swaitsandClaude Opus 5.5
c9772b70c1
docs/mcp-interface.md: the connect text and its budgets, the handbook (tool and resources), annotations, failing closed, import_local_file's roots and refusals, the set_w9 path rule and where TIN-bearing files may go. docs/testing.md: the P0 laws and the tests that check them. docs/architecture.md: the journal format gate and the server's fail-closed store lock. Round-2 review (2026-10-16): mcp-interface.md lists all 26 destructive tools (unlink_receipt, unlink_attachment and correct_asset were missing) and openWorldHint false on every tool; the connect text states the parties rule and the exact receipt rule, and how it fits its budget; the lint covers every refusal; the import roots refuse /, the home folder and folders holding the books or a key, the open is a component walk, every missing or outside path gets one refusal, and the TIN-file registry and text scan (and what they can't catch) are described. architecture.md: the frozen envelope, the lenient next-line prev, no repair of newer books. testing.md: the new laws and tests. verification.md and parity.md: P0 notes (no new law to prove, no status moved). The TIN text check, narrowed (2026-10-16; rules.md B-26, mcp-interface.md): text is refused for a TIN the vault holds, in its common renderings, or a stand-alone printed SSN or EIN shape, so bank exports go in; the oracle a vault comparison gives is recorded as a known limit for P7. testing.md: the P0 mutation run of the final diff, per crate, and how each survivor was resolved (killed by a named test, removed by simpler code, or argued equivalent). The probe bound (rules.md B-26, mcp-interface.md): rule (a) answers for at most two candidates a filing, so finding a TIN from its last four takes on the order of 50,000 refused filings; the open limit now names that path, closed by P2's access log and P7's human-only TIN channel. testing.md: the law and its tests; tin_files.rs mutated again, 74 of 74 caught. Review round 3: rules.md B-28 (four-digit years, amounts a line can carry, checked products, bounded recurring runs) and B-26/B-27 updates (BOM-less UTF-16, when set_w9 records); mcp-interface.md matches the code (25 destructive tools, the five idempotent ones, no reverse_entry receipt checklist, strictly above the threshold, no tool text names wanderbooks.toml, dot names and roots inside the books refused, the new settings lists); testing.md lists the new laws and this round's mutation run. Review round 4: rules.md B-28 says the bounds are for new commands only, replay accepts every line an earlier version accepted, and derived figures refuse rather than panic; mcp-interface.md documents plan_error and the nullable totals; testing.md lists the old-books contracts and this round's mutation run (core, store, MCP, config, GUI). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
wanderbooks
Canonical double-entry bookkeeping, rebuilt from scratch. The engine is verified against oracles, parity is tracked automatically, and the primary interface is AI-first: you talk to an agent, the agent drives the books through MCP. Self-hosted, no rent.
What it is
ledger-core— the semantics engine. Exact decimal money, five account types, balanced journal entries, immutable posted log with reversing entries, trial balance, P&L, balance sheet, indirect-method cash flow, fixed-asset register with straight-line depreciation.ledger-store— durable storage. The source of truth is an append-only, hash-chained JSONL journal; SQLite is a disposable projection rebuilt from it; receipts live in a content-addressed blob store. Drafts are durable, corrupt projections self-heal.ledger-mcp— the AI-first interface. 64 MCP tools over stdio or Streamable HTTP: draft → preview → post workflow, server-side receipt policy (>$25), anti-splitting detection, asset management, digest for human review, plus read-only resources for dashboards.ledger-import— QuickBooks interop. Import bank/credit-card CSVs and chart-of-accounts CSVs; export IIF journal entries (for QuickBooks Desktop) and XLSX reports.ledger-config— the company settings in a books dir'swanderbooks.toml.ledger-gui— the viewer: a read-only, live window on the books for owners and their accountant (egui). Answers first (cash, runway, profit, what needs attention), then every number drills down to its lines, its entry, the rendered receipt, and the journal history. Follows the journal live: a change the agent makes appears in milliseconds.--demoopens a realistic demo company.
Verification
This is the storytold way applied to accounting: agents verify against oracles, never ship what they haven't demonstrated.
- 190+ workspace tests: invariant tests encoding accounting law, golden-file report snapshots, property tests (proptest), end-to-end MCP stdio tests.
- Formal proofs of the accounting laws (
just prove,docs/verification.md). Kani model-checks the engine's arithmetic kernel exhaustively within bounds. Lean 4 proves that, after any sequence of commands, the trial balance balances, the accounting equation holds, posted entries are forever, closed periods stay frozen, and a close moves exactly the net income to equity. The first pass found three engine bugs that every test, fuzzer, and mutant had missed: negative amounts accepted, and two crashes on oversized amounts. It also caught a soundness bug in the model checker itself. - Mutation testing via cargo-mutants; fuzzing via cargo-fuzz (11.5M+ executions, zero crashes).
- GnuCash is a behavioral cross-check, never the authority and never a code source (clean room, MIT OR Apache-2.0). Accounting principles, proved, are the spec. Where any other program disagrees, we work out from principles who is right.
Status
Parity is 12 of 12 against the spreadsheet workflow it replaces
(docs/capability-parity.md). The parallel-run plan
(docs/parallel-run.md) is written; the Google Sheet stays the system of
record until a clean month and explicit cutover approval.
See plan/STATUS.md for where things stand and
plan/roadmap.md for where it's going.
Quickstart
cargo test # whole suite
cargo run -p ledger-mcp -- --data-dir /tmp/books # MCP server over stdio
cargo run -p ledger-gui -- --demo # the viewer, on a demo company
New books, one command (installs Rust and wanderbooks if needed, makes the books dir with the MCP server registered for it, and starts the agent):
curl -fsSL https://wanderbooks.ai/install.sh | sh -s -- ~/Books/Acme
Already installed: wanderbooks new ~/Books/Acme (add --codex to keep the
books with Codex instead of Claude Code). In a books dir, just
lists the shortcuts (just gui, just agent, just backup).
Developing
Every tool the recipes use (Rust, just, Kani, Lean, cargo-mutants,
cargo-fuzz, Node for deploys) is pinned in mise.toml:
mise install
just # lists the recipes: check, build, prove, demo, ...
Layout
crates/ledger-core/ # engine: money, accounts, journal, ledger, reports
crates/ledger-store/ # journal + SQLite projection + blob store
crates/ledger-mcp/ # MCP server: 64 tools, resources, receipt policy
crates/ledger-import/ # QBO CSV import, IIF/XLSX export
crates/ledger-config/ # wanderbooks.toml company settings
crates/ledger-cli/ # `wanderbooks new`: one-command books dir + agent
crates/ledger-gui/ # the viewer: read-only, live (egui)
docs/ # requirements, architecture, oracle, parity, glossary
plan/ # STATUS.md, roadmap.md
AGENTS.md # operating manual for agents working here
License
MIT OR Apache-2.0. See LICENSE-MIT and LICENSE-APACHE.