Seeing runs ​
To list recent runs, run orc runs. Orc keeps its records in ~/.orc, outside every project repo. Set ORC_HOME to keep them somewhere else. To print the same facts as JSON, run orc runs --json. Every JSON document names its version first as schema, starting at orc.runs.v1. orc runs --json also reads the runs of each other machine in dispatch.json and adds them to runs. It reads each machine with the command that reaches it, as orc runner result does, and asks it for orc runs --json --local. A machine whose start names no reach command, such as orc runner start with nothing after --, is not read. A machine that fails, does not answer within 10 seconds, or prints JSON orc cannot read adds no runs. It is named in machines with its machine and its error, what it printed or why its answer did not read, and the command still exits 0. machines is empty when every machine answers, when dispatch.json names none, or when dispatch.json cannot be read, and the reason is on stderr. To print only this machine's runs, run orc runs --json --local. Poll with --local when you read every few seconds. Each run in it also gives title, the first line of its task file, and endReason, the reason it ended. Its step is the step running now, else the step that started last, as an id, name, attempt and status, and null before the first step. Its state is waiting while a running run's folder holds asking.json, else running, succeeded or failed from its status, and a stopped run reads failed. Each backlog task that waits in the runner's orc pueue group for this ORC_HOME adds one entry with state queued, listed before the recorded runs. A queued entry gives the backlog id as task, its orc/<task-id> branch, the checkout it will run in as project, and 0 questions. Its id, status and every other fact a run gains once it starts are null. When pueue cannot be reached, orc runs --json adds no queued entry and still prints the recorded runs. Its currentStep gives the name and attempt of the step running now, and null when no step runs or the run has ended. Its questions counts the questions a running run waits on you to answer, and is 0 otherwise. Its task is the backlog task id its orc/<task-id> or orc/<task-id>-parent branch names, and null on any other branch. Its machine is the dispatch machine the beat claimed for the run, which reaches orc run as ORC_RUN_MACHINE. A run you start by hand on this machine without ORC_RUN_MACHINE shows null. A run from another machine names that machine when its own record names none. The Agents tab and orc-dash read with --local, so they show only this machine's runs. Its worktree is the checkout the run works in: the task copy for a ship run, the project checkout for a report run. Runs recorded before these facts show each as null. To see one run step by step, run orc show <run-id>. It lists each step with its agent, session, signals, latest score and evidence, every file as a full link. To print the same facts as JSON, run orc show <run-id> --json. Its JSON document names its version first as schema, starting at orc.show.v1. Its run gives the same task, machine and worktree as orc runs --json. Each step gives the model and thinking level its agent ran with, and null for a step no agent ran. A resumed run lists first the steps it carries from each earlier run. The implement step comes first and the rest follow in the order they ran there. Each carried step gives the id of the run it ran in as from, read from the folder its evidence names, and every other step gives null. An implement step beside carried steps gives their run even when its own evidence names this run. orc show <run-id> prints from run <run-id> under each carried step. To print one step's output, run orc log <run-id> <step-id> with the step's id from that JSON. An implement step prints agent.log and a fix step its fix-<attempt>.log, which Pi appends to as it works. A checks or no-mistakes step prints the file it wrote. A step whose output is not written yet prints nothing. A carried step prints the log from the evidence path it records. A carried step recorded without a log path prints the log from the run named by its from. To score a step, run orc score <run-id> <step> <1 to 5> [note]. Use fix#1 to score the first fix attempt and plain fix scores the latest attempt in execution order. Use the step id from orc show JSON to score one resumed checks row when an attempt ran twice. Orc score refuses a score outside 1 to 5 with no write. Orc score refuses an unknown run or an unknown step with no write. Orc score keeps one row per score and orc show prints the latest score beside the signals. To count the gates and rules past runs kept hitting, run orc refusals. To print the same facts as JSON, run orc refusals --json. Its JSON document names its version first as schema, starting at orc.refusals.v1. Each row counts the runs that hit one gate, and one rule where the failure names a stable rule code, so two failing rounds in one run count once.