Selected work / Product evidence

What is built. What is still being tested.

Completed proof cases stay separate from projects still earning their claims. Each active build shows what works now and the next evidence gate.

01 / Selected evidence

Three end-to-end cases ready to inspect.

These dossiers carry the clearest product, architecture, failure, and verification evidence.

AI-enabled coaching product

LeagueLoop

Validated locally

Turn one finished match into one better decision next game.

A post-match coaching product where deterministic code extracts and validates the evidence around one bounded model-generated review.

  • Structured validation
  • Explicit paid-call control
  • Local retention and deletion

Operational B2B product

Extra Work Capture

Portfolio demo

Preserve extra-work evidence before the context disappears.

A phone-first workflow that carries one request from field capture through requester acknowledgment, actuals, pricing, outcome, and a printable evidence packet.

  • Idempotent mutations
  • Append-only history
  • Visible failure recovery

Agent reliability lab

BoundaryLab

Portfolio demo

Only one architecture stayed useful under attack.

BoundaryLab holds eight synthetic support cases constant across three agent architectures, then shows how context, memory, review, and capability boundaries change safety and completion.

  • Bounded profile: 8/8
  • 0 attack successes
  • 0 false refusals

02 / Currently building

Four projects still earning the next claim.

Current proof and the next gate stay visible so early work does not read like a finished product.

Local-first idea admission

Roughwise

In progress

Decide what an idea deserves before creating another repository.

A local-first desktop product that turns rough ideas into evidence-backed dispositions, then keeps project creation behind an inspectable approval boundary.

Current proof
  • Packaged Electron skeleton
  • Five explicit dispositions
  • Local persistence and replay
Next gate
Connect the navigable builder profile to one deterministic idea-admission loop.
Read current build →

Deterministic autonomy simulation

Lunar Dispatch

In progress

Make rover dispatch decisions visible, reproducible, and reviewable.

An Unreal simulation for comparing dispatch policies under operational constraints, faults, and human oversight without confusing software evidence with physical-world validation.

Current proof
  • 7 passing Automation Specs
  • Repeated run, identical trace
  • Invalid commands preserve state
Next gate
Project the deterministic core into a visible graybox depot with rovers, controls, and an event timeline.
Read current build →

Deterministic generative music

Mapsonata

In progress

Turn a place into the same piece of music every time.

A browser composition engine that maps normalized coordinates to a versioned musical plan, while treating listening quality as a separate gate from deterministic correctness.

Current proof
  • Same place, same plan
  • 24-place listening gate
  • First candidate failed
Next gate
Run a six-piece directional listening pass before completing the corpus or adding the map.
Read current build →

Mobile conversation product

Topicall

In progress

Small scheduled conversations organized around topics, not profiles.

A mobile-first product with a deterministic scenario environment, pure domain packages, fake external services, and strict production boundaries.

Current proof
  • Expo and React Native
  • Deterministic scenario runtime
  • Production export gate
Next gate
Validate one complete scheduled-conversation path across both mobile platforms.
Read current build →