A system proposes
A learned field, heuristic, or working belief can suggest where to look and what representation may need repair.
Start here
PRIME is the current programme. TS is its wider lineage.
The simplest way to understand the work is through one rule: Learned systems may propose. Declared verifiers decide what becomes accepted state. Everything else—selective information, representation repair, abstention, receipts, and replay—exists to make that boundary operational.
Evidence constrains it
Fresh observations, risk filters, and sequential tests decide whether the proposal has earned support.
A verifier authorises
Only declared authority may change canonical state, restore a representation, or record abstention.
A receipt preserves the boundary
The accepted transition keeps its inputs, provenance, hashes, limitations, and replay path.
Flagship release
PRIME v1.0.0
A finite, verifier-governed architecture for selective information acquisition, persistent representation repair, bounded cognition, and auditable abstention.
Open the Zenodo recordResearch foundation
Adaptive state abstraction
Anytime-valid repair, retain, or wait decisions under partial observability, with explicit abstention, provenance, and preserved failed gates.
Read the published artifactCanonical source
Thinking System
The public monorepo for verifier-first reasoning, typed authority, residual accounting, receipts, and structured revision.
Inspect the sourceResearch programme
Understand the current architecture, evidence, and claim boundary.
Published artifacts
Open the permanent Zenodo records and citation details.
Open source
See current repositories separated from preserved predecessors.
Project lineage
Trace earlier TS, model, graph, systems, and verifier work.
Looking for the exact current release? Start with PRIME v1.0.0 on Zenodo.