fix(submissions): t-paliad-365 picker quality — re-kind 8 court-act rules (mig 164) + group-header contrast

P1: 8 anchor/trigger rows (Zustellung des Urteils / Veröffentlichung der
Erteilung) were mislabeled event_type='filing'+primary_party='both' and leaked
into the /submissions/new draftable-submission picker. Migration 164 re-kinds
them to event_type='decision'+primary_party='court', aligning them with the 16
sibling court-act rows the model already has; the picker's event_type='filing'
filter then excludes them. Defensive guard added to loadSubmissionCatalog
(primary_party IS DISTINCT FROM 'court') as belt-and-braces against future drift.
Safety: the only event_type coupling for these rows is ruleAnchorKind
(projection_service.go) — they now anchor as appointments (correct, sibling-
consistent); deadline computation keys off is_court_set not event_type, and the
child sequence guard parentHasAnchoredActual UNIONs both anchor tables, so no
chain breaks. Verified: catalog 113->105, 8 court acts gone, 48 'both' party
submissions (incl. UPC appeal briefs) retained.

P2: the grouped picker table already renders a correct colspan group-header row
(a911a2d); the defect was contrast — the band used --color-bg-subtle, the SAME
token as the thead, so groups read as undelimited floating text. New
--color-bg-group-header token (cool/deeper in light, raised-cream in dark) +
heavier top divider + neutral left accent + darker label make each proceeding a
distinct section. Live deployed bundle confirmed current (not stale).

frontend/src/styles/global.css

@@ -26,6 +26,7 @@
        theme swap for free. */
     --color-bg:           var(--hlc-cream);
     --color-bg-subtle:    #f7f3f0;        /* slightly off-cream for table headers / soft surfaces */
+    --color-bg-group-header: rgb(var(--hlc-midnight-rgb) / 0.055); /* cool band for grouped-table section headers — deeper + cooler than the warm-cream thead so groups read as distinct sections (t-paliad-365 P2) */
     --color-bg-lime-tint: rgb(var(--hlc-lime-rgb) / 0.10);
     --color-surface:      #ffffff;        /* cards stay white for contrast */
     --color-surface-2:    #fafafa;        /* slightly raised surface (alt rows, code blocks) */
@@ -180,6 +181,7 @@
 :root[data-theme="dark"] {
     --color-bg:           var(--hlc-midnight);
     --color-bg-subtle:    #00304a;        /* slightly raised for table headers */
+    --color-bg-group-header: rgb(var(--hlc-cream-rgb) / 0.08); /* raised cream band over midnight — distinct from the thead's #00304a (t-paliad-365 P2) */
     --color-bg-lime-tint: rgb(var(--hlc-lime-rgb) / 0.12);
     --color-surface:      #0a3047;        /* card surface — distinct from body */
     --color-surface-2:    #0d3a55;        /* one step further raised */
@@ -6064,16 +6066,24 @@ dialog.modal::backdrop {
    proceeding with a lime border so the lawyer sees their primary
    context at a glance. */
 .entity-table-group-header th {
-    padding: 0.65rem 1rem;
+    padding: 0.85rem 1rem 0.7rem;
     text-align: left;
     font-size: 0.75rem;
     font-weight: 600;
     text-transform: uppercase;
     letter-spacing: 0.04em;
-    color: var(--color-text-muted);
-    background: var(--color-bg-subtle);
-    border-top: 1px solid var(--color-border);
+    color: var(--color-text);
+    /* t-paliad-365 P2: the group-header band used --color-bg-subtle, the SAME
+       token as the thead, so over the cream page each proceeding header read as
+       undelimited floating text and the columns looked broken. A distinct,
+       cooler/deeper band + a heavier top rule + a neutral left accent make each
+       proceeding read as its own clearly-separated section. The --own modifier
+       (lime accent, below) still overrides this to mark the project's own
+       proceeding. */
+    background: var(--color-bg-group-header);
+    border-top: 2px solid var(--color-border-strong);
     border-bottom: 1px solid var(--color-border);
+    border-left: 3px solid var(--color-text-muted);
 }
 
 .entity-table tbody tr.entity-table-group-header,

internal/db/migrations/164_rekind_court_act_rules.down.sql

@@ -0,0 +1,20 @@
+-- Down for 164_rekind_court_act_rules — restore the pre-fix mislabel
+-- (event_type='filing', primary_party='both') on the 8 court/trigger rows.
+-- Scoped to the re-kinded state so it only reverses this migration's effect.
+
+UPDATE paliad.deadline_rules_unified
+   SET event_type    = 'filing',
+       primary_party = 'both',
+       updated_at    = now()
+ WHERE submission_code IN (
+         'de.inf.bgh.urteil_olg',
+         'de.inf.olg.urteil_lg',
+         'de.null.bgh.urteil_bpatg',
+         'dpma.appeal.bgh.entsch_bpatg',
+         'dpma.appeal.bpatg.entscheidung',
+         'epa.opp.boa.entsch',
+         'dpma.opp.dpma.publish',
+         'epa.opp.opd.grant'
+       )
+   AND event_type = 'decision'
+   AND primary_party = 'court';

internal/db/migrations/164_rekind_court_act_rules.up.sql

@@ -0,0 +1,64 @@
+-- 164_rekind_court_act_rules — t-paliad-365 (P1), diagnosis t-paliad-363.
+--
+-- Re-kind 8 anchor/trigger rows in paliad.deadline_rules_unified that are
+-- MISLABELED event_type='filing' + primary_party='both'. They are COURT /
+-- registry acts — the service of a lower-court judgment ("Zustellung … Urteil"
+-- / "Zustellung … Entscheidung") or the publication of grant ("Veröffentlichung
+-- der Erteilung") — that START a deadline chain. They are NOT party
+-- submissions, yet they surfaced in the global /submissions/new picker, whose
+-- catalog query filters event_type='filing' (loadSubmissionCatalog,
+-- internal/handlers/submissions.go). m noticed "Service of OLG Judgment"
+-- (de.inf.bgh.urteil_olg) listed as draftable.
+--
+-- The model already carries a correct court-act tier: 16 sibling rows with
+-- event_type='decision' + primary_party='court' (de.inf.olg.urteil_olg,
+-- upc.inf.cfi.decision, …). This migration aligns the 8 mislabeled rows with
+-- that tier. The picker filter (event_type='filing') then excludes them
+-- automatically — no handler change is required for the core fix (a defensive
+-- primary_party guard is added in submissions.go as belt-and-braces).
+--
+-- m's fork LOCKED to data-correction (t-paliad-363 §P1, fork i).
+--
+-- submission_code is KEPT (the 16 sibling decision rows keep theirs as stable
+-- anchor identifiers; the deadline chain links via parent_id (uuid FK), which
+-- this migration does not touch — so anchoring is unaffected).
+--
+-- ── SAFETY (grep of the deadline engine, t-paliad-365 Step 1) ──────────────
+-- The ONLY code coupled to event_type for these rows is ruleAnchorKind
+-- (projection_service.go:1825): event_type IN ('hearing','decision','order')
+-- → the rule anchors as a paliad.appointments row; everything else anchors as
+-- a paliad.deadlines row. After this migration, clicking "Datum setzen" on one
+-- of these 8 court/trigger events records it as an APPOINTMENT (milestone)
+-- instead of a deadline — which is the semantically correct, sibling-consistent
+-- behaviour.
+--   * Deadline COMPUTATION is NOT affected: the forward projection
+--     (computeProjections) keys off the is_court_set boolean, never event_type.
+--   * Child chains are NOT broken: the sequence guard parentHasAnchoredActual
+--     (projection_service.go:1803) UNIONs paliad.deadlines AND
+--     paliad.appointments, so a child (e.g. Revisionsfrist) still finds its
+--     re-kinded parent's anchor regardless of which table it lands in.
+--   * Existing data: exactly 1 paliad.deadlines row is already anchored to one
+--     of these 8 rules (0 appointments) as of 2026-06-01. It stays valid and
+--     is still found by the dual-table guard. The only cosmetic effect is that
+--     a future RE-anchor of that same event would write an appointment beside
+--     the old deadline; benign, and only if re-anchored.
+-- Verdict: re-kinding does not break deadline computation. Surfaced to head.
+--
+-- ADDITIVE / data-only. No schema changes. Reversible (see .down.sql).
+-- Idempotent: scoped to the mislabeled state (event_type='filing').
+
+UPDATE paliad.deadline_rules_unified
+   SET event_type    = 'decision',
+       primary_party = 'court',
+       updated_at    = now()
+ WHERE submission_code IN (
+         'de.inf.bgh.urteil_olg',        -- Zustellung OLG-Urteil
+         'de.inf.olg.urteil_lg',         -- Zustellung LG-Urteil
+         'de.null.bgh.urteil_bpatg',     -- Zustellung BPatG-Urteil
+         'dpma.appeal.bgh.entsch_bpatg', -- Zustellung BPatG-Entscheidung
+         'dpma.appeal.bpatg.entscheidung',-- Zustellung DPMA-Entscheidung
+         'epa.opp.boa.entsch',           -- Zustellung der Beschwerdeentscheidung
+         'dpma.opp.dpma.publish',        -- Veröffentlichung der Erteilung (DPMA)
+         'epa.opp.opd.grant'             -- Veröffentlichung der Erteilung (EPA)
+       )
+   AND event_type = 'filing';

internal/handlers/submissions.go

@@ -205,6 +205,12 @@ func loadSubmissionCatalog(ctx context.Context, projectProceedingTypeID *int) ([
 		  WHERE dr.is_active = true
 		    AND dr.lifecycle_state = 'published'
 		    AND dr.event_type = 'filing'
+		    -- Defensive guard (t-paliad-365 P1): even if a court/trigger act is
+		    -- ever mis-kinded event_type='filing' again, primary_party='court'
+		    -- keeps it out of the draftable-submission picker. mig 164 re-kinded
+		    -- the 8 known offenders to event_type='decision'+primary_party='court';
+		    -- this is belt-and-braces against future drift.
+		    AND dr.primary_party IS DISTINCT FROM 'court'
 		    AND dr.submission_code IS NOT NULL
 		    AND dr.submission_code <> ''
 		    AND pt.is_active = true