fix(adr-084): the two AmendmentOrderService call sites #506 did not reach

[rubenvdlinde]

fix(adr-084): the two AmendmentOrderService call sites #506 did not reach

This PR was reduced on 2026-08-16 by S29. It originally carried S19's repair
of eleven ADR-084 call sites. While S19 was killed by the session crash,
decidesk#506 fix/mcp-chain-lost-the-object-service landed the same repair
and reached development first, which is what made this branch
mergeable_state=dirty.

Their version was taken wholesale. Every file this branch touched has been
reset to origin/development — lib/Mcp/* (all five), PublicationService,
TranscriptionService, VoteCastingService, MotionForwardingService,
MotionService, ActionItemWriter and psalm.xml. None of this branch's
version of any of them survives. Two equivalent fixes are not worth a conflict,
and this is the fourth duplicate-work collision on the board today.

What #506 did NOT reach — verified, not assumed

Before dropping the branch as redundant I measured origin/development
(d1957ca1, the merge itself) with each tool's own command. It is not clean:

lib/Service/VotingRoundOpener.php:77   Missing parameter $objectService (ObjectServiceInterface)
                                       in call to AmendmentOrderService constructor
lib/Service/VotingRoundOpener.php:78   Undefined variable: $container
lib/Service/VotingRoundOpener.php:78   Unknown parameter $container in call to
                                       AmendmentOrderService constructor
lib/Service/VotingRoundResults.php:86  … the same three …

$container is neither a parameter of those two constructors — ADR-084 replaced
it with ObjectServiceInterface — nor an argument AmendmentOrderService
declares. Constructing either class raises Undefined variable: $container at
request time, so opening a voting round and computing a tally both answer 500.
That is the backend fault behind the ten extra E2E failures S19 traced to
publication-workflow and voting-quorum-workflow (the skip count was 58 on
both sides, which is what made those ten a regression rather than a shuffle).

This PR is now two files, +19/−3, and nothing else.

Before / after — origin/development vs this branch

Same vendor/, same php:8.3-cli container, each tool's own composer script,
207 lib/ files on both trees:

check	origin/development d1957ca1	this branch
phpstan	6 errors (rc 1)	[OK] No errors (rc 0)
psalm	4 errors (rc 2)	No errors found! (rc 0)
phpmd	11 findings	7 findings

Two independent analysers agree on the same two files and nothing else, and both
go to zero. The four phpmd findings that disappear are exactly the
UndefinedVariable/UnusedLocalVariable $container pair in each of the two
files. The 7 that remain are pre-existing CouplingBetweenObjects = 13 on
MeetingService, MinutesDocumentService, MotionService,
ParticipationPublicationService, TranscriptionService, VoteCastingService
and VotingRoundCloser — they fail on development too and are not touched
here, so PHP Quality (phpmd) stays red on both sides, strictly improved.

Tests — identical on both trees, by identifier and not by count

Unit Tests, origin/development : Tests: 961, Assertions: 3339, Errors: 187, Failures: 36, Warnings: 17, Skipped: 25
Unit Tests, this branch        : Tests: 961, Assertions: 3339, Errors: 187, Failures: 36, Warnings: 17, Skipped: 25

The 183 unique failing-test identifiers diff clean between the two trees, so
this change neither fixes nor breaks a test. The 187+36 are the separate ADR-084
test-rewiring debt R2 catalogued (hollow createMock(ObjectServiceInterface)
doubles replacing store-backed fakes), which is its own slot and is deliberately
out of scope here.

Deliberately not done

• The ~223 broken unit tests. Patching constructor arguments there turns
  ArgumentCountErrors into assertion failures, not greens — it is a rewiring
  job, not an arity sweep.
• The five remaining CouplingBetweenObjects = 13 classes. Each is refactor(deps): type-hint OpenRegister's published contract (ADR-084) #495 adding
  one type to a class already at 12; each needs its own DI pass.
• Integration Tests (Newman), red on development since refactor(deps): type-hint OpenRegister's published contract (ADR-084) #495.
• No baseline entry, no suppression and no gate exclusion anywhere.

---

[S29] Re-measured in CI — zero introduced, three jobs fixed, eight E2E tests fixed

PR run 31948665299 (36 jobs) vs base run 31944146603 (d1957ca1, 36 jobs):

jobs failing on BASE but not on PR:
  quality / PHP Quality (phpstan)
  quality / PHP Quality (psalm)
  quality / Integration Tests (Newman)     ← now `success`

jobs failing on PR but not on BASE:
  (none)

base failures 13  →  PR failures 10

The local measurement reproduces exactly in CI: phpstan and psalm are both
success on this PR and both failure on the base.

PHP Quality (phpmd) stays red on both sides, strictly improved — CI names
11 findings at base and 7 here, and the four that disappear are exactly the
UndefinedVariable / UnusedLocalVariable $container pair in each of the two
files. The 7 that remain are pre-existing CouplingBetweenObjects = 13.

PHPUnit is byte-identical, cell for cell:
Tests: 969, Assertions: 3366, Errors: 177, Failures: 37, Warnings: 17, Skipped: 33
on both the PR and the base.

E2E — 11 failures → 3, and the eight that go green are the two repaired paths

BASE : 11 failed / 58 skipped / 109 passed
PR   :  3 failed / 58 skipped / 117 passed

Identical skip count (58), so this is a real delta and not a skip shuffle.
Diffing the failing-test identifiers rather than the counts: nothing is
introduced, and the eight that go green are

open voting round succeeds for a seeded meeting chair (guard resolves the chair)
open voting round is blocked (403) when caller is not a meeting chair/secretary
casting a vote returns the persisted vote (no return-type 500)
tally math — majority for → adopted
tally math — majority against → rejected
tally math — equal for/against, default tie-break → rejected
tally math — equal for/against, tieBreakRule=revote → tied
show-of-hands tally math — for=5 against=2 abstain=1 → adopted

Every one of them walks VotingRoundOpener::__construct or
VotingRoundResults::__construct — the two constructors this PR repairs. That
is causal, not correlational: the fix is two call sites, and the tests that
recover are exactly the ones that construct those two classes.

Gates

BASE run 31944146603 : FAIL(2)  gate-7 (12), gate-19 (983)
PR   run 31948665299 : FAIL(1)  gate-7 (12)
INTRODUCED: (none)

gate-19's disappearance is not claimed — it is diff-scoped, and this branch's
diff is now two files, so the two-dot artefact is working in our favour.

[github-actions]

Quality Report — ConductionNL/decidesk @ aa9f4e4

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	❌
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
check-manifest		✅
test-l10n		✅
format		✅
composer			✅	✅ 104/104
npm			✅	✅ 532/532
app:check-code	⏭️
info.xml	✅
REUSE	❌
PHPUnit					❌
Newman					✅
Playwright					❌
Hydra gates					❌

---

Quality workflow — 2026-08-16 11:51 UTC

Download the full PDF report from the workflow artifacts.

[github-actions]

Quality Report — ConductionNL/decidesk @ f6529a9

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	❌
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
check-manifest		✅
test-l10n		✅
format		✅
composer			✅	✅ 104/104
npm			✅	✅ 532/532
app:check-code	⏭️
info.xml	✅
REUSE	❌
PHPUnit					❌
Newman					✅
Playwright					❌
Hydra gates					❌

---

Quality workflow — 2026-08-16 13:39 UTC

Download the full PDF report from the workflow artifacts.