Skip to content

mxr classify

Generated from mxr classify --help. After editing crates/daemon/src/cli/, update and review the snapshots with scripts/cargo-test -p mxr --test cli_help, then rebuild site/.

Measure how well mxr sorts mail: today’s classifier against labelled scenarios

mxr classify [OPTIONS] <COMMAND>
Subcommand Purpose
report Count stored fallback decisions, grouped by home, sender kind and rule
eval Grade the mail already in this mailbox against labelled scenarios: how often today’s rules agree on who sent it, which mode it goes to and whether it makes a record; for what an email asks of you, precision and recall per kind of ask and how often the amount, due date and counterparty are exactly right; and every email they get wrong. Read-only and local. Run mxr demo first to get mail with labels
help Print this message or the help of the given subcommand(s)
Flag Description
--account <ACCOUNT> Limit to one account; the default covers every account
--format <FORMAT> [possible values: table, json, jsonl, csv, ids]
-h, --help Print help

Grade stored sender, home and record answers against labelled scenarios, and inspect each mismatch. Measure asks with per-kind precision and recall and field exact-match. Read-only. Only the demo mailbox carries labels. See Measure how well mxr sorts.

Terminal window
mxr demo --no-tui
mxr classify report --limit 1000 --format json
mxr classify eval
mxr classify eval --set full
mxr classify eval --compare POLICY --format json
mxr classify eval --set full --format json # every mismatch
mxr demo stop

Grade the mail already in this mailbox against labelled scenarios: how often today’s rules agree on who sent it, which mode it goes to and whether it makes a record; for what an email asks of you, precision and recall per kind of ask and how often the amount, due date and counterparty are exactly right; and every email they get wrong. Read-only and local. Run mxr demo first to get mail with labels

mxr classify eval [OPTIONS]
Flag Description
--set <SET> Which labelled set to grade against Possible values: - quick: The labelled demo mail and every near-miss twin - full: Every labelled email in the mailbox: on the 50,000-message demo, the labelled demo mail and all of the generated mail, with held-out phrasings reported on their own line. Takes up to a minute [default: quick]
--compare <COMPARE> Compare homes with a policy version actually retained in this mailbox
--account <ACCOUNT> Limit to one account; the default covers every account
--format <FORMAT> [possible values: table, json, jsonl, csv, ids]
-h, --help Print help (see a summary with ‘-h’)

Count stored fallback decisions, grouped by home, sender kind and rule

mxr classify report [OPTIONS]
Flag Description
--limit <LIMIT> [default: 1000]
--account <ACCOUNT> Limit to one account; the default covers every account
--format <FORMAT> [possible values: table, json, jsonl, csv, ids]
-h, --help Print help