Phase 6: theme trend-scan (B) + competitor gap research (A)

- theme_scan.py: tags entries by 4 practitioner themes (tool-call/context/
  compute/trust), counts NEW arrivals per cron cycle (falsification check for
  one-day-cluster vs trend). Idempotent: re-run = 0 new.
- schema.sql: theme_tags table (separate from core entries schema)
- oracle-pipeline.sh: wire theme_scan after summarize
- A result (footnote): unified discipline layer unoccupied; adjacent OSS
  entrants exist (agentgateway, lelu) but no portable unified layer.

Verified: theme_scan classifies 5 seed/extra items, 2nd run = 0 new.

schema.sql

@@ -31,3 +31,16 @@ CREATE TABLE IF NOT EXISTS run_log (
     sources_failed TEXT,   -- JSON list of sources that errored/skipped
     notes TEXT
 );
+
+-- Theme tags: Phase 6 trend-tracking. Tags entries by the 4 practitioner
+-- resource-discipline themes so we can measure RECURRING theme frequency
+-- across FRESH entries (not persistence of specific rows). Counts new
+-- arrivals per cron cycle -> the falsification check for the "one-day cluster
+-- vs real trend" question. Separate table, never mutates the core entries schema.
+CREATE TABLE IF NOT EXISTS theme_tags (
+    entry_id INTEGER NOT NULL,
+    theme TEXT NOT NULL,
+    first_seen_cycle TEXT DEFAULT (strftime('%Y-%m-%dT%H:%M:%SZ', 'now')),
+    PRIMARY KEY (entry_id, theme),
+    FOREIGN KEY (entry_id) REFERENCES entries(id)
+);