books.start passed Integrity::default() to the attention engine, so it
asserted "database matches, no file failures". On a book whose database
differs from the journal, or whose stored file is missing or corrupt, the
viewer showed the database/stored-files items and the server did not: the
two lists diverged exactly where they matter most.
Call find, which computes the check: every distinct stored blob is read
and hash-checked once (a damaged blob is still reported per attachment
filed from it), and Store::disk_check now makes the viewer's comparison
through a snapshot of its data directory instead of reporting a match.
The parity test now damages its own fixture (removes the receipt's blob,
then changes a projected account row) and asserts both lists still agree
item for item and both name the damage. On the demo books (138
attachments, 138 distinct blobs, debug build) books.start takes about
0.26 s warm, 0.29 s cold, so the real check costs no decision.
Needs attention was computed by the viewer alone, while books.start
hand-picked a handful of next steps. The human's list and the agent's
could disagree, and a new item had to be written twice (or, as
books.start showed, not at all).
Move the engine to ledger-package over PackageBooks: one function,
find/find_with(books, today, settings, integrity), reads the ledger, the
journal's event types (an entry's origin), the attachments and their
links, the statements, the reconciliations, and the database check. The
viewer's attention module now only maps Target to Route; books.start
turns the same items into its next steps (why = the item's title, top 12),
so the list the human sees and the steps the agent gets cannot drift.
PackageBooks gains settings, journal, statements, reconciliations and
disk_check, implemented for Store and Snapshot, so both views answer
alike. The W-9, reconciliation and 1099 predicates move beside the items
to keep one copy. ledger-package gains ledger-config and serde_json (the
receipt policy and the tool arguments the items carry).
Tests: view_model.rs and spec_kept.rs pass unchanged; the new
attention_parity.rs compares the viewer and books.start item for item on
the demo books and on a book written entry by entry. just check green.
server.rs had grown to 12,566 lines: every tool handler, every helper and
the request machinery in one file, so every unit that adds a tool touched
it and parallel worktrees conflicted on every merge (plan/served.md 8.2
U0a, 8.0.3). The 98 handlers still in it now live one file per group under
tools/, following the pattern tools/kept.rs and tools/projects.rs already
use (`#[path]` declaration, `use super::*`, one
`#[tool_router(vis = "pub(super)")] impl` per file):
documents (invoices, bills, payments, voids, books.documents,
books.document, the control check, the agings), entries (draft, line,
post, reverse, get, search, the bank CSV import), reports (the
statements, books.show, verify, digest, export, the CPA package, the
P&L by class, with the export builders), assets, schedules, periods,
reconciliation, parties (parties, vendors, customers, W-9s, the TIN
vault, 1099s, entry parties), attachments (receipts, filed documents,
links), settings (the handbook, books.settings, FX rates), chart,
payroll, payouts, tax; add_class and list_classes join the existing
projects file, since a project is a class with its facts.
The helpers only one group uses moved into that group's file; the ones
several groups share moved to tools/helpers.rs, re-exported by server.rs
so every group reaches them through `use super::*` and no group file
reaches into another's (resolve_instrument and the attachment/statement
shapers moved there too, since the payment, entries and attachment tools
all use them). server.rs keeps BooksServer, the store and its lock, the
request machinery, call_tool/get_info, tool_router (which now composes
the group routers; core_router is gone), and the shared helpers the plan
names (as_of, hand_over, resolve_vendor, parse_doc_ref, preview_lines_for).
This is a move, not a change: no accounting behaviour, tool name,
description, annotation or schema changed. A dump of
tool_router().list_all() — every tool's name, title, description, input
and output schema and annotations, in order — is identical to main's
(122 tools; the plan's "118" is stale). server.rs is 1,909 lines and has
no #[tool] handler. The server's cross-group tests moved to
server_tests.rs beside it (a #[path] module; fixtures in
server::test_support) and each group's own tests into its group file, so
server.rs stays under the 3,000-line bar. cargo test -p ledger-mcp: 99
passed; cargo test -p ledger-cli: 25 suites, 0 failed.
The owner (2026-10-23, via the question tool): "Uploads only"; and
before it: "What we would DEFINITELY want to avoid is passing a bunch
of BLOBs through AI context windows. That's just a pure waste." No file
passes through an agent's context, in either direction.
In: the inline arguments are gone with the 256 KiB limit (INLINE_MAX,
take_file's inline branch): content_base64 and its sha256 on
books.attach_receipt and books.file_document, w9_content_base64 and
w9_sha256 on books.set_w9, csv_text on the two CSV imports. A file comes
only by books.begin_upload and `wanderbooks upload`, named by its
upload_id; existing_attachment_id still names bytes stored already. The
five filing tools deny unknown fields, and any inline argument to any
tool is refused before it runs as an unknown argument pointing to
books.begin_upload (access log: refused inline_file).
Out: books.export (every format), books.asset_labels and
books.get_receipt view bytes hand the file over (HandedFile: download
command and expiry, D-29), as the CPA package and the W-9 do. Resource
reads stay text. A token keeps up to 64 handed files (was 16). The
access log's digest no longer exempts file bytes.
Docs: handbook files-in-out and parties-and-tins, mcp-interface.md,
rules.md, receipt-store.md, parties-and-attachments.md, plan/served.md
(D-46; the inline decisions marked superseded), STATUS.md, testing.md
(D-46: tests migrated to uploads and downloads, rewritten, removed,
added).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
The owner (2026-10-23): "We should probably just delete the stdio path
altogether. THere's no point in that. And it sitting there unused will
just lead to bitrot. Nuke it?"
Books are served only by the books server (`wanderbooks server`),
reached by agents through the bridge (`wanderbooks mcp --book ID`),
which stays.
Removed: `wanderbooks dev-stdio`, the `ledger-mcp` binary, and
`ledger_mcp::run` (stdio and `--http` over a bare data dir), with all
that only that mode had:
- files the books make coming back inline: `HandedFile.content_base64`
is gone, and `download` and `expires_at` are always there;
- books with no access log or backups: `books.access_log`,
`books.backup` and `books.back_up_now` are always offered, by scope;
- the settings file read and reloaded (`legacy_files`), the vault key
named by a file or `--vault-key-file`, and the books dir's
key-inside-the-books and git checks (with `paths.rs`);
- `wire.rs` (the bridge frames an agent's lines itself);
- `books.settings`' file fields and `books.start`'s `settings_source`.
`BooksServer` is built one way, from `Served` (a book as the books
server serves it), in production and in the unit tests. An unknown tool
is now a protocol error (mcp-tools.md). Inline file input is unchanged.
Then the owner's three decisions (2026-10-23, the question tool):
(a) `books.import_settings` is removed entirely (118 tools): old books
come in whole by `wanderbooks books adopt`, which reads their settings
file once; ledger-config keeps its settings-file parsing only as far as
adoption uses it (`load` and the path-reading `effective_receipt_policy`
are gone), and `books.chart_template` gives its settings as
`books.set_setting` changes. (b) The viewer's loose-books-folder mode is
removed (`--data-dir`, `--books-toml`, `--vault-key-file`, the settings
file and config.json followed): it opens only books of the data root,
reading settings from the journal and the vault from the book's own
`vault/` and `keys/`. (c) `wanderbooks dev-stdio` has no stub: it is an
unknown command like any other word. Also: an upload's expiry past the
calendar is an error, never an empty string.
Tests: every suite that drove the stdio server moved to
crates/ledger-cli/tests/ and drives a real `wanderbooks server` on a
data root of its own (tests/common), handed files by their download;
mcp_stdio.rs is mcp_journeys.rs. Tests of what is gone are deleted and
listed in docs/testing.md. Docs: plan D-44 (D-18 superseded), STATUS,
testing, rules, mcp-interface, the handbook, README; justfile recipes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
A client project's invoiced revenue and a contractor's billed cost could
never count toward the project: books.create_invoice and books.create_bill
posted at once with untagged lines, and only a draft's line takes a class.
Job costing (B-37 item 3) collects a job's billings and the supplier
invoices for its direct costs on its cost sheet, so each document line
now takes an optional project.
The owner decided it on 2026-10-22 (plan D-43, rules.md B-37 item 8):
each line takes one project or none; a document spanning N projects is N
lines, each with its own amount; no percentage split, so every figure is
exact; a cost truly shared among projects names none on the document
and reaches projects only as entries of their own under the company's
monthly allocation convention, when books.conventions has one (otherwise
it stays untagged and the agent asks the human). The reasoning, stated
in B-37: job costing charges a job the costs traceable to it, whole, and
reaches shared costs only by allocation on a stated basis. ASC
340-40-25-7 counts as directly related both the costs traceable to a
contract (25-7(a), (b), (d), (e)) and allocations of costs that relate
directly to it (25-7(c)); 25-8 expenses general and administrative cost.
The books carry the first whole on the document's line and the second as
the allocation convention's own entries; neither is a percentage on the
document (sources/fasb-asc.md, 25-7 and 25-8 added). The handbook's
projects rule and the line's class parameter say the same, and no
handbook sentence lets a shared cost be split across a document's lines.
The accounting: the document's entry posts one revenue (an invoice's) or
expense (a bill's) line per project its lines name, in the order each is
first named, each the sum of that project's lines, and the untagged
lines' sum as one more line (arap::document_parts, conservation checked
by a proptest). A document naming no project therefore posts exactly the
one line every document posted before. The receivable or payable line is
never tagged: it is a balance-sheet line, which item 3 counts nothing;
the payments (receive_payment, pay_bill) move cash against it and take no
project either. Nothing but the projects moves: the P&L, the 1099
worksheet (payments made, B-19) and party totals (what moved the money,
B-35) read the same with or without the tags (twin-books test); the
projects with the untagged lines still partition the P&L.
- ledger-core: DocumentLine.class_id (serde-omitted when none, so old
documents serialize as before); create_invoice / create_bill split the
revenue or expense side by project and tag it; an unknown project is
refused whole (atomically).
- ledger-store: a document line's project is journaled as class_id on
invoice.created / bill.created, and the entry's line as
entry.line_class_set right after its entry.line_added (the entry.posted
lines carry class_id, as a hand-built entry's do). Behind a NEW journal
feature, document_line_projects (KNOWN_FEATURES, declared by new books,
asked for before the first journaled line, not even the control
accounts written on refusal); not `projects`, which the live books
already declared: the owner consents again at the next upgrade. A
malformed class_id is corrupt; null or absent is none. Untagged
documents journal byte for byte as before; old journals replay
unchanged.
- ledger-mcp: each line takes class (id or name, resolved like
add_line's, every one before anything is written); the result gains
notes for a finished or abandoned project, and on an invoice for an
on-spec project (is there now an agreement? on spec a hoped-for
payment is no receivable, B-37 item 4), tagged anyway; books lacking
the feature refuse with format_upgrade_pending. books.get_entry shows a
line's class_id and project. The handbook's "an invoice or bill takes no
project yet" caveat now depends on the books: gone where the feature is
declared, an "until their format upgrade" note where it isn't.
- ledger-gui: an invoice's or bill's page (its entry's) lists the
document's lines, each project a link that opens it; the demo's client
invoices are tagged to "Client work", whose snapshot is now its revenue
less its contractors. Looking at the page in a narrow window found a
tagged line's tag pushing the amounts past the pane (any tagged entry):
the name is now cut to leave the tag room, the tag itself at most half
the cell.
- bookspec: an `invoice` event, and `lines` with `project` on a bill or
an invoice; example projects-invoice-and-bill-lines-count-by-project,
worked by hand, both projects client work (Zeta 3,500.00 - 1,200.00 =
2,300.00; Harbor 200.00;
untagged -1,750.00; together the P&L's 750.00; Dana's 1099 payments
1,250.00).
- fuzz: fuzz_store_replay can declare journal features.
- Tests: core, store, MCP, viewer, the P2-served cutover fixture and a
viewer/server parity test (its month lists the same length, not
empty); a tagged document's reversal nets its projects to zero, live
and replayed; books.get_entry's class_id; the handbook served as a
resource both ways; each fix broken to confirm its test fails
(docs/testing.md). Docs: rules.md B-27 and B-37 items 3 and 8,
sources/fasb-asc.md, plan/served.md D-43,
mcp-interface.md, testing.md, parity.md, plan/STATUS.md.
No proof changed (no kernel arithmetic: document_parts sums with the
existing document_total).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
The project page opens on a snapshot: revenue, direct costs, the
contribution before overhead and its margin of revenue (a segment
margin: revenue less every cost traced to the project, fixed and
variable; B-37 item 7), in the black or red, and for on-spec work the
cost unrecovered if repaid. To date or this fiscal year. By month,
revenue and costs with the net to date in its own strip; costs by
category; the list colors each net by its sign. books.project carries
the same figures, and a parity test runs the viewer and the MCP on the
same books to the cent. Tests honor TMPDIR.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
A project is a class with facts (description, status, kind, party,
dates, links, notes) in one project.set event behind the format
feature projects. Four tools (books.add_project, update_project,
projects, project); documents and decisions link to projects; B-37:
on-spec costs are expensed as incurred, repayment is revenue, and the
unrecovered cost is never below zero (Kani). Viewer: a Projects list
and a project page with its figures, by month and by account. The
sidebar measures its footer instead of reserving a fixed 172 px.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
Books an update could write a newer format for open on their old format
and are asked at every open; anything needing a new feature is refused
until the owner agrees with the passphrase, after a verified backup that
includes the keys. New books start on the current format. Each projection
schema has its own database file, so an older version still opens the
books until the owner agrees.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
A vendor's 1099 decision is its own dated fact (files NEC or MISC, not
reportable with its reason, or undetermined), applied by each payment's
date. Payment methods are recorded; card and network payments are left
out and listed; a member's check through financing counts; a bank payment
with no method counts (D-40); $2,000 for payments after 2025. One
worksheet for the tool, the viewer and the CPA package.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
Mutation testing pauses until P12 (D-33, owner 2026-10-20), which reruns
the whole max-testing pass over everything since P1. Before the pause,
pass 1 on slices 3-6 and its triage gave tests in ledger-store,
ledger-package and ledger-mcp, and three real bugs, fixed:
- the command line took an admin answer longer than four MiB, or cut off
before its newline, whenever the bytes that came still parsed;
- The CPA package put a twin's number on the last dot of the whole path,
so in a folder with a dot the second file landed in another folder;
- a document or category named of dots alone ("..") could climb out of
its folder when the CPA package is unpacked.
The plan gains P13, bookspec (D-34): a formal, sourced accounting
specification first (rules with symbolic forms and worked examples), then
its own formal validation and a conformance harness any engine can be
tested against, seeded by P3's regression tests. docs/testing.md records
slices 3-8 and what P12 must cover; plan/STATUS.md says where P2 stands.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
No Landlock, no seccomp, no helper: isolation is "same_user
(advisory)". The lease socket keeps two intents (close, format-upgrade
consent). The viewer holds the books open while it runs and gains a
Books menu: New books, Back up now, Restore (preview, then confirm),
Close, Connect Claude Code / Codex. New commands `wanderbooks books
back-up` and `wanderbooks connect claude|codex`. The scaffold
(`new`/`init` and its templates) is gone; install.sh needs no sudo.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
The server backs each book up itself: a per-book git archive, pushed
fast-forward only to a git remote and to a folder, when the books open,
after a quiet spell, before they close and on request; every commit,
push and drill is a record in ops/backup.jsonl. Drills clone each
target and verify the chain, the blobs, the vault and the head, weekly
and after a failure. `wanderbooks books restore` previews, then
restores beside or (with --replace) over, never moving books something
holds. The archive refuses anything but a plain bare repository and
pins git against the owner's configuration. books.backup and
books.back_up_now for the agent.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
`wanderbooks gui [BOOK]` finds the data root, reads settings from the
journal, the vault of the book, presence through ops/, Save a copy /
Open on every document, Save the W-9, the Export menu, and colour
schemes. One builder for the CPA package and the 1099 worksheet
(ledger-package), shared by the server and the viewer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
Files reach the books through the server: uploads by hash and size,
then filed; exports (the CPA package, the 1099 worksheet, W-9s) handed
over as downloads from the server, never read off disk by the agent.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
Agent tokens (wb_ and 64 hex digits), one book and one name each, kept
only as SHA-256 in a chain log and in the OS keyring; made, listed and
revoked only through the admin socket (`wanderbooks token ...`). The
door checks them, scopes cover every tool, every call is in the access
log (fail closed), and the owner passphrase is an Argon2id hash.
Setup's terminal test waits for its prompts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
One wanderbooks executable. The books are open while the viewer, an
agent session or a terminal holds them: a host process takes leases on
a socket, starts the server (the hidden mode `wanderbooks server`) with
a pipe on its input, and closes it 10 s after the last lease ends. No
background service. One host even in a race (run/host.lock).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
Settings, conventions (posting templates, approval rules, policies,
close steps, calendar, mappings, notes), decisions, open items and
entry approvals are journal events (journal format 2). Books that keep
their settings never read wanderbooks.toml or config.json; the server
and the viewer take the vault key from --vault-key-file instead.
books.import_settings moves a settings file in once, exactly as the
server reads it. Approval rules, when enforced, gate every posting
path. Replay refuses what it can't read as newer, never as corrupt.
Lean: a record changes no figure. Mutation on the whole diff: every
survivor killed by a test but two equivalent (docs/testing.md). The
Company page says a late open item is overdue in words, not colour
alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T3gXMVt7XJPi6ens1or9Dj
swaits
