Project
Cogniform
A 3D engine that lets software and AI tools change a scene, then inspect images and object data from the same scene version.
Engineering reading paths
Agent-operable worlds
Connect bounded agent intent, authoritative state and revision-linked observations.
Start here
Key idea
An agent needs explicit authority to change a world and enough context to interpret the resulting observation. Explore Cogniform and LudoWeave through their documented contracts, inspect the recorded observations, and distinguish state identity from caller trust.
Curated connections
Systems
Project
LudoWeave Engine
A Python engine for small 2D worlds with verified replay, pause and single-step controls, seek/resume, and a tick-aware playback status panel.
Curated connections
Engineering notes
Engineering update
LudoWeave makes verified playback inspectable
M242 through M244 add pause/step, seek/resume and a current/final tick panel with playback state and contextual hints. Recorded world input stays authoritative.
Engineering update
Verified replay gains a presentation pass
M241 adds visible playback of verified Clockwork Arena recordings. A full headless check precedes presentation; recorded transactions and inputs remain authoritative, while live gameplay input is ignored.
Engineering update
Record play sessions, verify them headlessly
Clockwork Arena can record the input snapshots its play loop consumes and verify them in a fresh headless process. Recording is opt-in, capped at 3,600 requested ticks, and keeps rendering and audio outside the artifact.
Engineering update
Optional audio beside the canonical world
Clockwork Arena now documents optional synthesized effects through an explicitly pumped, callback-free audio path. The default Null audio mode remains headless and unpaced; device playback and human listening require separate verification.
Engineering update
Bounded local agent sessions
The CLI now documents inherited-stream sessions and a local MCP adapter for queries, imagination, atomic patches, and revision-linked observations. These adapters add no network listener or remote security boundary.
Curated connections
Long-form writing
Article
An Observation Needs a World Revision
What Cogniform's revision-linked observations clarify about agent feedback, local transport and evidence.
Article
A Replay Is More Than a List of Inputs
How consumed input snapshots, world transactions and independent verification make a recording useful evidence.
Article
Where AI Fits in the Larger System
What recent work on agent standards, on-device inference, and reliability has reinforced for me about building AI into real products.
Curated connections
Experiments and explorers
Bounded evidence
Explore four synchronized observations
Recorded capture. Inspect color, depth, normals and entity identity from the recorded reference frame, with the limits of each view.
Bounded evidence
Inspect a replay boundary
Interactive illustration. Step through an illustrative consumed-input recording and locate a divergent checkpoint. This teaching model does not execute LudoWeave.
Bounded evidence
Inspect the Clockwork Arena capture
Recorded capture. Compare the existing local frames and recorded hashes. The capture retains its original environment and limitations.
Curated connections
Evidence
System map and source
Cogniform: inspect the evidence
Explore the documented boundaries and pinned sources in this case study. Evidence reviewed 30 September 2026.
System map and source
LudoWeave Engine: inspect the evidence
Explore the documented boundaries and pinned sources in this case study. Evidence reviewed 30 September 2026.