across the main site, halos-spec, and northharbor.dev.

Two layers, made explicit. The framework has always had two distinct, versioned parts — they just weren’t surfaced:

• HALOS Principles — v1.0, Stable. The normative foundation: human primacy, attribution, transparency of AI involvement, and ethical guardrails. These do not version; they anchor everything else.
• Provenance Spec — v0.1, Active. The technical standard for recording who the accountable human is, what AI contributed, and whether a human reviewed an artifact before use.

The main site now leads with both, with version numbers, rather than treating them as a single undifferentiated framework.

Where HALOS fits. The ecosystem comparison — previously buried in the halos-spec README — is now front-and-center on the main site:

• vs. CycloneDX/SBOM — CycloneDX answers “what’s in this software?” HALOS adds “who was responsible and what AI contributed.” HALOS provenance embeds directly as component.evidence.
• vs. SLSA — SLSA answers “how was it built?” HALOS adds “what decisions were made and by whom.”
• vs. W3C PROV — The v0.2 graph model aligns with W3C PROV (Entities, Activities, Agents) but uses plain JSON — no RDF required.
• vs. NIST AI RMF / ISO 42001 — Governance frameworks define policy. HALOS operates at the artifact level inside those processes.

Eight domain examples. halos-spec now includes eight real-world scenario narratives, each with a .md account and a .halos.json provenance record using the v0.2 graph model:

• Journalism — Carlos Medina (investigative reporter) corrected an AI-generated violation dataset before publication
• Enterprise software — Tomoko Hayashi (platform engineer) rejected an AI-recommended CQRS architecture and rewrote the state machine from scratch
• Scientific research — Dr. Nkechi Okonkwo (computational epidemiologist) excluded a confounded feature the AI had ranked as highly predictive
• Real-time critical systems — Luis Herrera, PE (control systems engineer) reduced an AI-proposed dosing recommendation and added safety logic the AI had omitted
• Government policy — Fatima Al-Rashid (policy analyst) determined how to frame a politically consequential housing supply finding
• Education — Mateo Rivera (undergrad) rejected an AI-fabricated citation after verification failure
• Music production — Maya Reeves (composer) rejected an AI harmonic sketch, keeping only the structural idea
• Nonprofit / humanitarian — Amara Diallo (field coordinator) overrode an AI priority ranking based on ground-truth field data

Each narrative covers the scenario, collaboration process, artifact description, AIVSS-style risk interpretation, and framework crosswalk (NIST AI RMF, ISO/IEC 42001, CycloneDX).

Coming next:

• First tagged release of halos-spec — the March reorganization is complete, a release is overdue.
• halos/spec/ content replaced with redirect stubs pointing to halos-spec (deferred from the March migration).
• software-dev domain profile fleshed out — full SLSA + CycloneDX + Chainloop mapping.
• Phase 2 adoption: introducing .halos.json provenance records as the reference implementation of CORE-3 and CORE-4.

The halos site updated in step with the halos-spec launch.

Whitepaper published. The HALOS whitepaper — positioning, motivation, and architecture — is now a first-class page on the site, linked from the homepage. It was revised to be implementation-agnostic before publishing.

halos-spec surfaces throughout. The spec page, governance page, and Getting Involved section now point directly to halos-spec: adoption guide, templates, schemas, and agent prompt. Visitors who want to implement or adopt HALOS have a clear path.

Explore UX. Copy rewritten for non-technical audiences — “Who are you?” / “What’s on your mind?” in place of category labels. The community illustration moved into the Explore frame as a welcoming visual. Mode icons redesigned with branded assets, always stacked vertically for consistency at every screen size.

Agent discovery simplified. FOR_AGENTS.md and AGENTS.md consolidated; external links now point to halos-spec rather than duplicating content.

halos-spec is now the single source of truth for everything normative. The separation is real: halos holds the community site, narrative, and governance process; halos-spec holds the machine-readable spec, schemas, adoption toolkit, and examples.

What landed today:

• Adoption toolkit — adopt/GUIDE.md, templates for HALOS-ADOPTION.md and halos.yaml, and adopt/AGENT-PROMPT.md so any AI agent can adopt HALOS in a new repo without human hand-holding.
• halos-profile schema — spec/schema/halos-profile.schema.json formally defines the halos.yaml governance profile. Adopters and agents can now validate what they generate.
• Spec migration — manifest.json, core.json, and changelog.json moved from halos into halos-spec. spec/CANONICAL.md declares this the authoritative source.
• Generated principles — scripts/generate-principles.js renders core.json to PRINCIPLES/halos-principles-v1.0.md. The principles are no longer maintained by hand.
• Domain profiles scaffolded — profiles/software-dev/ provides a SLSA/CycloneDX/Chainloop mapping. Other domains can follow the same pattern.
• CI validation — GitHub Actions workflow validates core.json, manifest.json, changelog.json, provenance examples, and drift between generated principles and core.json on every push.
• FOR_AGENTS.md — outward-facing agent discovery entry point for any agent working in halos-spec or adopting HALOS externally.

Two rounds of site polish landed between March 18–20. NorthHarbor parent branding added to the nav: beacon icon, wordmark, and separator give HALOS its organizational home without competing with the primary navigation. Reading controls moved from the nav to a fixed left-edge tab — cleaner on mobile and desktop alike.

Messaging sharpened on the homepage: the hero now leads with AI governance urgency and human agency rather than framework description. “Everyday Humans” renamed to “For Everyone” throughout — a small change that opens the door wider.

Signatory registry implemented: signatories.json, CONTRIBUTING.md, validation script, CI checks. Supporters page with table layout and pagination (first 100, 25 per page). Badges for Founder, Founding Supporter, Early Supporter. Bob Hong and HALOS Agent added as founding signatories.

Planned: Proposals 0001–0003 under review. HALOS Index and signatory web form when backend exists. Organizational signatories, cryptographic verification, trust graph on the horizon.

Proposal 0004 submitted: HALOS Verify — a conformance test suite mapping HALOS-CORE-1 through HALOS-CORE-8 to evaluable scenarios. Today, alignment is self-declared; Verify introduces a shared methodology so adopters, researchers, and tooling can measure and compare HALOS adherence objectively. An optional public benchmark would enable rankings, discovery, and data syndication. Status: draft, open for community input.

Homepage section and footer updated to clarify engagement paths. New Getting Involved block between Dive deeper and Context: four cards—Read the Spec, Become a Signatory, Contribute (GitHub), Follow Updates (LinkedIn). Footer now links to GitHub and NorthHarbor LinkedIn company page. Plan documented in docs/_proposals/GETTING-INVOLVED-PLAN.md.

Three proposals submitted: 0001 (Explore input sanitization—allowlist, Unicode NFC, WAF), 0002 (HALOS Index—persist anonymized roles/concerns for trend analysis), 0003 (Signatory registry). Explore guardrails module added: sanitize, blocklist, low-quality detection. Unit tests for guardrails.

Explore flow shipped: visitors choose role and concerns, get a HALOS-seeded prompt and deep links to Claude, ChatGPT, or copy. Help system, LEARN_HALOS prompt, Docker build with live reload. Agent discovery files and agent-convention-sync skill added so AI agents can adopt HALOS when working in this repo. Reading controls (text size, page width) and mobile-friendly layout.

Today we launch the HALOS journal: a living history of the project. Here we will record decisions, milestones, contributions, and progress—in the project’s own voice—so that anyone can follow how HALOS evolves and why.

This is more than a changelog. The spec changelog tracks versions and technical changes. The journal tells the story: what we chose, what we learned, and who helped along the way.

Initial HALOS framework landed: vision, principles, governance, and the proposal process. Machine-readable spec v1.0.0 added, GitHub Pages configured for halos.northharbor.dev, identity mark and rationale published. HALOS went from document to addressable resource.