Next up
/:owner/:name/next is the one ranked list the crew builds from (#8694). It reads as one numbered queue in the dispatcher's order (#11619): each row has its position, title, who is on it (or "next"), and one word of state: building, checking, in review, landing, waiting on #N, held or ready. A pull that is open on the issue sets building, checking or in review. Three native folds hold the rows, each with a count: "Up next" (the first ten, open), "After" (the rest) and "Later" (behind the Later gate). There is no script for the folds and no ?all=1; the folds are keyboard and screen-reader reachable at any width. The terminal's Next up screen shows the same numbers and state words.
- Role: each row is tagged Blocker (it holds other open work up: "finishing it frees N") or Enabler (it holds nothing up; its line says what it serves, else that the owner ranked it). The tag is a label only: it never changes the order the picker takes work in.
- Move:
Top(t),Up(u),Down(d) andNot now(l) are forms on each row, with the key shown beside each (Shift+J/K or Alt+↓/↑ also move down and up; Alt+Shift combos do nothing);j/kchoose the row, and a row can be dragged. Each is onePOST /:owner/:name/next(number,to=top,not-noworafter:<number>) and saves at once. Writers only; the same-origin check applies. - Top lifts the gate:
Topon a deferred issue removes itswait_foredge to the Later gate (and the Later umbrella's edge to it) in the same action. Work that only deferred work needs stays deferred, and the refusal says so. - Owner's pick: a row a person placed (page, terminal or MCP) is marked "Owner's pick" on the page and in the terminal. Automatic ranking leaves it at its place and orders only the other rows around it; only a person's move changes it, and
Not nowclears it along with the rank. A move of another row does not mark the rows the planner placed. Nothing calls the automatic mode yet (writeOrderwithowner: false): the guarantee is in place and tested, ahead of its first caller. - Not now links the issue to wait for the open Later gate and drops its rank.
- The picker takes the highest ranked startable issue before anything else, small work included (
readyNow,slotOrder). A ranked issue something still blocks is skipped, never waited on. Same state, same pick. - MCP:
next_up { app, issue?, to?, all? }reads the list, or moves one issue with the same words. - A rank is cleared when the issue closes. Dependencies order within the list and are shown, never edited here.
Terminal: the palette's "Next up in a repository" opens the same list; t, u, d and l are Top, Up, Down and Not now on the row under the cursor, through the same action as the page and MCP. A reader sees the order and the keys do nothing; a move made anywhere reorders an open screen live. At most 200 issues are ranked; a move past that is refused and says so.
Filter (#11867): a plain GET form above the list narrows /next to one pillar (?pillar=P7 or Later, typed in a box) and/or a word (?q=cache, matched without regard to case, shown as typed, at most 80 characters). A filter only hides rows: each keeps its real place in the whole ranking, the cut line stays where the whole list puts it (drawn only between shown rows), and a move made under a filter moves in the whole order and returns to the same filter. A line says "N of M shown", and that the week's capacity bar is left out while filtered. An unknown or malformed pillar shows an empty list that says so. Terminal: / asks for a pillar and/or a word (P7 cache); an empty answer clears it; ranks, t u d l and the header count behave as on the page.
The chapter line (#12681): under the title, once the repository has chapters, Ranked for <open chapter>: <progress sentence> (linked), or "No chapter is open" with a link to draft one; a repository that never drafted a chapter shows no line. Terminal: the screen prints the same line under the title, from the same function (chapterLineParts in core/chapters.ts), wrapped to the width and never cut (#13254, #12991).
The growth line (#12290): under the title, two dl rows: Week (issues opened in the last seven days, who opened them, how many closed, and what is open now: "2 issues opened (2 by owner), 0 closed. 2 open (2 issues, 0 pulls).", #12867: one row, so the numbers are said once) and Planner (on, paused until a time in UTC, or paused until the owner switches it on). The owner (a person who administers the repository) also gets the switch in the Planner row: a button in a post form to the same planner_switch action as the queue page, sent back to /next, so the pause is one tap and the owner stays on the list. A reader sees both rows and no button. Both rows come from one read (weekLine, which also counts open issues and pulls and what opened and closed today), sealed issues left out for a viewer who may not see them. Terminal: Next up shows the same sentence (growthWords, pauseSentence) above the list, and the owner presses p to pause the planner until they switch it on, and p again to switch it on; for a reader and for an agent the key does nothing.
The week is the sprint (#8704)
The subscription's weekly reset is the sprint boundary; what the crew may still spend before it is the capacity (flow/week.ts, from core/standing.ts). No ceremony and no locked scope: rerank any time and the line moves on the next paint.
- Estimates: each row shows a cost range (the 25th to 75th percentile of what past closed issues of the same size and kind cost, every agent run on them summed,
flow/estimate.ts). With fewer than three comparable issues it falls back to size alone, then says "too little history". Measured, never asked of an agent: no run is spent. - Cut line: walking from the top, the running median estimate is held against the capacity; the line falls before the first issue that does not fit and everything under it reads as next week. An issue with no estimate counts as the median of the others; with none estimated, or no capacity read, there is no dollar line and the sentence says why. Same list and history, same line.
- Pace line: when the dollar line is unsure (no capacity read, or nothing estimated), the line is drawn by pace instead (
paceCut): how many issues landed in the last seven days, carried over the days left to the reset. It is worded "By pace: this week ends here" and never says "fit", because it makes no dollar claim; with fewer than three landed the sentence says the pace is a guess. Once the cost line exists, it replaces the pace line. - Projection: one sentence above the list: how many should ship by the reset, as a range from the high and low ends of the estimates, and a confidence (high at eight or more samples on every row).
- Retrospective: once an hour the forge keeps the projection of the open window (
week_projection) and, for a window that has ended, writesweek_retroonce (flow/retro.ts): what shipped against the projection, how many shipped issues cost inside their range, what open work cost most, what is stuck (two failed runs or more, and the class of the last failure, never the run's own output), and at most three proposals./:owner/:name/weekand the MCPweektool show it. A proposal is one tap (POST /:owner/:name/week, orweek { app, take }) by a writer who is a person; it puts the issue behind the Later gate, the same move asNot now, and is taken at most once. Nothing changes without that tap. Only a person who writes can take one:takeProposalrefuses an agent and a reader for every door. - Not done here: the line falls between issues, not whole scopes, because bets and scopes (#8713) are not in the forge yet; when they land the line moves to their edges. The health mark per group is left for the same reason.
- Live: a drag posts the move and the page reloads itself from the server, which redraws the line; there is no manual reload, but it is a reload, not a client-side recompute.
- Terminal: Next up shows each row's range and a muted "this week ends here" line under the last issue that fits (by pace, "by pace: this week ends here"), with the projection in the header. "The week in a repository" (palette) shows the retrospective;
j/kchoose a proposal andenteroratakes it, through the sametakeProposal. A reader or an agent sees it and the key does nothing; a new retrospective (week.retro) or a proposal taken anywhere updates the open screen. MCPnext_uprows carryestimateandthisWeek(above the pace line when the line is by pace), with aprojectionsentence. - Sealed issues are left out of history and of the retrospective, so their cost never shapes a visible range.