impl-tool:measurement
impl-tool:measurement@0.0.0
modelRunRecord
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
task | string | Yes | The kind of call this was, as the runner names its tasks -- iteration for a session turn, pr_description for a pull request description | |
provider | string | Yes | The agent that served the call, by its configured name | |
model | string | No | "" | The model that answered; empty where the provider names none |
attempt | integer | No | 1 | Which try at the task this call was, counted from 1 |
outcome | enum one of ok, invalid, error | Yes | How the call ended: ok when it answered and the answer validated, invalid when it answered and the answer did not validate, error when the call itself failed | |
elapsed | duration | Yes | How long the call ran | |
promptChars | integer | No | 0 | Size of the prompt sent, in characters; 0 where it was not measured |
responseChars | integer | No | 0 | Size of the answer received, in characters |
tokens | integer | No | 0 | Tokens the call consumed as the provider reported them, with input, output, cache reads and cache writes summed. 0 means the provider reported no usage, not that the call was free |
inputTokens | integer | No | 0 | Input tokens the provider reported, counted apart from cache reads and cache writes |
outputTokens | integer | No | 0 | Output tokens the provider reported |
cacheReadTokens | integer | No | 0 | Tokens the provider reported as read from its prompt cache |
cacheWriteTokens | integer | No | 0 | Tokens the provider reported as written to its prompt cache |
scope | string | No | "." | The repository the call was made for: . for the top-level repository, otherwise the submodule's path |
issue | string | No | "" | The id of the issue the call was made for, as it was spelled at the time; empty where the call belonged to no issue |
at | string | No | "" | When the call ended, as an ISO 8601 timestamp to the second |
detail | string | No | "" | Why the answer was rejected or the call failed; empty for a call that succeeded |
date | date | No | "0001-01-01" | The day the call ran, in the configured timezone |
timeStarted | time | No | "00:00:00" | When the call began, timezone included |
timeEnded | time | No | "00:00:00" | When the call finished, timezone included |
weightedTokens | number | No | 0 | The consumption in input-equivalent tokens |
planCost | currencyValue | No | 0 | The estimated share of the subscription the call consumed |
tokenCost | currencyValue | No | 0 | What the tokens cost as metered spend |
iteration | integer | No | -1 | The session turn the call belongs to, numbered from 0; -1 where no turn owns the call |
increment | integer | No | -1 | The commit increment the call belongs to, numbered from 0; -1 where it belongs to none |
succeeded | boolean | No | false | Whether the work landed. For a session turn: its result parsed and its commit and push completed. For any other call: the answer satisfied its contract |
planObserved | number | No | 0 | Used percentage of the account's seven-day plan window, as last read when the row was written; 0 means no fresh reading was on hand |
sessionUuid | string | No | "" | Identifier of the runner session the call was made in; empty where it ran outside one |
invocationUuid | string | No | "" | Identifier of the impl-tool invocation that made or recorded the call; empty where there was none |
conversationUuid | string | No | "" | Identifier the provider gives the conversation the call belonged to; empty where the provider has none |
issueUuid | string | No | "" | The issue's permanent identifier, taken from its queue entry; empty where the run carried none |
rating | number | No | 0 | A quality score for the run. Reserved: nothing writes it, and 0 means unrated |
modelRunLog
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
runs | list[modelRunRecord] | No | [] | Every call on record, one entry per call |
modelRunSummary
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
task | string | Yes | The task the calls served | |
provider | string | Yes | Which backend served them | |
model | string | No | "" | Which model, empty where the provider names none |
calls | integer | No | 0 | How many calls the group holds |
ok | integer | No | 0 | How many answers satisfied their contract |
seconds | number | No | 0 | Total elapsed over the group, in seconds |
promptChars | integer | No | 0 | Total prompt size over the group |
tokens | integer | No | 0 | Total measured tokens; zero means unmeasured, not free |
weightedTokens | number | No | 0 | Total consumption in input-equivalent tokens |
planObserved | number | No | 0 | The account reading stamped on the latest row |
planDerived | number | No | The tokens as a share of the plan window, once priced | |
planCost | number | No | The share in subscription dollars, once priced | |
tokenCost | number | No | 0 | Total metered dollars over the group |
planSpent | number | No | 0 | Total stamped plan dollars over the group |
modelTiming
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
task | string | Yes | The task the calls served | |
provider | string | Yes | The agent that served them | |
model | string | No | "" | The model that answered; empty where the provider names none |
calls | integer | Yes | How many calls the group holds | |
ok | integer | No | 0 | How many of the calls returned an answer that satisfied its contract |
elapsed | duration | Yes | Total time the calls ran, summed over the group | |
promptChars | integer | No | 0 | Total prompt size over the group, in characters |
tokens | integer | No | 0 | Total tokens the providers reported over the group; 0 means unmeasured, not free |
weightedTokens | number | No | 0 | Total consumption in input-equivalent tokens, each kind of token weighted by its price relative to an input token |
planDerived | number | No | 0 | The group's tokens as an estimated percentage of the plan's weekly window; 0 means not priced |
planCost | number | No | 0 | That share in subscription dollars, at the rates in force when the table was made; 0 means not priced |
planSpent | number | No | 0 | Plan dollars stamped on each call when it was made, summed over the group |
tokenCost | number | No | 0 | Metered dollars summed over the group; 0 while the account is inside its plan limits |
planObserved | number | No | 0 | Used percentage of the plan window as read when the group's latest call was recorded; 0 means no fresh reading |
timingDisplay
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
unit | enum one of microseconds, milliseconds, seconds, minutes, hours | No | "milliseconds" | The unit rendered durations are read in |
column
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
name | string | Yes | The column's name in its table | |
ddl | string | Yes | The column's SQL declaration: its type, constraints and default | |
fill | string | Yes | The SQL literal a query fills the column with for a row that does not carry it | |
field | string | No | "" | The record field the column stores; empty for a column that stores none |
kind | string | No | "plain" | How a value converts between the field and the column: plain passes through, stamp stores a timestamp as text, micros stores a duration as whole microseconds, bool stores 0 or 1, and currency_value and currency_code store the two halves of a currency amount |
checkRun
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
uuid | string | Yes | The run's own identifier, assigned when the command ran; a run is filed once under it | |
issue | string | Yes | The id of the issue the run was made for, as spelled when the row was filed | |
issueUuid | string | No | "" | The issue's permanent identifier; empty where the session's launch carried none |
increment | integer | No | 0 | The commit increment of the turn that ran it; for a turn that committed nothing, the increment of the commit that lands its work |
kind | enum one of check, test | No | "check" | Whether the run was one of the gate's checks (check) or a test run (test) |
name | string | Yes | The command and the flags it was given, as the gate names the check | |
arguments | string | No | "" | The rest of what the run was given, as it would be typed -- a test run's selection and tiers, whether it crossed into pinned repositories. A withheld or overlong argument is named without its value |
repository | string | No | "." | The checkout the run acted on, relative to the run's root: . for the root itself |
startedAt | datetime | Yes | When the run started | |
elapsed | duration | Yes | How long the run took, measured by the tool that ran it | |
exitCode | integer | No | 0 | The exit code the command ended with |
passed | boolean | No | false | Whether the run exited 0 |
checkAverage
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
kind | enum one of check, test | No | "check" | Whether the runs were of one of the gate's checks (check) or test runs (test) |
name | string | Yes | The check as the gate names it: the command and the flags it was given | |
arguments | string | No | "" | The rest of what the runs were given, as it would be typed |
repository | string | No | "." | The checkout the runs acted on, relative to the run's root: . for the root itself |
runs | integer | Yes | How many runs of the check are filed for the issue | |
passed | integer | No | 0 | How many of those runs exited 0 |
elapsed | duration | Yes | Total time the runs took, summed; divide by runs for the average |