fix(aggregations): translate the legacy DSL — batch 1 of the #1261 sweep

[rubenvdlinde]

Stacked on #1287. Base is feat/glline-fiscal-year, whose baseline changes this builds on. Retarget to development once #1287 merges.

Forty-one aggregations move from keys AggregationRunner never reads to keys it does. None of them errored before — they returned an empty result, or nothing at all, under HTTP 200.

19 @self correlations → groupBy dimensions

filter: { poolId: "@self.id" } on an intra-schema aggregation does not resolve. PlaceholderResolver acts only on $-prefixed values, so @self.id stayed a literal string and matched no row — an empty result that reads as "no data" rather than "this never worked".

And it could not have worked: no caller supplies a parent row. AggregationController, ReportRenderService and ThresholdEvaluationService are the only three call sites of run() in OpenRegister, and none passes one.

Grouping by the same field needs no parent row and answers for every record at once; a caller narrows to one through the extraFilter query params the REST endpoint already passes through.

22 operations maps → metrics

operations is not an engine key either. Each entry is {field, operation, target} — exactly {field, metric, as} — so the translation is mechanical. target duplicated the map key; both became as. Qualified fields (RetainerDrawdown.drawdownAmount) are stripped to bare names, since from already resolves them. Conditional entries carry through as condition, which the runner honours since openregister #2917.

Deliberately NOT translated

• 33 operations maps containing an expression op. The engine has no equivalent, and their conditions are SQL-ish strings ("VATLine.type = 'collected'") where computeMetrics() takes a filter object. Guessing either would produce a confident wrong number — the exact failure mode this sweep exists to remove. BtwAangifte.totalsByReturn is pinned in its test as still-untranslated so the gap stays visible.
• ExpenseClaimEntry.settlementTotals. It uses a multi-source sources array whose amountField/customerField are config keys of a source descriptor, not fields. An earlier pass of this change translated them into metrics; that was nonsense and is reverted.

Two defects the gates caught mid-change

Moving those field names into a key the bare-reference gate can see pushed the count 102 → 111. That is not new debt — it is debt that was previously invisible:

ProvisionDisclosureTabel.provisionDisclosureGeneration	kept source instead of from, so its five movement fields resolved against the wrong schema
IBProfitAttribution.innovatieboxAdministratie	summed kwalificerende_winst_na_nexus; the schema declares qualifying_profit_after_nexus — a Dutch/English mismatch that matched nothing
GhgInventory.scope3ByCategory	tripped the string-groupBy gate the moment it gained metrics; a string groupBy is silently dropped and yields one ungrouped total

All three fixed here, so the count returns to 102 with no baseline raised.

Ratchets

AGG_PLACEHOLDER_BASELINE   81 -> 62    (exactly the 19 converted)
AGG_NO_METRIC_BASELINE    206 -> 185
AGG_BARE_REF_BASELINE     102 -> 102   (unchanged, after the three fixes)

Verification

• Parsed-tree diff against HEAD: exactly the intended aggregations changed, zero non-aggregation leaf changes across all 14 files
• Only files whose JSON round-trips byte-identically were rewritten, so nothing was silently reformatted
• Five fragment tests updated from the old vocabulary to the engine's — each asserting the inert key is gone, not merely that the new one is present
• Full suite 4990 tests, 0 failures
• validate-registers exits 0 with every gate at its baseline

Remaining after this batch

28 of the original 55 @self aggregations still need remodelling (missing fields, recursiveOver, @descendant, string expressions like "@self.periodYear-@self.periodQuarter"). Those are judgement calls per aggregation, not a mechanical pass, and each needs checking against real rows.

Refs #1261

[github-actions]

Quality Report — ConductionNL/shillinq @ b2d1820

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
check-manifest		✅
check-manifest-shell		✅
format		✅
check-l10n-js		✅
check-schema-l10n		✅
check-manifest-budget		✅
check-markers		✅
check-registers		✅
check-seeds		✅
check-fragment-required		✅
check-nav-reachability		✅
check-job-registration		✅
test-l10n		✅
composer			✅	✅ 106/106
npm			✅	✅ 551/551
app:check-code	⏭️
info.xml	✅
REUSE	❌
PHPUnit					✅
Newman					✅
Playwright					✅
Hydra gates					✅

---

Quality workflow — 2026-08-27 20:39 UTC

Download the full PDF report from the workflow artifacts.