Human Kernel

Browser-native, local-first knowledge environment that compiles a person's own notes into an evidence-linked model of their behavioral patterns — visualized as a dashboard, not asserted as a personality score. No install, no account, no server-side storage of personal data.

Reference implementation of Forage DeepMind's Digital Human Modeling thesis. Built for a courseware context first (Forage's Digital Human Modeling module), productization deferred until validated (ADR-0004).

Status

MVD (Minimum Viable Demonstration) in active development. Planning and specification phase is complete and Approved (Founder). All five modules — vault-reader, evidence-parser, compiler, store, dashboard — are implemented and wired end-to-end (CI badge above). Covered by unit tests with fakes/jsdom, not yet by a manual browser click-through with a real vault — see src/dashboard/README.md for the precise "what's real / what isn't yet" line. See docs/agile-backlog.md for the MVD critical path (9 stories, 37 points) and current build order.

Live build: deploys to GitHub Pages on every push to main via deploy-pages.yml → https://forageopen.github.io/human-kernel/. First deploy needs one manual, owner-only step: repo Settings → Pages → Build and deployment → Source → GitHub Actions. Nobody but the repo owner can flip that, so treat the link as "will go live once that's set," not "already live."

Changelog: see CHANGELOG.md — current milestone is v0.1.0-mvd.

Start here: docs/INDEX.md — full map of every document in this repo, what it's for, and its current status.

Core Principles

1. Extract or compile, never invent. No claim renders without a linked Evidence object.
2. Every claim needs a witness. confidence is required, not optional.
3. Open or not at all. MIT-licensed, no closed telemetry, no server dependency for core function.

Full detail: docs/project-brief-v2.md §10.

Architecture

Local Files (.md vault)
   → vault-reader
   → evidence-parser
   → compiler (Parameter Compiler / Pattern Engine / Relationship Compiler)
   → store (.human-kernel/index.json)
   → dashboard

Module boundary rule: only vault-reader touches the file system, only store touches the index file, dashboard touches neither directly. Full spec: docs/specification-v0.1.md.

Repo Layout

• docs/ — brief, spec, ADRs, process docs, planning docs. Read docs/INDEX.md first.
• schema/ — JSON Schema (2020-12) for Evidence, Parameter, Relationship, and the root index file.
• wireframes/ — clickable HTML wireframe covering the data states not shown in the original interaction-model prototype.
• src/ — application code, split along the module boundaries above. All five modules are implemented and wired end-to-end; dashboard is real code now, not just a README (though src/dashboard/README.md still documents the honest gaps). See docs/agile-backlog.md for the exact build order.

Contributing

See CONTRIBUTING.md — every PR is checked against the three principles above, not just style.

Credits

• Adam Rosman — Founder, Forage DeepMind. Product direction and every Founder Override decision recorded in this repo (see docs/INDEX.md).
• Claude (Anthropic) — AI pair-programmer. Drafted the docs, schema, and everything currently in src/ under the Founder's direction. Flagging the honest mismatch rather than papering over it: GitHub's "Contributors" graph is commit-authorship-based and tied to a real account, which Claude doesn't have — so this line is the accurate record, not that graph. Commits authored with Claude's involvement carry a Co-Authored-By: Claude <noreply@anthropic.com> trailer starting from this one.

License

MIT — see LICENSE.