Synaplex

synaplex.ai is the system — a self-learning, self-evolving feedback system for uncovering the structure of AI systems.

Not a publication. Not a portfolio. The system itself, with a public face at synaplex.ai and an operator face at command.synaplex.ai. Pods (atlas for systematic crypto, skillfoundry for commercial discovery) are bidirectional exploratory probes; the canon, knowledge system, lab, publication, and command surfaces are the load-bearing layers.

This repository is the codebase for the system — publication machinery, evaluation lab, and (in progress) the five-layer operational pipeline that ingests external signal, converts it into candidate canon envelopes, validates them, and publishes accepted claims as reader-facing writeups.

What gets published

• Editorial: topology-not-timeline deep-dives on agent platforms, memory systems, context infrastructure, orchestration, and the broader structure of AI systems. New pieces refine the latent shape; old pieces become foundations. Backed by the same canon envelopes the lab emits.
• Lab: systematic third-party evaluations, pre-registered as canon Claims, executed against bounded task suites, recorded as typed Evidence, decided via canon Decisions that cite all contradictory evidence. Replayable and hash-bound.
• Directory: quality-graded (not star-count-graded) catalog of agent platform components, curated from the lab's evaluation corpus.

Canon

Every envelope this repo emits conforms to the discovery-framework spec at context-repository/spec/discovery-framework/ (v0.1.0, frozen). Object types: Claim, Evidence, Decision, Promotion, Realization, Policy, EventLogEntry, ArtifactPointer.

See CLAUDE.md for operating principles; CURRENT_STATE.md for live status; supervisor/decisions/0027-synaplex-is-the-system.md for the framing.

Repository profile

Synaplex is an active, agentic, flat multi-surface monorepo. It intentionally does not imitate a single Python service: intake, reasoning, lab, canon, knowledge projection, integrity, and the Astro publication surface have different authority and runtime boundaries. See repo.toml and docs/architecture.md.

make help
make check

make check is the merge gate. It runs deterministic lint, scoped static typing, tests and canon guards, prompt-baseline contract checks, the site build, and deployment-source validation. It does not run models or change epistemic state.

Historical lineage

The public /lineage/ surface records four predecessor repositories at exact commits and snapshot digests. Those records preserve useful design history while explicitly excluding scientific Claims, Evidence, findings, and current-study status.