Recursive Discovery v1.0.0

First public release of Recursive Discovery — a research system for scientific problems that need both mathematical reasoning and empirical evidence.

Included

• Artifact ledger — SQLite graph with content-addressed blob storage, so proposals, sources, results, and decisions share one representation and survive individual model invocations.
• Kernel-signed execution records — every declared job is executed, hashed, and signed, with the tool, command, runtime, platform, environment, and seed recorded as provenance.
• Model-owned research sessions — persistent branches, memos, and replay-ready decision/outcome traces.
• Context compiler — task-local retrieval using graph proximity, lexical hits, and authority labels.
• Mathematical checkers — Python, SymPy, Lean, Z3, and cvc5 adapters, discovered when installed.
• Empirical laboratory — declared-input snapshots, seeded runs, and statistical summaries with intervals, contrasts, and stopping rules.
• Reusable instruments — built-in diagnostics plus model-defined expression instruments.
• Prospective evaluation — a separate sealed evaluator with one-use test handles and service attestation.
• Literature search — arXiv, OpenAlex, and Crossref, with optional source reading.
• CLI and reports — recursive-discovery commands and Markdown reports.

Scope

This release is the general research machinery. It does not bundle a model provider, scientific demonstrations, or frozen benchmark results: model credentials stay in your own bridge, and results depend on the checkers, protocols, and data you supply.

Requirements

Python 3.11+, NumPy, SciPy, and SymPy. Lean, Z3, cvc5, and a container runtime are optional and discovered when available. The worker module imports POSIX resource, so native Windows support is not established.

Notes on what this is not

Signed execution records authenticate that a run happened; they are not a claim that a scientific interpretation is correct. The default process worker is not a sandbox. Persistent branches are records, not a distributed swarm. See SECURITY.md and docs/FORMALISM.md.

Getting started

python -m venv .venv
source .venv/bin/activate
python -m pip install -e .

recursive-discovery init ./research-workspace
recursive-discovery status ./research-workspace

See the README for the model bridge contract and a first research session.