Read the JSON report ​
To print the report once, run orc-dash report session-quota. The report is one JSON object on one line, and SessionQuotaReport in packages/schema/src/entities/session-quota/contract/session-quota.ts decodes it. A reader checks schemaVersion first, and version 1 is the only version.
| Field | What it holds |
|---|---|
schemaVersion | 1 |
generatedAt | The ISO instant of the read |
scope | local, since the report covers only the machine it ran on |
estimate | basis local-logs, upperBound true, and the priceRatios of input, output, cache read and cache write tokens |
sources | The status of the quota, herdr and processes reads, each ok, unreachable or invalid with a detail, and a herdr not installed reads ok with no panes, the same as an installed herdr that lists none |
ledger | How many bridge copies, repeated lines, empty rows, fork copies and repeated Codex counts the count dropped, and the unjoinedCalls no account took |
accounts | Each account with its provider, accountKey, credentialKey, accountKeys, title, the joins that put calls on it and its quota read status |
accounts[].credentialKey | The account's own credential key, which stays the same whichever source read the account: the harness's native login, codex-home for Codex or default for Claude, when the account holds it, or else the first of quota-axi's accountKeys |
accounts[].accountKeys | Every credential key quota-axi lists for the account, so composition can tell that two machines reading one subscription under different credentials read one account |
windows | Each tracked window with its account, start, reset, used and left points, pace, elapsedFraction and gate |
windows[].gate | aheadOfPace, pastFirstTenth, and open when both hold |
windows[].exhaustsBeforeReset | Whether the window runs out before its reset, from the quota-axi verdict or the live sessions' total rate |
windows[].verdict | The quota-axi verdict for the window, null when the live rate decides instead |
windows[].source | quota-axi with its readAt, or harvested with the harness, via log, statusline or headless, the observedAt of the reading, its file and the join that put it on the account |
windows[] pace fields | sustainablePointsPerHour, the live burning activeSessions and the fairPointsPerHour each of them gets, both rates null once the window's reset has passed |
windows[].observed | The calls and measured tokens the logs hold for the window |
windows[].unattributedPoints | Used points the local logs cannot explain |
windows[].sessions | Each session's calls, measured tokens, trailingHourTokens, raw and priced estimates, models and flagged |
windows[].sessions[].models | Each model the session called in the window, most tokens first, with its model, calls, tokens, trailingHourTokens and priced estimate, whose points add up to the session's |
raw and priced | share of the window's local calls, contributionPoints as an upper bound, and pointsPerHour over the trailing hour |
sessions | Each session's id, kind, live, pane or pid, harness, cwd, place, label, log files and modelsSinceMidnight |
sessions[].modelsSinceMidnight | Each model the session called since local midnight on the machine that wrote the report, most tokens first, with its model, tokens and billing provider, across every route and quota reset |
unmatchedProcesses | Each log folder whose headless processes no log could be matched to, with its harness, cwd, place and pids |
flags | Each flagged session with its account, window, priced points per hour, fair points per hour and both shares |
unmetered | Calls on routes no subscription meters, such as an API key, with their tokens |
A report written before window sources and credential keys has no windows[].source, accounts[].credentialKey or accounts[].accountKeys, and still reads. Its windows count as read by quota-axi at its generatedAt. A report written before the capture ledger has no windows[].source.via, and its harvested windows read as read from a log. A report written before per-model rows has no windows[].sessions[].models, and its rows read as naming no model. A model is null where the log names none, as a Codex rollout's token counts do. A report written before them has no sessions[].modelsSinceMidnight, and its sessions read as naming no model since midnight. A modelsSinceMidnight entry written before billing attribution has no provider, and it reads with no provider.
A session id is pane:<pane id> for a Herdr pane, pid:<pid> for a headless process, and file:<log path> for a session that ended. Raw weighs every token alike, and priced weighs them by estimate.priceRatios. The ranking and the flag use the priced estimate.