Call for field testers: shared human-AI work receipts

[HarperZ9]

Project Telos is looking for field testers for telos: a shared perceive-and-make surface where a person and model can inspect the same artifact and produce a re-checkable certificate.

The question this repo should help answer:

Can human-AI work produce a visible receipt trail that is useful in real domains, not only in demos?

Useful test lanes:

• Doctor/admin: summarize and check public policy or guidance material. Do not use private patient data.
• Artist/studio: track transforms, exports, gallery artifacts, and human-vs-automatic decisions.
• Media/newsroom: separate supported claims, source links, assumptions, and unverifiable claims.
• Token economy/routing: record which tool/service was chosen, why, what it cost, and what receipt came back.
• Reasoning workflows: compare a model's explanation against shared witnessed state.

What I am looking for:

• Verification and testing against real public workflows.
• Clear examples where the receipt trail helps a human make a safer decision.
• Cases where the UI/report hides too much, exposes too much, or labels uncertainty poorly.
• Early traction from doctors/admins, artists, journalists, engineers, researchers, and builders who need checkable AI-assisted work.
• Pointers to modest grassroots research funding or labs that would stress-test this properly.

Project links:

• Main site: https://harperz9.github.io
• Field guide: https://harperz9.github.io/field-guide.html
• GitHub profile: https://github.com/HarperZ9
• Related flagships: https://github.com/HarperZ9/gather https://github.com/HarperZ9/index https://github.com/HarperZ9/forum https://github.com/HarperZ9/crucible

Stage label: public, solo, early, pre-revenue, not independently audited. Please test with public, synthetic, or non-sensitive material first.

[HarperZ9]

Current state update for field testers, 2026-06-29:

The public MCP surface is now easier to exercise from a source checkout:

• npm start or node demo/telos-mcp.mjs runs the Telos stdio MCP server.
• node demo/catalog.mjs --summary lists the current tool surface.
• node demo/server-manifest.mjs --summary maps the five-server MCP launch plan across gather, index, forum, crucible, and telos.
• node demo/action-receipt.mjs shows the external-action receipt contract with intent, admission, execution, review, compensation, trace joins, and typed failure codes.
• node demo/context-pack.mjs shows a bounded context packet with source refs, hashes, budgets, and verdicts for large-codebase handoffs.
• node demo/creative-engine.mjs --summary and node demo/creative-kernels.mjs --summary cover the creative/graphics lane: dithering, pixel sorting, plotter paths, clustered-light bins, sound/typography/media hooks, and measurement-backed demos.
• node demo/model-foundry.mjs --summary covers the model-foundry lane: hosted frontier APIs where appropriate, local/open-weight runtimes where useful, post-training experiments, and promotion gates.

The most useful testing pressure right now is not a generic star or launch upvote. It is small public/synthetic workflows where someone can say whether the receipt made the work easier to trust:

• Did the packet preserve the source, transform, decision, and verifier boundary?
• Did it distinguish MATCH, DRIFT, and UNVERIFIABLE in a way a real reviewer could use?
• Did the tool expose too much, hide too much, or make the wrong thing look certain?
• For creative/math/graphics workflows, which measurements should be visible first: color/tonal metadata, dithering artifacts, splat/point-cloud state, shader parameters, audio features, or replay seed/state?

Please test only with public, synthetic, or non-sensitive material unless you are running locally and understand the boundary. Good next users are maintainers building agent tooling, eval/observability engineers, artists/studio people, graphics programmers, researchers, and anyone who needs AI-assisted work to leave a checkable trail instead of just a fluent summary.

Main site: https://harperz9.github.io
Field guide: https://harperz9.github.io/field-guide.html
Glama listing: https://glama.ai/mcp/servers/HarperZ9/telos