Built as a field guide, not a funnel. Every track has a working surface.
Some pieces here are engines. Some are demos, papers, generated images, graphics systems, or notes from ongoing research. The point is not to flatten every artifact into one product. It is to make the range navigable and show how independent engines can pass verified outputs forward.
That connection is the Flywheel thesis: parallel systems run alone, then improve the next run when the problem calls for it. A compiler can inform an agent harness. A graphics pipeline can become a measurement surface. A local-model workflow can become a benchmark. A generated specimen can become a design system.
The front door is range: systems, models, graphics, research, generated media, and the tooling that makes them usable.
The work still carries dates, source links, tests, papers, and limits where they matter. That discipline stays. What changes is the public frame: rigor is support structure, not the whole thesis. Start with what you want to understand or use, then follow the links into the deeper material.
Where to enter.
Different readers need different doors. This page keeps the routes close together so the work can be scanned as a system instead of as isolated project tiles.
Engine room.
public tools that map workspaces, ingest difficult sources, run agent teams, judge claims, and compile typed-effects systems code.
Use the flagship list when you want names, roles, source links, package pages, and a fast read on what is already public.
Open an engine, run a demo, then follow the source link when the behavior matters more than the summary.
Model workflows.
local endpoints, harnesses, benchmark loops, context maps, and agentic work patterns that can be measured instead of narrated.
Start with flywheel for routing and verification, relay for coding across endpoints, plexus for tool wiring, and mneme for memory. Use telos, forum, gather, index, and crucible for the shared workflow loop.
Bring a real workflow, repo, or document set. The useful question is what should become a repeatable loop.
Graphics and generated media.
procedural fields, render pipelines, color systems, engine instrumentation, and generated specimens that become site material.
Use the Studio, generative demos, Quanta Color, and graphics pages when you want to see the visual side of the workshop.
Look for outputs that are generated by the system itself: canvases, textures, captures, and tool-made artifacts.
Research and writing.
papers, essays, field notes, release notes, and the public record of what is stable, forming, or still speculative.
Use research when you want formal artifacts and writing when you want the plain-language version of the thinking.
Read one piece, then jump sideways into the engine or demo that made the idea concrete.
Consulting and collaboration.
freelance work, test runs, technical strategy, prototype development, and research collaboration across unusual domains.
Use work routes when you need a concrete engagement: audit a workflow, build a prototype, shape a benchmark, improve a site, or pressure-test a tool.
Send the artifact or problem. The first pass should turn ambiguity into a scoped run, not a vague sales call.
Private-line work.
larger internal systems, lab-only infrastructure, and public-safe slices that are described by capability without exposing sensitive internals.
Use Checkpoint, Runtime, Vault, Boundary, and the private-line pages for the public portions of the larger platform.
Stay with the public-safe slices. Anything private is described at the boundary, not opened by accident.
Different doors, one workshop: systems that can be tried, inspected, adapted, or discussed without pretending the whole body of work is one product.
Flagships.
Everything below is part of the public map: engines, private-line slices, graphics systems, papers, essays, and product experiments. The workspace behind them is mapped by index itself, while public and private surfaces stay explicitly separated. The private-line flagships publish only the pieces that can be safely shown, while Lab (Seed) and Ledger (Sofer) remain private until a clean public split is ready. The map is meant to help you choose a door, not memorize the whole building.
The fourteen engines
Private-line flagships
Secondary substrates
Bricks
Also on the site: the catalog, the resume, and the CV. WARDEN is one private-line branch of the larger Project Telos map.
Two small live demos. Specific tools, not the whole thesis.
These examples show one track inside the larger workshop: byte integrity and action boundaries. They stay here because they are concrete, runnable, and useful, but they no longer carry the whole site’s public framing.
# re-derive a file's bytes from scratch and compare to the seal >>> witness(path, seal) MATCH sha256 re-derived, identical to the sealed digest # the same file, one byte changed >>> witness(tampered, seal) DRIFT re-derived digest differs from the seal # asked to confirm a property with no evidence to re-derive it from >>> witness(path, seal=None) UNVERIFIABLE no seal to check against, not a pass, not a fail
This demo fingerprints text locally and compares it to a stored seal. The point is practical: tiny changes become visible, and missing comparison material is labeled as missing.
EMET ships a tamper-seal that rebuilds a file’s bytes from scratch across four independent language implementations. Honest limit: a plain seal can show whether bytes changed relative to that seal, but it cannot stop a determined forger who controls both the file and the seal. For that, the seal needs an anchor outside the file.
# a gate request whose authorization receipt is absent >>> evaluate_gate(request) GateDecision(decision="deny") authorization: no receipt, deny budget: unknown, needs-human # a delegate trying to widen its own granted scope >>> verify_delegation(chain) DENIED hop scope is not a subset of its parent, privilege escalation $ python -m pytest # run the repository suite for the current result
The library models action boundaries: missing permission denies, empty permission grants nothing, and a delegated step cannot quietly widen its own scope.
The check ships in proof-surface: standard-library only, zero dependencies, with its current suite and source at github.com/HarperZ9/proof-surface. Honest limit: the check gives advice, a careful recommendation the real system still has to carry out. The library itself doesn’t hold the power to force it.
The shared backbone (system/spine.js) lets these pieces show up in the studio, where generated drawings can carry fingerprints and boundary checks without turning the whole site into one security product.
Run them yourself: the demos run locally in the browser where possible. Open the recorded demonstrations → · open the studio route →
How it checks itself
The same checking runs underneath everything here. Try it, then forget about it.
Byte fingerprint demo. Type, seal, change, compare.
Try it. Type something, then seal it. The demo takes a fingerprint of those exact words and remembers it. A fingerprint here is a short code that any change, even one character, would scramble into something completely different. Change one letter and compare again: the fresh fingerprint no longer matches the one it kept. Clear the seal and there is nothing to compare against.
This demo needs JavaScript and the Web Crypto API. The full implementation ships in EMET, linked below.
The fingerprinting is the real thing: crypto.subtle.digest("SHA-256"), a tool built into your browser, run on what you typed and never sent anywhere. EMET does the same across whole files in four independent language implementations; the source and frozen specification live at github.com/HarperZ9. Honest limit: a plain seal like this only compares bytes against the seal you kept.
The check. Allow is earned, not assumed.
An action arrives at the check, hoping to go ahead. The check says no by default. It is a checkpoint that says no until it is sure. When anything is unclear it still says no: no permission slip means deny; anything it can’t be sure about becomes needs-human (stop and ask a person); and allow comes back only when every single thing checks out. Flip the conditions below and watch the answer move. Try to earn an allow, and notice just how much has to be true at the same time.
This is the real decision logic: a single no settles it; anything unknown becomes needs-human; and an allow needs every check to pass at once. The tested original lives in proof-surface: plain Python, nothing extra to install, with current source and suite at github.com/HarperZ9/proof-surface. Honest limit: the check only gives a recommendation; the program around it does the enforcing. This page can never grant a yes on its own.
The standing invitation.
I’m putting the public portions of the workshop in one place so the range can be understood without a private tour. Some projects are mature, some are forming, and some are deliberately held back until they can be separated from sensitive internal work. The public map should make those boundaries legible.
If you are evaluating the work, start anywhere concrete: run a demo, open a package, read a paper, inspect the generative surface, or send a real problem. The best conversations begin with an artifact on the table.
Bring a workflow, a research thread, a prototype, a site, or a hard technical knot. The point is to make it tangible.