@plot-pm/board 0.14.1 → 0.14.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@plot-pm/board",
3
- "version": "0.14.1",
3
+ "version": "0.14.3",
4
4
  "description": "Local Kanban board for Plot — a glanceable view of plan phases from docs/plans, with sprint and story filters",
5
5
  "type": "module",
6
6
  "license": "MIT",
package/plot-deliver.sh CHANGED
@@ -419,6 +419,70 @@ decide_transition() { # $1=file → prints "<Phase>\t<record>\t<write|already>"
419
419
  # writes — and appending a second is how one plan came to hold two `Delivered:`
420
420
  # lines (2026-09-01). The domain returns the written record unchanged in that
421
421
  # case, so the phase still flips and nothing is inserted.
422
+ # Would the PARSER read this phase out of the file we are about to land?
423
+ #
424
+ # THE WRITE SUCCEEDING IS NOT THE OUTCOME HOLDING, and that gap is #924. A plan
425
+ # carrying BOTH front matter and a `## Status` block was delivered in a project
426
+ # repo: `flip_phase` wrote `Delivered` into the block, `mv` landed it, the
427
+ # summary said `phase=flipped`, and `plot-plan-meta.sh` went on answering
428
+ # `approved` — because it preferred front matter wherever it existed. The write
429
+ # took effect on bytes nobody reads.
430
+ #
431
+ # THAT PRECEDENCE INVERTED IN #933, so the two-record plan now delivers rather
432
+ # than refusing here: the parser reads the `## Status` block, which is the field
433
+ # every lifecycle script writes. This gate is unchanged and is not softened —
434
+ # its condition simply stops holding for that shape. It still fires on a scratch
435
+ # copy the parser cannot read, and it still asks the parser rather than trusting
436
+ # that awk changed a line.
437
+ #
438
+ # `flip_phase`'s awk matches only inside `section == "status"`. That one guard
439
+ # IS the defect: on a front-matter plan it edits the block and leaves the front
440
+ # matter untouched, and returns 0 for having changed something.
441
+ #
442
+ # SO THE TEST IS WHAT THE PARSER ANSWERS, NEVER WHETHER AWK CHANGED A LINE.
443
+ # Refusing on `flipped=0` would break every re-run of a correct delivery — a
444
+ # plan already carrying `Delivered` flips nothing and is fine. This asks the one
445
+ # question that distinguishes them: read the scratch copy the way every later
446
+ # consumer will read the plan, and compare.
447
+ #
448
+ # IT RUNS ON THE SCRATCH COPY, BEFORE THE `mv`, and the caller passes whichever
449
+ # file that arm is about to land — `$a` on the `recorded=yes` arm, `$b` on the
450
+ # other. Parsing `$a` on the record arm would check content that never reaches
451
+ # the plan. After the `mv` is too late twice over: the script's own header
452
+ # documents exit 0 as *"the plan is Delivered on the default branch"*, so
453
+ # refusing there would exit 1 on a run meeting the documented success
454
+ # condition — and `runAutoDeliver` spawns this detached, logging a non-zero exit
455
+ # to nobody while the plan sits delivered on main.
456
+ #
457
+ # AN UNREADABLE SCRATCH COPY REFUSES. `decide_transition` already takes that
458
+ # line — *"refusing rather than guessing"* — and a file the parser cannot read
459
+ # is exactly the state this gate exists to keep off the plan.
460
+ phase_would_read() { # $1=scratch file $2=expected phase (lowercase) → 0 agrees, 1 refuses
461
+ local scratch="$1" want="$2" m got
462
+ m=$(bash "$script_dir/plot-plan-meta.sh" "$scratch" 2>/dev/null) || m=""
463
+ if [ -z "$m" ]; then
464
+ echo "plot-deliver: $rel — the written file does not parse, so the delivery was not landed." >&2
465
+ echo " Nothing was written. Re-run after fixing the file." >&2
466
+ return 1
467
+ fi
468
+ got=$(printf '%s' "$m" | jq -r '.phase // ""')
469
+ [ "$got" = "$want" ] && return 0
470
+
471
+ # THE REFUSAL NAMES BOTH VALUES AND THE FILE. A message saying only "delivery
472
+ # failed" throws away the half a person acts on: which phase was written, and
473
+ # which one the parser still reads. The cause is named too, because the file
474
+ # holding two records of one fact is the thing to fix — and which format ought
475
+ # to win is a decision this gate deliberately leaves to a person.
476
+ echo "plot-deliver: $rel — wrote phase '$want', but the parser still reads '$got'." >&2
477
+ echo " The write landed and the parser reads something else, so the delivery" >&2
478
+ echo " would have reported a success it did not achieve." >&2
479
+ echo " Nothing was written — the plan is unchanged. Check what the plan says" >&2
480
+ echo " its phase is, and where: a plan stating it in two places reports the" >&2
481
+ echo " '## Status' block, which is the field every lifecycle script writes." >&2
482
+ echo " See what the parser reads: $script_dir/plot-plan-meta.sh $rel" >&2
483
+ return 1
484
+ }
485
+
422
486
  write_transition() { # $1=file $2=record $3=recorded(yes|no) → sets phase_report record_report
423
487
  local f="$1" record="$2" recorded="$3" a="$1.plot-phase" b="$1.plot-record" flipped=0
424
488
 
@@ -428,6 +492,11 @@ write_transition() { # $1=file $2=record $3=recorded(yes|no) → sets phase_repo
428
492
  [ -s "$a" ] || { rm -f "$a"; echo "plot-deliver: could not read $rel" >&2; return 1; }
429
493
 
430
494
  if [ "$recorded" = "yes" ]; then
495
+ # THE DRY RUN, on the file this arm is about to land. A refusal discards the
496
+ # scratch copy and leaves the plan byte-identical — and never reaches
497
+ # `record_state_receipt`, which would otherwise license a commit of a state
498
+ # that was refused.
499
+ phase_would_read "$a" delivered || { rm -f "$a"; return 1; }
431
500
  mv "$a" "$f" || { rm -f "$a"; return 1; }
432
501
  record_report="already"
433
502
  else
@@ -438,6 +507,9 @@ write_transition() { # $1=file $2=record $3=recorded(yes|no) → sets phase_repo
438
507
  echo " with no record is invisible to the scan. Fix the section and re-run." >&2
439
508
  return 1
440
509
  fi
510
+ # `$b` AND NOT `$a`: this arm lands the file carrying the record, so `$a`
511
+ # is content that never reaches the plan.
512
+ phase_would_read "$b" delivered || { rm -f "$a" "$b"; return 1; }
441
513
  mv "$b" "$f" || { rm -f "$a" "$b"; return 1; }
442
514
  rm -f "$a"
443
515
  record_report="written"
package/plot-dispatch.sh CHANGED
@@ -56,6 +56,25 @@
56
56
  # tradition of --allow-local: a gate with no exit is one people
57
57
  # route around by never annotating at all. It says so on the
58
58
  # line it overrides, so the override is on the record.
59
+ # --agent <name> dispatch this run's agents under the charter
60
+ # `.plot/charters/<name>.json`. It SETS `PLOT_AGENT`, which was
61
+ # the input all along and which nothing chose: the charter
62
+ # mechanism shipped in v2.17.0 with 16 readers for `harness` and
63
+ # no selector at all, so a kind could be declared and never
64
+ # asked for. The choice is EXPLICIT and never inferred — a
65
+ # matcher reading a plan could guess a kind, and a guess that is
66
+ # usually right produces a fleet whose wrong answers cannot be
67
+ # explained. An already-exported PLOT_AGENT is overridden, since
68
+ # a flag on this run is the more specific answer.
69
+ # AMENDED 2026-09-15: a slice may now DECLARE its kind, and
70
+ # `start_worker` reads it where this flag left `PLOT_AGENT`
71
+ # unset. That is not the matcher refused above — a declaration is
72
+ # a field a person wrote, and nothing here ranks candidates or
73
+ # infers a kind from a slice's contents. The flag still wins,
74
+ # because a flag typed on this run is more specific than a field
75
+ # written when the plan was drafted. A charter this clone does
76
+ # not hold is REPORTED and dispatched anyway; see
77
+ # `plan_declared_agent`.
59
78
  # <slug> the plan to fan out
60
79
  # Output: one line per branch, each optionally followed by an indented
61
80
  # `in flight:` line naming a branch that already holds files, then the
@@ -230,6 +249,7 @@ allow_waiting=0
230
249
  max=0
231
250
  slug=""
232
251
  migrate_yes=0
252
+ agent=""
233
253
  while [ $# -gt 0 ]; do
234
254
  case "$1" in
235
255
  --dry-run) dry_run=1 ;;
@@ -270,17 +290,56 @@ while [ $# -gt 0 ]; do
270
290
  --offline|--no-fetch) offline="--offline" ;;
271
291
  --allow-local) allow_local=1 ;;
272
292
  --allow-waiting) allow_waiting=1 ;;
293
+ # THE VALUE IS REQUIRED AND ALWAYS CONSUMED, which is the opposite of the
294
+ # rule `--stop`, `--restart` and `--start` follow — and the difference is
295
+ # the value's SHAPE rather than a change of mind. Those three take a branch
296
+ # (`*/*`) or a count (digits), each recognisable on sight, so an absent one
297
+ # can be left for the parser. An agent name is a bare word and so is a plan
298
+ # slug: nothing tells `--agent reviewer` from `--agent` followed by the
299
+ # slug. Leaving it unconsumed would let the `*)` arm below silently take
300
+ # the agent name as the plan to dispatch, which reads as "no such plan" and
301
+ # names neither what was asked nor what went wrong. So a missing value
302
+ # REFUSES rather than guessing.
303
+ --agent) agent="${2:?--agent needs a charter name, e.g. --agent reviewer}"
304
+ case "$agent" in
305
+ -*) echo "plot-dispatch: --agent needs a charter name, got '$agent'" >&2; exit 1 ;;
306
+ esac
307
+ shift ;;
273
308
  --max) max="${2:?--max needs a value}"
274
309
  case "$max" in
275
310
  ''|*[!0-9]*) echo "plot-dispatch: --max needs a number, got '$max'" >&2; exit 1 ;;
276
311
  esac
277
312
  shift ;;
278
- -h|--help) sed -n '2,59p' "$0"; exit 0 ;;
313
+ # THE RANGE MOVED WITH THE HEADER IT PRINTS. It ended at `<slug>` and still
314
+ # does; adding `--agent` above pushed that line from 59 to 69, and
315
+ # documenting the plan-declared kind pushed it from 69 to 78. Two records
316
+ # of one fact, and nothing compares them — a stale number here silently
317
+ # truncates the help rather than failing, so it is checked by a test.
318
+ -h|--help) sed -n '2,78p' "$0"; exit 0 ;;
279
319
  *) slug="$1" ;;
280
320
  esac
281
321
  shift
282
322
  done
283
323
 
324
+ # `--agent` SETS THE INPUT THAT ALREADY EXISTED, and it sets it in ONE place.
325
+ #
326
+ # `PLOT_AGENT` was read in three places before this flag — `resolve_launch`'s
327
+ # call in `start_worker`, the `--capabilities` block, and the forwarded export
328
+ # into the worker's environment — and assigned in none: `plot-dispatch.sh` held
329
+ # `PLOT_AGENT="${PLOT_AGENT:-}"`, a pass-through of whatever the operator had
330
+ # already exported. So the selector existed as a variable nothing chose.
331
+ #
332
+ # ONE ASSIGNMENT RATHER THAN A THREADED ARGUMENT. Every reader already asks
333
+ # `${PLOT_AGENT:-}`; exporting here reaches all three with no further change and
334
+ # leaves no call site that could be missed when a fourth reader is added. The
335
+ # flag's whole meaning is "set PLOT_AGENT", which is also what keeps it from
336
+ # becoming a second selector competing with the variable.
337
+ #
338
+ # THE FLAG WINS OVER AN INHERITED VALUE. A dispatched worker runs with its own
339
+ # `PLOT_AGENT` in the environment, so a run launched from inside one would
340
+ # otherwise inherit a kind nobody asked for on this dispatch.
341
+ [ -n "$agent" ] && export PLOT_AGENT="$agent"
342
+
284
343
  # THE TWO PRECONDITIONS OF A DISPATCH, and a sourcing test has neither.
285
344
  # `PLOT_DISPATCH_SOURCED=1` is taking the definitions below rather than running
286
345
  # a dispatch, so it has no slug and often no git repository; the guard that
@@ -930,6 +989,42 @@ start_worker() {
930
989
  local branch="$1" wt="$2"
931
990
  local cmd
932
991
 
992
+ # THE PLAN IS THE DEFAULT AND THE FLAG IS THE OVERRIDE.
993
+ #
994
+ # `[ -z "${PLOT_AGENT:-}" ]` is the whole precedence rule: `--agent` exported
995
+ # the variable before this line was reached, so a flag typed on this run stops
996
+ # the plan being read at all. A field written when the plan was drafted is the
997
+ # less specific answer — the same order `--agent` already applies to an
998
+ # inherited `PLOT_AGENT`.
999
+ #
1000
+ # ONE ASSIGNMENT, FOR THE REASON `--agent` GIVES ONE FIELD OVER. Four readers
1001
+ # ask `${PLOT_AGENT:-}` — `resolve_launch` below, the capabilities block, the
1002
+ # manifest, and the export into the worker environment — so setting the
1003
+ # variable reaches all four and leaves no call site to miss when a fifth
1004
+ # appears.
1005
+ #
1006
+ # A FREE AGENT HAS NO BRANCH and so has no plan to read: `--start` calls this
1007
+ # with `branch` empty, and `plan_declared_agent` returns 1 on it immediately.
1008
+ if [ -z "${PLOT_AGENT:-}" ] && [ -n "$branch" ]; then
1009
+ local declared charter
1010
+ if declared=$(plan_declared_agent "$branch"); then
1011
+ export PLOT_AGENT="$declared"
1012
+ # A MISSING CHARTER IS REPORTED AND DISPATCHED ANYWAY. Said HERE rather
1013
+ # than only in the manifest, because this is where a person watching the
1014
+ # run can still act on it: the launch below falls back to the repo default
1015
+ # and is otherwise silent about a declaration that reached nothing.
1016
+ charter=$(charter_file_for "$declared")
1017
+ if [ -n "$charter" ] && [ ! -f "$charter" ]; then
1018
+ echo " $branch declares agent '$declared', and no charter answers to that name"
1019
+ echo " looked for $charter"
1020
+ echo " starting it on the repo's 'Worker command' — a plan written where that"
1021
+ echo " charter exists must stay dispatchable on a clone that lacks it."
1022
+ else
1023
+ echo " $branch declares agent '$declared' — its plan selected the charter"
1024
+ fi
1025
+ fi
1026
+ fi
1027
+
933
1028
  # RESOLVED BEFORE `Worker command`, because the charter is the more specific
934
1029
  # answer and the config key is the fallback rather than the sole source.
935
1030
  #
@@ -1359,6 +1454,78 @@ start_worker() {
1359
1454
  return 0
1360
1455
  }
1361
1456
 
1457
+ # WHICH KIND OF AGENT A BRANCH'S PLAN DECLARES — the selector nobody has to type.
1458
+ #
1459
+ # `--agent` was the only one, and an operator is the only thing that can type a
1460
+ # flag. The registry hands a queued slice to a free agent with no `--agent`
1461
+ # anywhere in the path, so an unattended fleet ran every slice as the same
1462
+ # undifferentiated worker. A plan that names the kind is what reaches a dispatch
1463
+ # nobody is watching.
1464
+ #
1465
+ # IT ANSWERS AND EXPORTS NOTHING. The caller decides, because the precedence is
1466
+ # the caller's: `--agent` already won over an inherited `PLOT_AGENT` and must win
1467
+ # over this too — a flag typed on this run is more specific than a field written
1468
+ # when the plan was drafted.
1469
+ #
1470
+ # THE PLAN DIRECTORY IS SEARCHED, NOT THE ACTIVE INDEX, and the candidates are
1471
+ # grepped before any is parsed — both `plot-open-pr.sh`'s rules, measured there:
1472
+ # a plan governing a branch may carry no symlink, and parsing 253 plans took
1473
+ # 103 s against 0.6 s for one `grep -lF` over the directory. A branch name is a
1474
+ # literal string, so a plan that does not contain it cannot name it.
1475
+ #
1476
+ # ABSENCE IS THE COMMON ANSWER AND NEVER AN ERROR. Every plan on the estate
1477
+ # names no kind, a `--restart` may run on a branch no plan names at all, and the
1478
+ # annotation is optional by design. All three print nothing and return 1.
1479
+ plan_declared_agent() { # $1 = branch → prints the declared kind, or nothing
1480
+ local branch="$1" plan_dir root cands f found
1481
+ [ -n "$branch" ] || return 1
1482
+ root=$(git rev-parse --show-toplevel 2>/dev/null) || return 1
1483
+ plan_dir=$("$script_dir/plot-config.sh" get "Plan directory" "docs/plans/")
1484
+ case "$plan_dir" in /*) ;; *) plan_dir="$root/$plan_dir" ;; esac
1485
+
1486
+ cands=$(grep -lF "$branch" "$plan_dir"*.md 2>/dev/null || true)
1487
+ for f in $cands; do
1488
+ [ -e "$f" ] || continue
1489
+ # The PARSER answers, never a grep of the plan file. A `grep agent:` would
1490
+ # read the marker out of prose documenting it and out of a DIFFERENT branch
1491
+ # line in the same plan — the annotation binds to one branch, and only the
1492
+ # parser knows which.
1493
+ found=$(bash "$script_dir/plot-plan-meta.sh" "$f" 2>/dev/null \
1494
+ | PLOT_WANT_BRANCH="$branch" node -e '
1495
+ let s = "";
1496
+ process.stdin.on("data", (d) => (s += d)).on("end", () => {
1497
+ let meta;
1498
+ try { meta = JSON.parse(s); } catch { return; }
1499
+ const want = process.env.PLOT_WANT_BRANCH;
1500
+ for (const wave of meta.waves ?? []) {
1501
+ for (const b of wave.branches ?? []) {
1502
+ // PRESENCE, not truthiness. The parser emits no `agent` key where
1503
+ // the plan declares none, which is every plan on the estate.
1504
+ if (b.branch === want && "agent" in b) { console.log(b.agent); return; }
1505
+ }
1506
+ }
1507
+ });
1508
+ ') || found=""
1509
+ if [ -n "$found" ]; then printf '%s\n' "$found"; return 0; fi
1510
+ done
1511
+ return 1
1512
+ }
1513
+
1514
+ # DOES THIS CLONE HOLD THE CHARTER? Answered so a MISSING one can be REPORTED.
1515
+ #
1516
+ # A DELIBERATE ASYMMETRY, and the plan settles it: `resolve_launch` refuses a
1517
+ # charter it cannot BELIEVE (malformed) and a harness not on PATH, and both
1518
+ # stay. A charter that simply does not EXIST is the adoption case — a plan
1519
+ # written where `reviewer` is declared, dispatched on a clone that declares
1520
+ # nothing — and refusing it would make that plan undispatchable on every such
1521
+ # clone. So the kind still travels, the launch falls back to the repo default,
1522
+ # and the run SAYS which name it looked for.
1523
+ charter_file_for() { # $1 = agent name → prints the path it would read
1524
+ local name="$1" root
1525
+ root=$(git rev-parse --show-toplevel 2>/dev/null) || return 1
1526
+ printf '%s/.plot/charters/%s.json\n' "$root" "$name"
1527
+ }
1528
+
1362
1529
  # `PLOT_DISPATCH_SOURCED=1` STOPS HERE, so a test can take `resolve_launch` and
1363
1530
  # `start_worker` without dispatching anything — `plot-worker-loop.sh` states
1364
1531
  # this idiom for `resolve_prompt_file`, and this is the same one applied to the
@@ -1737,62 +1904,70 @@ EOF
1737
1904
  esac
1738
1905
  fi
1739
1906
 
1740
- # THE COUNT IS THE RULE'S, and the rule is `packages/domain/src/rules/
1741
- # fleet-size.ts` — imported directly, the same shape `plot-reap.sh` uses for
1742
- # `reapable.ts`. Node 24 strips the types, so there is no build step between
1743
- # this script and the decision it asks for, and there is no second copy of the
1744
- # default, the subtraction or the machine's veto living in shell.
1907
+ # THE COUNT IS THE RULE'S, and the rule is asked through its BUNDLE —
1908
+ # `board/plot-fleet-size.mjs`, tracked in git beside the other 24. There is no
1909
+ # second copy of the default, the subtraction or the machine's veto living in
1910
+ # shell.
1911
+ #
1912
+ # A SOURCE IMPORT CANNOT REACH A PLUGIN INSTALL, and this block used to be
1913
+ # one. It imported `rules/fleet-size.ts` and `entities/machine.ts` as `file://`
1914
+ # sources. Node 24 strips types, so the TypeScript was never the obstacle — the
1915
+ # SECOND import is: `machine.ts` opens with `import { z } from 'zod'`, and an
1916
+ # install carrying no `node_modules` cannot resolve it. Measured 2026-09-17
1917
+ # against a copy with no `node_modules` on the path:
1745
1918
  #
1746
- # A RULE THAT CANNOT BE ASKED STARTS NOTHING AND SAYS SO. Missing node, a
1747
- # failed import, a module that throws all leave the answer empty. The
1748
- # direction is the reaper's: silence is never permission, and here permission
1749
- # would spawn detached processes.
1919
+ # machine.ts FAILED: Cannot find package 'zod'
1920
+ # fleet-size.ts: imported
1921
+ #
1922
+ # So the bundle carries BOTH rules with `zod` bundled in. `a-shell-script-asks
1923
+ # -the-domain` settled the shape: a bundle under `skills/plot/scripts/board/`
1924
+ # is how a shell script reaches a rule, and a skill's own script directory is
1925
+ # what a plugin ships.
1750
1926
  #
1751
1927
  # TWO MODULES, BECAUSE THE VERDICT AND THE COUNT ARE TWO RULES. `headroomFor`
1752
1928
  # owns what a fork cost MEANS and `fleetSize` owns what to do about it; the
1753
1929
  # count rule takes the verdict as a reading rather than deriving it, so the
1754
- # thresholds have exactly one home and this block is the join.
1755
- #
1756
- # IMPORTED AS `.ts` WITH NO `.js` REWRITING. Node 24 strips types but does not
1757
- # remap a relative specifier, so `fleet-size.ts` may only `import type` from
1758
- # its neighbours — which is why the verdict arrives as a value here rather
1759
- # than being computed inside the rule.
1760
- start_domain="$(cd "$script_dir/../../.." 2>/dev/null && pwd)/packages/domain/src"
1761
- start_rule="file://$start_domain/rules/fleet-size.ts"
1762
- start_answer=$(PLOT_REQUESTED="$start_count" PLOT_RUNNING="$start_running" \
1763
- PLOT_COST="$start_cost" PLOT_RULE="$start_rule" \
1764
- PLOT_MACHINE="file://$start_domain/entities/machine.ts" \
1765
- node --input-type=module - <<'NODE_EOF' 2>/dev/null
1766
- const { fleetSize, DEFAULT_FLEET_SIZE } = await import(process.env.PLOT_RULE);
1767
- const { headroomFor } = await import(process.env.PLOT_MACHINE);
1768
-
1769
- // AN ABSENT COUNT IS THE RULE'S DEFAULT, resolved here rather than in the
1770
- // shell: the number and the argument for it have one home.
1771
- const requested =
1772
- process.env.PLOT_REQUESTED === "" ? DEFAULT_FLEET_SIZE : Number(process.env.PLOT_REQUESTED);
1773
-
1774
- // An UNMEASURED cost is null, never zero: zero is the fastest fork there is and
1775
- // would read as the clearest possible machine.
1776
- const spawnCostMs = process.env.PLOT_COST === "" ? null : Number(process.env.PLOT_COST);
1777
-
1778
- const answer = fleetSize({
1779
- requested,
1780
- running: Number(process.env.PLOT_RUNNING),
1781
- spawnCostMs,
1782
- headroom: headroomFor(spawnCostMs),
1783
- });
1784
-
1785
- process.stdout.write(`${answer.start}\t${answer.headroom}\t${answer.shortfall}`);
1786
- NODE_EOF
1787
- )
1930
+ # thresholds have exactly one home and the bundle's entry is the join.
1931
+ #
1932
+ # A RULE THAT CANNOT BE ASKED STARTS NOTHING AND SAYS SO. A missing bundle, a
1933
+ # missing node, a module that throws all leave the answer empty. The direction
1934
+ # is the reaper's: silence is never permission, and here permission would spawn
1935
+ # detached processes.
1936
+ start_bundle="$script_dir/board/plot-fleet-size.mjs"
1937
+ start_answer=$(printf '%s\t%s\t%s' "$start_count" "$start_running" "$start_cost" \
1938
+ | node "$start_bundle" 2>/dev/null)
1788
1939
 
1789
1940
  if [ -z "$start_answer" ]; then
1941
+ # THE REFUSAL NAMES THE CONDITION THAT FAILED, never two that hold. The
1942
+ # message this replaced said *"it needs node 24 and a readable checkout of
1943
+ # packages/domain"* to an operator whose node was 24.4.1 and whose checkout
1944
+ # was readable — the import it could not resolve was named nowhere. A
1945
+ # refusal that names the wrong condition costs more than one that says
1946
+ # nothing, because it looks actionable. So the two causes are separated and
1947
+ # tested in the order that distinguishes them: an absent bundle is a broken
1948
+ # or partial installation, and a present bundle that answered nothing is the
1949
+ # runtime underneath it.
1790
1950
  echo "plot-dispatch: --start could not ask how many agents to start — starting none." >&2
1791
- echo " The rule is $start_rule" >&2
1792
- echo " It needs node 24 and a readable checkout of packages/domain." >&2
1951
+ if [ ! -f "$start_bundle" ]; then
1952
+ echo " The rule's bundle is missing: $start_bundle" >&2
1953
+ echo " Every bundle is tracked in git, so this is a broken or partial installation." >&2
1954
+ echo " In a development checkout, run 'pnpm build:board'." >&2
1955
+ else
1956
+ start_node_v="$(node --version 2>/dev/null)" || start_node_v=""
1957
+ echo " The rule's bundle is $start_bundle" >&2
1958
+ if [ -z "$start_node_v" ]; then
1959
+ echo " No usable 'node' was found on PATH. The bundle needs node 20 or newer." >&2
1960
+ else
1961
+ echo " The bundle is present but answered nothing under node $start_node_v." >&2
1962
+ echo " Run it directly to see why: printf '%s\\t%s\\t%s' '$start_count' '$start_running' '$start_cost' | node '$start_bundle'" >&2
1963
+ fi
1964
+ fi
1793
1965
  exit 1
1794
1966
  fi
1795
1967
 
1968
+ # THREE FIELDS, AND THE SENTENCE IS LAST. `start_why` is printed to the
1969
+ # operator below, so it travels; taking it as the whole remainder means a
1970
+ # shortfall can never be truncated by its own punctuation.
1796
1971
  start_n=${start_answer%%$'\t'*}
1797
1972
  start_rest=${start_answer#*$'\t'}
1798
1973
  start_headroom=${start_rest%%$'\t'*}
@@ -89,6 +89,9 @@
89
89
  # Output: per-plan wave report on stdout, terminated by a machine-countable
90
90
  # summary line:
91
91
  # summary: plans=1 waves=3 branches=5 claimed=1 eligible=2 blocked=1 deferred=1 waiting=1 prereq_missing=0 merge_detect=pr-merge host=ok main=main
92
+ # `host` is one of ok, partial, throttled, secondary, failed, unasked —
93
+ # `partial` means some of the host's states answered and some did not,
94
+ # so the PR readings below are incomplete rather than absent.
92
95
  # `blocked` counts WAVES an earlier wave holds; `waiting` and
93
96
  # `prereq_missing` count BRANCHES their `waits:` annotation holds.
94
97
  # merge_detect names how merged-and-deleted branches were detected:
@@ -631,6 +634,19 @@ PR_LIST_LIMIT="${PLOT_PR_LIST_LIMIT:-1000}"
631
634
  # secondary — a burst refusal (`plot-host.sh` exit 6). Nothing is broken
632
635
  # either, and it clears in seconds rather than minutes.
633
636
  # failed — any other failure (exit 3, or anything unclassified).
637
+ # partial — SOME of the host's states answered and some did not
638
+ # (`plot-host.sh` exit 7). The rows that arrived are real and
639
+ # are parsed; what is missing is a whole state, so the reading
640
+ # is incomplete rather than absent. Only Bitbucket can produce
641
+ # it: `bb pr list` has no `all` state, so the arm asks once per
642
+ # state, while GitHub takes `--state all` in one call.
643
+ #
644
+ # IT IS NOT `ok` AND IT IS NOT `failed`. Reporting `ok` would
645
+ # serve a page missing a state as a complete answer — #912, where
646
+ # nine branches read `commits, no PR ever opened` and two had
647
+ # live PRs. Reporting `failed` would throw away rows that
648
+ # arrived and make every branch `unknown`, which is not
649
+ # startable — the right refusal about the wrong thing.
634
650
  # unasked — no host to ask, or --offline WITHOUT `--next`. Not a
635
651
  # degradation: the scan was never going to ask, and saying
636
652
  # `failed` would report a fault where there is a configuration.
@@ -644,6 +660,43 @@ PR_LIST_LIMIT="${PLOT_PR_LIST_LIMIT:-1000}"
644
660
  # secondary limit waits minutes for a ceiling that cleared in seconds.
645
661
  HOST_VERDICT=unasked
646
662
 
663
+ # The remote branches this scan tracks, read once for every question that asks.
664
+ #
665
+ # MOVED UP FROM ITS OLD POSITION (#333) so the host call below can be told which
666
+ # branches to ask about. It is the same single `for-each-ref` it always was —
667
+ # see the commentary at its old site — and the reasons it exists are unchanged:
668
+ # `git show-ref --verify` was asked once per branch from two places, and
669
+ # `%(objectname)` rides along free for the commit walk.
670
+ REMOTE_REFS=$(git for-each-ref --format='%(refname:strip=3)%09%(objectname)' \
671
+ "refs/remotes/origin" </dev/null 2>/dev/null)
672
+
673
+ # The branch names alone, space-separated — what the sweep asks the host about.
674
+ #
675
+ # THE JOIN'S OWN KEYS, AND NOTHING WIDER. `prefill_pr_states` indexes the host's
676
+ # reply by branch and every row it cannot key is discarded, so the set this asks
677
+ # about is exactly the set that could ever be used. Measured 2026-09-20 on
678
+ # `quatico/quaweb-website`: 11 remote branches against 902 pull requests, of
679
+ # which a listing hands over 50 — the sweep asks 11 questions and gets 11
680
+ # answers, where the listing asked one and answered for 5%.
681
+ #
682
+ # `HEAD` IS DROPPED. `refs/remotes/origin/HEAD` is a symbolic ref naming the
683
+ # default branch, not a branch of its own; asking the host about a branch called
684
+ # `HEAD` spends a query to learn that nothing is named that.
685
+ #
686
+ # SPACE-SEPARATED, AND GIT IS WHAT MAKES THAT SAFE. Both this list and the
687
+ # adapter's `PR_LIST_BRANCHES` are read by an unquoted `for`, so a name carrying
688
+ # whitespace would split into two branches that do not exist. `git
689
+ # check-ref-format` REFUSES a ref name containing a space or a tab — verified
690
+ # 2026-09-20, both exit non-zero — so no such branch can reach this, and the
691
+ # separator is git's guarantee rather than a hopeful convention.
692
+ #
693
+ # EMPTY IS A REAL ANSWER AND IT DISABLES THE SWEEP. A checkout with no remote
694
+ # refs has no branches to ask about, and a sweep of nothing would state that
695
+ # every tracked branch answered — a completeness claim over an empty set, which
696
+ # would license `NONE` for branches nobody asked about. `prefill_pr_states`
697
+ # falls back to the listing there, which is what it has always done.
698
+ TRACKED_BRANCHES=$(printf '%s\n' "$REMOTE_REFS" | cut -f1 | grep -v '^HEAD$' | grep -v '^$' | tr '\n' ' ')
699
+
647
700
  prefill_pr_states() {
648
701
  [ "$HOST_LOOKUP_OK" = 1 ] || return 0
649
702
  [ -n "$HOST_STATE_CACHE" ] || return 0
@@ -672,10 +725,34 @@ prefill_pr_states() {
672
725
  # and /dev/null keeps the call working with the text simply unavailable.
673
726
  host_list_out="${HOST_STATE_CACHE:+$HOST_STATE_CACHE/pr-list.json}"
674
727
  host_list_out="${host_list_out:-/dev/null}"
728
+ # THE BRANCHES THIS SCAN TRACKS, HANDED TO THE HOST (#333). The adapter uses
729
+ # them only where it can — the Bitbucket arm sweeps its REST endpoint once per
730
+ # branch per state — and ignores them everywhere else, so the GitHub arm makes
731
+ # the single call it always made. Passing them unconditionally keeps one call
732
+ # shape here rather than a backend test this script has no business making.
733
+ #
734
+ # AN EMPTY SET PASSES NOTHING and the adapter lists as before. See
735
+ # `TRACKED_BRANCHES`: a completeness claim over an empty set would license
736
+ # `NONE` for branches nobody asked about.
737
+ _branch_args=()
738
+ for _tb in $TRACKED_BRANCHES; do _branch_args+=(--branch "$_tb"); done
675
739
  host_err=$("$script_dir/plot-host.sh" pr-list --state all --limit "$PR_LIST_LIMIT" --rich \
740
+ ${_branch_args[@]+"${_branch_args[@]}"} \
676
741
  </dev/null 2>&1 >"$host_list_out"); rc=$?
677
742
  js=$(cat "$host_list_out" 2>/dev/null)
678
- if [ "$rc" -ne 0 ]; then
743
+ # A PARTIAL ANSWER TAKES THE PARSE PATH AND STILL DEGRADES THE VERDICT, which
744
+ # is a control-flow change rather than another `case` arm below: every other
745
+ # non-zero rc sets a verdict and returns BEFORE `$js` is read, because there
746
+ # is nothing to read. Here there is — the rows of the states that answered are
747
+ # already in `host_list_out`, since stdout is redirected to a file and stderr
748
+ # captured separately.
749
+ #
750
+ # BOTH HALVES ARE REQUIRED. Falling through without setting the verdict would
751
+ # report `ok` over a page missing a whole state, which is #912; returning
752
+ # early would throw away rows the host did answer with.
753
+ if [ "$rc" -eq 7 ]; then
754
+ HOST_VERDICT=partial
755
+ elif [ "$rc" -ne 0 ]; then
679
756
  # THREE OUTCOMES, NOT TWO. `unasked` already means "the question was
680
757
  # never put" (see HOST_VERDICT above: *not a degradation, the scan was
681
758
  # never asking*), and a host that cannot be ASKED AT ALL belongs there
@@ -743,7 +820,12 @@ prefill_pr_states() {
743
820
  return 0
744
821
  fi
745
822
  # The list arrived. An empty one arrived too — that is the whole distinction.
746
- HOST_VERDICT=ok
823
+ #
824
+ # A PARTIAL VERDICT IS NOT OVERWRITTEN HERE. Exit 7 reaches this line
825
+ # deliberately, because its rows must be parsed; an unguarded `ok` would
826
+ # undo the one thing that distinguishes an incomplete page from a whole one
827
+ # and report #912 as a healthy reading.
828
+ [ "$HOST_VERDICT" = partial ] || HOST_VERDICT=ok
747
829
  # `pr-list` emits one compact JSON object per line. PARSED IN ONE PASS, and
748
830
  # that is a correctness-of-cost property rather than a style preference:
749
831
  # measured 2026-08-18 on this repo's 221 PRs, a `sed` per field per row —
@@ -864,9 +946,39 @@ EOF
864
946
  # A repository genuinely holding zero PRs loses nothing by being asked: it has
865
947
  # no branches with PRs for the join to serve either, so the cost is zero calls
866
948
  # in both readings.
867
- if [ "$_pr_rows" -gt 0 ] && [ "$_pr_rows" -lt "$PR_LIST_LIMIT" ] 2>/dev/null; then
868
- printf '1' > "$HOST_STATE_CACHE/.list-complete" 2>/dev/null || true
869
- fi
949
+ #
950
+ # A SWEEP STATES ITS COMPLETENESS; A PAGE ONLY EVER IMPLIED IT (#333). The
951
+ # test above reads completeness off one page's size, which is the only
952
+ # evidence a listing offers — and on Bitbucket it is evidence the listing
953
+ # cannot give at all, since `bb pr list` returns a fixed 50 whether or not
954
+ # more exist. Where the adapter swept per branch it says so on stderr, naming
955
+ # both counts, and that sentence is a stronger claim than any row count: every
956
+ # tracked branch was asked and each one answered.
957
+ #
958
+ # THE ROW COUNT IS NOT CONSULTED ON THAT PATH, and it must not be. A sweep
959
+ # over 11 branches of which 2 have pull requests emits 2 rows — a true and
960
+ # complete answer that `0 < rows < PR_LIST_LIMIT` would also accept, but for
961
+ # the wrong reason, and which a sweep of 0 matches would fail outright despite
962
+ # being equally complete. Reading the claim the adapter made is exact where
963
+ # re-deriving it from the output is a coincidence.
964
+ #
965
+ # THE WORDING IS A CONTRACT between this script and `plot-host.sh`'s
966
+ # `pr_sweep_report`, pinned on both sides. A partial sweep never prints it, so
967
+ # a match is licence and a miss is silence — never a guess.
968
+ #
969
+ # WHAT IS LOST BY GETTING THIS WRONG IS COST, NOT CORRECTNESS. Without the
970
+ # marker, `host_pr_state --ask` falls through to one `pr-state` call per
971
+ # unjoined branch and still answers correctly — the per-branch N+1 that #216
972
+ # removed. That is why withholding the marker is always the safe direction and
973
+ # is what every failure path here does.
974
+ case "$host_err" in
975
+ *"pr-list sweep complete"*)
976
+ printf '1' > "$HOST_STATE_CACHE/.list-complete" 2>/dev/null || true ;;
977
+ *)
978
+ if [ "$_pr_rows" -gt 0 ] && [ "$_pr_rows" -lt "$PR_LIST_LIMIT" ] 2>/dev/null; then
979
+ printf '1' > "$HOST_STATE_CACHE/.list-complete" 2>/dev/null || true
980
+ fi ;;
981
+ esac
870
982
  }
871
983
  prefill_pr_states
872
984
 
@@ -1487,8 +1599,12 @@ worktree_locked() { # $1=worktree path → 0 when a lock is held there
1487
1599
  # on the population that must stay free. The walk here is a subject/emptiness
1488
1600
  # question rather than a timestamp read, but the guard is deliberately broad and
1489
1601
  # loosening it to fit this change is how a guard rots.
1490
- REMOTE_REFS=$(git for-each-ref --format='%(refname:strip=3)%09%(objectname)' \
1491
- "refs/remotes/origin" </dev/null 2>/dev/null)
1602
+ # READ ABOVE `prefill_pr_states`, not here. The per-branch sweep (#333) hands
1603
+ # the host the branches this scan tracks, and that list is exactly what this
1604
+ # batch already answers — so the assignment moved up rather than a second
1605
+ # `for-each-ref` being added beside it. Everything documented above still
1606
+ # describes it; only the line's position changed, and it depends on nothing but
1607
+ # git, so nothing between the two points can read a different answer.
1492
1608
 
1493
1609
  # Whether `origin/$1` exists, answered from the batch rather than by spawning.
1494
1610
  #
@@ -4265,6 +4381,17 @@ elif [ "$HOST_VERDICT" = failed ]; then
4265
4381
  echo " branch below reads from local evidence alone, and a branch whose"
4266
4382
  echo " PR is unknown reads 'unknown' rather than 'open'. This is not a"
4267
4383
  echo " rate limit — waiting will not clear it; check the host and auth."
4384
+ # THE PAGE IS SHORT, NOT ABSENT, and that is a different instruction to a
4385
+ # reader. The three notes above all say *no PR could be read*; here some were,
4386
+ # so the branches below are a MIXTURE — a branch shown without a PR may have one
4387
+ # in the state that failed. Telling a reader to treat this as an outage would
4388
+ # discard the rows that arrived; telling them nothing is #912, where nine
4389
+ # branches read as having no PR and two had live ones.
4390
+ elif [ "$HOST_VERDICT" = partial ]; then
4391
+ echo " note: the git host answered for some states and not others, so the PR"
4392
+ echo " list below is INCOMPLETE. A branch shown without a PR may have one"
4393
+ echo " in the state that failed — do not read this as evidence that a"
4394
+ echo " branch is unreviewed. Re-run to get the whole list."
4268
4395
  fi
4269
4396
  # A STALE PULSE SAYS SO. The fetch used to fail silently, which made a scan of
4270
4397
  # hour-old refs read exactly like a scan of current ones — the same