MathResearch Ledger v1.2.2

Added

• Formal Problem lifecycle through research_create_problem and research_update_problem, with preview and explicit semantic revisions.
• Workspace-aware Codex bundled MCP and a per-Agent DeepSeek Harness Host plugin.
• An Adapter Contract and measured Harness capability matrix.

Changed

• Workspace binding comes from trusted Host/session context at server startup, rather than an install-time fixed research path. It remains immutable for that server's lifetime.
• The Codex plugin owns the shared Skill and MCP together; project-level direct MCP is a legacy fallback.
• The DSH Host plugin is the recommended default. Config-only integration remains a manual/single-workspace fallback. Do not enable both at once.

Fixed

• Exact MCP workspace binding no longer aliases .math-research-ledger to its parent workspace. Missing state returns NO_WORKSPACE without auto-init or parent fallback.

Compatibility and upgrade

• Validated on Windows with Codex 0.160.0 and DeepSeek Harness 0.2.1-rc.3.
• DSH rc.2 was not revalidated for the Host adapter. macOS/Linux Host behavior, UNC/network paths, GUI conversational acceptance and real-corpus mutations remain unverified.
• Install the matching wheel with [mcp] into your persistent runtime, refresh local plugin packages and start new sessions/restart. Updating source alone does not update an installed runtime.
• Core authority, verification and human-acceptance boundaries are unchanged.

Codex installation · DSH installation · Release details · Capability matrix

MathResearch Ledger 1.2.1

MathResearch Ledger provides durable structured memory, dependency tracking and verification infrastructure for mathematical research across AI harnesses.

Highlights

• Pure MCP reads no longer require a writable Ledger state directory.
• MCP sessions pin one workspace at startup and reject cross-workspace tool requests.
• DeepSeek Harness uses a configuration-only Bundle with its native MCP client and an external Python runtime.
• Windows Codex and DSH Desktop 0.2.1-rc.2 Host read-only acceptance passed; DSH synthetic writes and native OFF/ON passed.

Compatibility

• All concurrent Ledger writers should use 1.2.1+.
• Pin MCP deployments with startup --workspace or MRL_WORKSPACE; tool workspace is an assertion, not a selector.
• Python 3.12 is the recommended persistent runtime.

Install

Install the wheel with MCP dependencies: python -m pip install './math_research_ledger-1.2.1-py3-none-any.whl[mcp]'.
See the Codex guide or DSH guide.
The DSH .tgz is an unrendered configuration template. Use the preparation script in the source archive to install the runtime and render local paths before adding the Bundle.

Known limits

• DSH macOS/Linux Host acceptance is untested.
• Formal writes to a real research corpus were not separately exercised.
• Each MCP session has one pinned workspace; read-consistency fingerprinting adds latency.

Verify downloads with the attached SHA256SUMS. Final release hashes differ from the accepted candidate wheel; all 95 production Python files match that candidate and the release source. This release is published on GitHub only, not PyPI.

MathResearch Ledger 1.2.0

MathResearch Ledger preserves durable research state across external agent sessions.
It is not another research-agent runtime.

• Persistent structured Problems, Claims, Gaps, proofs, failures and checkpoint history.
• Qualified identities, pinned dependency versions and explicit freshness/premise states.
• Research Episode preview, original-source integrity checks and provisional ingestion.
• Continuation-aware retrieval, exact checkpoint/product navigation and bounded object reads.
• Free research by default; memory intervention by exception.
• Codex integration through the stable nine-tool MCP surface.

Human acceptance is a workflow boundary on the MCP surface, not authentication or
sandboxing against arbitrary local code/filesystem access. Checked is not accepted;
canonical is not verified.

DeepSeek Harness integration is experimental. Historical isolated Host checks passed,
but real Windows existing-workspace DSH acceptance remains incomplete at the reported
Low/Medium Integrity boundary. macOS/Linux DSH Host acceptance is untested. The plugin
does not change ACLs, integrity labels, authentication or Full Access settings.

Requires Python 3.11+. Download the wheel and install with
python -m pip install './math_research_ledger-1.2.0-py3-none-any.whl[mcp]', or use the
source distribution. See the README for explicit workspace initialization and Codex
installation. The experimental DSH tarball requires its separate documented setup.

Assets include the Python wheel, source distribution, experimental DSH Bundle and
SHA-256 checksums. This is a GitHub Release only; no PyPI publication is included.

See CHANGELOG
and known limitations.