Witness the seam
Scoped perturbations are checked against physical runtime evidence before a result is allowed into the conversation.
Solo-built evidence systems for human-AI symbiosis
Colby Fox, solo systems founder
AI needs receipts,
not vibes.
Probably AGI is where Colby Fox publishes pieces of a larger system: durable memory, episodic time, inference courts, and model anatomy maps that let AI learn with humans without pretending every change deserves authority.
Building evidence systems, not beliefs.
This is not a lab. It is one builder's public trail.
- no-authority by default
- receipt-backed
- local inference
- episodic memory
- model-porting autopilot
We publish evidence, not reproduction paths.
This is not a claim of production steering, AGI, or hidden model memory. It is a receipt-backed system for testing what actually happened inside local inference runs.
FoxEIS
Receipt-backed inference courts.
FoxEIS is the first public artifact: a solo-built evidence protocol for testing what models actually did, witnessing whether scoped interventions physically applied, and rejecting fake wins before anything earns authority.
Reject fake wins
Cleanup drift, zero-bias contamination, and router-only motion stay out of the winner's circle.
Port the court
Receipts support model-porting, autopsy, and autopilot work without dumping coordinates or reproduction paths.
Pipeline
Evidence moves through gates.
- Model manifest
- Runtime doctor
- Seam passport
- Island atlas
- Prompt geometry
- Rule mine
- Cleanup screen
- Pre-registration
- Court receipt
FDAG
Memory should be earned, not stuffed into context.
FoxEIS tests inference. FDAG remembers evidence over time. Together they point toward AI systems that can learn with humans without pretending every memory or influence deserves trust.
What I will show
- redacted receipts
- no-authority verdicts
- model-porting conveyor
- why prompt geometry matters
- why failures become evidence
What I will not dump
- exact coordinates
- prompt packs
- kernel details
- reproduction paths
- brittle gains
- unverifiable claims
Notes
The public trail now has rooms.
Founder notes, explainers, boundaries, and sanitized evidence examples stay precise: useful evidence, clear limits, no mythology.
Edge Injection Maturity Gate #1
Infrastructure complete, science pending: a sanitized first evidence example with authority still off.
Read the gateReceipt-Backed Inference Courts for Local Models
A public-safe explanation of FoxEIS: scoped inference perturbation, physical witness, clean effect measurement, and no-authority governance.
Read noteProbably AGI is not a prediction. It is a workshop.
Colby Fox on Probably AGI, FoxEIS, FDAG, episodic time, and evidence systems for human-AI symbiosis.
Read notePrivate demo
For people building the future of AI systems.
AI infrastructure builders, local inference people, safety and evals teams, interpretability researchers, and serious strategic backers can request a private demo.
No slides. Just receipts.