open research / ghost / work in progress
test the claim.
keep what survives.
Ghost is an open research program for discovering what engineering and intelligence claims actually survive explicit tests. The CLI and engine are experimental instruments for that work—not the story in place of it.
explore then proof
Hypothesis
proposed
Explore
in progress
Build + Test
gated
Proof
verified
work in progress
Research in progress
Ghost is source-first and experimental. Its research records the experiments, replications, controls, negative results, and blocked claims that shape the work. The CLI can be built locally, but it is not a packaged release, and research artifacts do not automatically become product capabilities.
Open Research
BitForge experiments test program synthesis, search limits, evaluation, and self-improving systems in public. Findings include controlled wins, replications, and many documented negatives—not product promises.
Reasoning Engine
The engine owns the low-level reasoning, support graphs, proof gates, verifier rules, and mutation logic. A useful response is not the same thing as a verified result: final validity stays with the relevant gates.
Operator CLI
A local Zig command-line shell for asking, inspecting, fixing, verifying, and working with project-scoped knowledge. It is intentionally a thin interface over the engine, not a second reasoning system.
research, honestly scoped
the research is Ghost's center.
The repository is not a marketing backlog. It is the record of the work: baseline studies, formal synthesis, meta-synthesis, adversarial checks, causal experiments, and the results that did not survive their controls. A result earns a place here by being scoped, replayable, and honest about its limits.
Verified synthesis
Formal program-synthesis experiments pair search with explicit specifications and verification. The archive includes reproducible mixer, sorting, and counterexample-guided synthesis work.
Limits and falsification
The research program actively records ceilings, failed controls, and blocked claims. The closure-principle work is a central example: changing search effort alone cannot escape a closed primitive family.
Current frontier
Recent rounds investigate autonomous knowledge, evaluator isolation, causal substrates, and organism-owned invention mechanisms. Those tracks are gated research, not shipped intelligence claims.
research instrument / cli
an experimental operator surface.
$ ghost askRoute a local question to the engine; an answer still needs its own evidence and verification.
$ ghost fixRequest a fix-oriented response, then inspect the proposed work and run the applicable gates.
$ ghost verifyCheck a workspace or verification workflow through the engine-backed interface.
$ ghost corpusStage, apply, and query project-scoped corpus material with explicit provenance boundaries.
$ ghost autopsyInspect a project’s structure and dependencies from the local operator workflow.
run it yourself
build the CLI and try it in your own terminal.
$ git clone https://github.com/Sylorlabs/ghost_cli\ncd ghost_cli && zig build && ./zig-out/bin/ghost --help
The CLI delegates to the engine; inspect both before depending on a workflow. View Engine Source ↗ · View CLI Source ↗
research that stays accountable.
Ghost keeps the research trail visible: hypotheses, gates, evidence, failures, and the distinction between a promising experiment and a demonstrated capability.
Explore the Catalog