[{"data":1,"prerenderedAt":37},["ShallowReactive",2],{"$f8nkn9bnu1dj2":3},{"href":4,"title":5,"description":6,"kind":7,"mark":7,"planned":8,"contributors":9,"provenance":7,"express":8,"bridge":8,"html":10,"headings":11},"\u002Fdocs\u002Fcontrol\u002Fself-improvement","Self-improvement","After an execution that went wrong, the OS proposes one change to one of its tasks. You decide, and an approved change is a new version you can see.",null,false,[],"\u003Ch2 id=\"what-it-does\">What it does\u003C\u002Fh2>\n\u003Cp>Self-improvement fixes how a task runs. After an execution that gave it something to learn from, the OS reviews the\nexecution against the definitions of its tasks and may propose \u003Cstrong>one\u003C\u002Fstrong> change to one of them: its skill, its tools,\nits gates, its spend or the outcomes that finish its work. Nothing changes until you approve it.\u003C\u002Fp>\n\u003Cp>It does not ask whether a team should exist, or chase a business number. See\n\u003Ca href=\"\u002Fdocs\u002Fcontrol\u002Fretrospectives\">Retrospectives\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch2 id=\"which-executions-get-a-pass\">Which executions get a pass\u003C\u002Fh2>\n\u003Cp>Every task is reviewed by the OS. You don't add anything to a task for it: a task you create, clone or copy is\ncovered from its first run. A pass is a run of the OS's review task, \u003Ccode>task-self-improvement\u003C\u002Fcode>, on your enterprise's\nmodel connection, and its cost counts like any run's.\u003C\u002Fp>\n\u003Cp>A task that names no edge to the review task is considered each time one of its runs ends. A task that names one as\na successor, for example \u003Ccode>{ &quot;slug&quot;: &quot;task-self-improvement&quot;, &quot;on&quot;: &quot;failed&quot; }\u003C\u002Fcode>, keeps that choice: one of its runs\nis considered only when one of those edges fires, on the run's outcome and the edge's \u003Ccode>when\u003C\u002Fcode>. With the example, a\nfailed run is considered and a successful one is not. Being considered is not the same as getting a pass: whether a\npass is due is still decided for the whole execution, below, not by the edge.\u003C\u002Fp>\n\u003Cp>A pass looks at the whole \u003Cstrong>execution\u003C\u002Fstrong>: every run of one piece of work, from the first task to the last. A loop\nthat sends work round three tasks is one execution, so it gets one pass that sees the loop, rather than three passes\nthat each see one clean lap. The pass may propose a change to any task of the execution, and for a loop that is often\nthe task whose outcome sends the work back.\u003C\u002Fp>\n\u003Cp>Not every execution gets a pass. The OS decides each time a run ends, and an execution gets one when:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>a \u003Cstrong>loop cap trips\u003C\u002Fstrong>, for one step or for the whole execution. The pass starts at once, while the work is still\nheld;\u003C\u002Fli>\n\u003Cli>its runs first show a \u003Cstrong>loop\u003C\u002Fstrong> (the same tasks round again) or the \u003Cstrong>same outcome repeated\u003C\u002Fstrong>;\u003C\u002Fli>\n\u003Cli>it \u003Cstrong>ended badly\u003C\u002Fstrong>: it failed, was blocked, was escalated to a person, or stopped on an outcome its task has no\nnext step for; or a run in it failed or had a failed tool call; or it looped, repeated an outcome or cost far more\nthan usual;\u003C\u002Fli>\n\u003Cli>or, as a \u003Cstrong>sample\u003C\u002Fstrong>, it finished cleanly with two or more runs. About one clean execution in 20 is sampled, and\nwhich ones is fixed by the execution itself, not chance. A clean execution of one run is never sampled, and neither\nis a cancelled one.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>There is no review every so many runs: a long execution is reviewed for its loops and repeats, once each.\u003C\u002Fp>\n\u003Cp>An execution gets no pass when:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>the run that ended is itself part of the loop: a task that decides recommendations, or a run carrying one out;\u003C\u002Fli>\n\u003Cli>its only problem is a run stopped because it went quiet (\u003Ccode>run_stalled\u003C\u002Fcode>), which no change to a task fixes;\u003C\u002Fli>\n\u003Cli>it already had a pass for the same problem, or another execution went wrong the same way in the last \u003Cstrong>24 hours\u003C\u002Fstrong>.\nThen the recommendation waiting for that problem counts one more recurrence instead;\u003C\u002Fli>\n\u003Cli>it is a sample, or a problem the OS has no name for, and its first task had a pass in the last \u003Cstrong>six hours\u003C\u002Fstrong>;\u003C\u002Fli>\n\u003Cli>it is one run long, and that task had a pass in the last six hours, so a chat or a conversational task costs no\nmore than before;\u003C\u002Fli>\n\u003Cli>every task in it already has a recommendation waiting. A task with one waiting is not proposed again, but the\nother tasks of the execution still can be.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Each time an execution is passed over, the run that ended says why in its events (\u003Ccode>improvement.skipped\u003C\u002Fcode>, with the\nreason).\u003C\u002Fp>\n\u003Cp>A pass keeps at most one recommendation: the one it is surest will help, and never a cosmetic one. The OS checks the\nanswer itself before keeping it: a change to a task outside the execution, to a task that already has one waiting,\nor to one your enterprise may not change, is dropped, and so is one the review is less than 80% sure of. A pass that\nfinds nothing worth changing proposes nothing, and that is a good result. Only tasks your enterprise owns are\nproposed: a \u003Ca href=\"\u002Fdocs\u002Fwork\u002Ftasks#clone-and-copy\">clone\u003C\u002Fa> is not, and neither are the OS's own tasks.\u003C\u002Fp>\n\u003Ch2 id=\"a-task-that-keeps-stopping\">A task that keeps stopping\u003C\u002Fh2>\n\u003Cp>A run that ends on an outcome its task has no next step for stops its pipeline, and the run's page says so. When a\ntask's successful runs do that three or more times in a row, the task itself is the problem: \u003Cstrong>Blockers\u003C\u002Fstrong> lists it\nunder task problems and the task's page shows the same warning, naming the outcome its runs end on and how many\nstopped. A run that routes clears it. A scheduled task shows this way too, though its runs have no issue or work\nitem for Blockers to list.\u003C\u002Fp>\n\u003Ch2 id=\"deciding\">Deciding\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>Operations → Self-Improvement\u003C\u002Fstrong> lists the recommendations waiting for you, each with its task, what it proposes,\nhow sure the review is, and when it was raised. \u003Cstrong>Self-Improvement\u003C\u002Fstrong> in the sidebar counts them. They are not\nblockers: nothing waits on them, so \u003Cstrong>Blockers\u003C\u002Fstrong> does not list them. \u003Cstrong>Tasks\u003C\u002Fstrong> shows each task's count, and a task's\npage lists its own.\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Decision\u003C\u002Fth>\n\u003Cth>What happens\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Decline\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Needs a reason. The recommendation is kept, declined, and the review sees it next time, so the same change is not proposed again without new evidence.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Approve\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>The OS carries it out: it writes the task's next version with the change.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>Only the enterprise that owns a task can approve a change to it.\u003C\u002Fp>\n\u003Cp>A task given the OS's recommendation tools can decide recommendations too, and can also mark one \u003Cstrong>actioned\u003C\u002Fstrong>:\nsomeone else takes the work, tracked by the issue or pull request it links. Every decision records who made it, a\nperson or a run, and \u003Cstrong>Decided in the last 60 days\u003C\u002Fstrong> lists them with their reasons.\u003C\u002Fp>\n\u003Ch2 id=\"statuses\">Statuses\u003C\u002Fh2>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Status\u003C\u002Fth>\n\u003Cth>Means\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Ccode>pending\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Waiting for a decision. While one waits, its task gets no new pass.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>declined\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Turned down, with the reason.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>actioned\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Handed to work tracked elsewhere, with its link.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>implementing\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Approved; the change is being written.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>complete\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>The new version is published.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>error\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>The change could not be made: the implementing run failed, was blocked or cancelled, or the change broke a rule below. You can approve or decline it again from its page.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Ch2 id=\"what-the-change-may-never-do\">What the change may never do\u003C\u002Fh2>\n\u003Cp>An approved change is carried out by one of the OS's own tasks, on your enterprise's model connection. It may change\nonly the task's skill, tools, gates, spend and the outcomes that finish its work (\u003Ccode>finishesOn\u003C\u002Fcode>), and it is refused\nif it would:\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Refused\u003C\u002Fth>\n\u003Cth>Code\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>Give the task a tool it did not have\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>tools_widened\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Leave a task that has tools with none\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>tools_emptied\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Remove a gate\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>gates_removed\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Raise the task's spend\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>spend_increased\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Raise its model tier, or remove it\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>tier_increased\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Write a repository command into the skill\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>repo_command_in_skill\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Write your own data (an id, an email address) into the skill\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>tenant_data_in_skill\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Leave a skill that cannot be read\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>unparseable_workflow\u003C\u002Fcode>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>The goal, the success metric, the guardrails, the inputs, the workspace, the successors and the schedule are never\ntouched. A change that needs any of those, or needs more access, is yours to make by editing the task.\u003C\u002Fp>\n\u003Ch2 id=\"the-new-version-and-going-back\">The new version, and going back\u003C\u002Fh2>\n\u003Cp>An approved change is published as the task's next \u003Ca href=\"\u002Fdocs\u002Fwork\u002Ftasks#versions\">version\u003C\u002Fa>, like any edit. Runs already\nunder way keep the version they started on; the next run uses the new one. Any other recommendation still waiting on\nthe task is declined as superseded, since it was written against the old version.\u003C\u002Fp>\n\u003Cp>To undo a change, open the previous version in the task's \u003Cstrong>Version history\u003C\u002Fstrong> and save its content as a new version.\nThere is no one-click rollback.\u003C\u002Fp>\n\u003Ch2 id=\"over-the-api\">Over the API\u003C\u002Fh2>\n\u003Cp>With an \u003Ca href=\"\u002Fdocs\u002Fapi\u002Fauthentication\">API token\u003C\u002Fa> scoped \u003Ccode>recommendations:read\u003C\u002Fcode> (and \u003Ccode>recommendations:write\u003C\u002Fcode> to decide):\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Call\u003C\u002Fth>\n\u003Cth>What it does\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Ccode>GET \u002Fv1\u002Frecommendations\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>The recommendations waiting for a decision, or with \u003Ccode>view=decided\u003C\u002Fcode>, those decided in the last 60 days.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>GET \u002Fv1\u002Frecommendations\u002F{id}\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>One recommendation, in full.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>POST \u002Fv1\u002Frecommendations\u002F{id}\u002Fdecision\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Approve or decline one. A decline needs a reason. Send an \u003Ccode>idempotencyKey\u003C\u002Fcode> and a retried call returns the first decision rather than deciding twice.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n",[12,16,19,22,25,28,31,34],{"id":13,"text":14,"level":15,"planned":8},"what-it-does","What it does",2,{"id":17,"text":18,"level":15,"planned":8},"which-executions-get-a-pass","Which executions get a pass",{"id":20,"text":21,"level":15,"planned":8},"a-task-that-keeps-stopping","A task that keeps stopping",{"id":23,"text":24,"level":15,"planned":8},"deciding","Deciding",{"id":26,"text":27,"level":15,"planned":8},"statuses","Statuses",{"id":29,"text":30,"level":15,"planned":8},"what-the-change-may-never-do","What the change may never do",{"id":32,"text":33,"level":15,"planned":8},"the-new-version-and-going-back","The new version, and going back",{"id":35,"text":36,"level":15,"planned":8},"over-the-api","Over the API",1794928142858]