The full breakdown behind today's tide. Updated every 5 minutes.
📰 Smart headline
auto-generated from historical analogs
Loading headline…
🧪 Macro components
the four readings driving the regime state
Loading components…
⏱️ Lag tracker
how late Tideline reacts to actual SPY drawdowns
Loading lag stats…
Each row = one SPY drawdown ≥3% since 1997. Warn-lag = trading days from peak to first ELEVATED/STRESS state. Drawdown caught = how much of the fall happened before the warning.
Loading drawdown episodes…
📓 Research record · what didn't ship
two predictive states tested in 2026, both rejected before launch
Tideline ships exactly one predictive claim — the CAUTION rule. Two more were tested in 2026. Both failed.
Killed before backtest
CRACKING — early macro weakness signal
External review (GPT-5 Thinking) identified it as a statistical paradox.
The rule would need to fire under calm-regime pre-conditions (NORMAL/EASY)
yet predict structural deterioration — but the moment structural deterioration
arrives, the regime panel definitionally flips to ELEVATED/STRESS. The 4-state
regime panel is the cracking signal.
Failed pre-registered backtest
DIP_BUY v2 — uptrend dip with credit + vol filters
Pre-registered spec: SPY ≥5% off 20-day high AND BAA10Y 20-day change ≤ 0
AND VIX/VIX3M ≤ 1.0, all under Faber=GREEN + regime ∈ {NORMAL, EASY}.
Pre-committed: ~45 fires training, ~30 OOS, +8.5pp H10 edge.
Result: 1 fire training, 2 fires OOS. Triple filter too strict — when SPY falls ≥5% in calm uptrends, credit and vol-curve almost always tighten with it.
Pre-registered FAIL action triggered: kill the rule, ship UX-only path with louder ELEVATED/STRESS visuals (cracked sand on the beach scene). Old code retained as graveyard artifact per integrity protocol.
See the verbatim entries (Entry 12, Entry 13) in the
research log.
Failed tests are never deleted.