> ## Documentation Index > Fetch the complete documentation index at: https://xerg.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Explain local evidence > Drill into a finding, signal, or repeated tool chain without exposing raw tool arguments or results. Every current terminal and Markdown finding, signal, and repeated tool-chain row includes an eight-character local item ID and a command hint: ```bash theme={null} npx @xerg/cli@latest explain ``` `xerg explain` reads the newest local audit snapshot by default and shows: * classification or explicit neutral-observation status * why the detector emitted the item * observed evidence and evidence still missing * economic meaning and pricing coverage * run/call IDs, local source locators, and the observed time range * digest-only tool input, result, and chain-invocation evidence when available * limitations and the next useful inspection step Human output is organized into titled `Why`, `Evidence`, `Economic meaning`, `Pricing and analysis coverage`, `Source/time range`, `Local source locators`, `Limitations`, and `Next step` sections instead of embedding a pretty-printed JSON object. Use `--json` for the byte-stable machine-readable packet. Raw tool arguments and results are never retained in an explanation packet. Xerg canonicalizes JSON by sorting object keys while preserving values and array order, then stores only SHA-256 digests and byte counts. ## Choose an audit or output format ```bash theme={null} npx @xerg/cli@latest explain 01abc234 --audit 98def765:2026-08-13T09:04:23.000Z npx @xerg/cli@latest explain 01abc234 --db /path/to/xerg-snapshots.json npx @xerg/cli@latest explain 01abc234 --json ``` Prefixes must contain at least eight characters. Ambiguous prefixes are rejected with matching IDs rather than guessed. Because repeated observations can share a deterministic full audit ID, an exact full ID selects the newest matching snapshot. Report hints append the observation timestamp to that ID, matching the persisted `:` key so a saved report always resolves its original local snapshot. When the report was persisted, the hint also carries its absolute, shell-quoted `--db` path, including when the audit received a relative custom-store path. `--json` returns that snapshot selector together with the packet, audit IDs, runtime, generation time, and diagnostic version. Pre-0.21 snapshots remain readable, but they do not contain evidence packets. Xerg shows the legacy item and asks you to rerun. If a stale snapshot contains `tool_sequence_repetition_v1`, Xerg warns that the tool-name-only monetary method is retired and its amount should not be acted on. ## Snapshot requirement Later lookup requires the local snapshot written by `xerg audit`. If you audit with `--no-db`, the current report still includes item IDs and inline evidence, but replaces executable explain commands with a lookup-unavailable message because `xerg explain` cannot retrieve that report later. Rerun without `--no-db`, or use `--db` consistently for both commands. ## Argument-diversity limitation Xerg deliberately preserves timestamps, nonces, temporary paths, and other apparently volatile input values in v1. This prevents different invocations from being collapsed into false exact repetitions. It also means volatile values can make semantically repeated work look distinct, which inflates the distinct ratio and under-reports repetition. A 100% ratio means every observed canonical `(tool name, input)` pair differed. It does not prove that the calls were semantically different or successful. Generic field dropping is out of scope; future semantic normalization requires tool-specific allowlists and calibration.