Stretched

Scans every cross-asset input — vol surfaces, credit spreads, drawdowns, valuation ratios — and ranks them by how far the latest reading sits from its own recent history. The further from the median, the less cushion if the tape turns. It does not predict; it surfaces the single loudest level extremity in the book right now.

A pick at z = +2.5 means today's reading is two-and-a-half deviations richer than its rolling norm — a level extreme, not a positioning one. Positioning, sentiment and macro inputs are excluded by design (those belong to Crowded). Calm-regime valuation stretch (CAPE flat-but-high for years) is invisible to a percentile axis — a known blind spot.

• FRED HY-spread truncation — ICE BofA series start 2023; mitigated by the HYG/LQD-derived hy_oas_proxy but full coverage isn't possible.
• Moderate-z saturation — VIX=26 at SVB detrends to z≈1.2; engine correctly calls it 'elevated, not extreme'. Lowering the gate further was inert.
• Calm-regime valuation blind spot — percentile-rank scoring can't see a ratio that's stretched but flat for years.

Drives the glance anchor (How rich are equities?) and owns the Extremes leaderboard — the one dynamic chapter that re-ranks week to week. In the commentary it's a SUPPORTING voice: it supplies the specific input name ("VIX term slope at a multi-year high") when its confidence gate fires, and stays quiet when the analyst-canonical signal isn't loud.

Graded on the Tier-1 regime-test harness — curated macro events × 18-yr backfill × recall@K. Verdict PARTIAL (33%).