Every signal we publish is measured with walk-forward backtests — replayed through history with zero look-ahead, judged on data the system never saw. The failures are listed too. Updated weekly, automatically.
🧭 Morning Compass
VALIDATED — LIVE
Out-of-sample window514 days (2024-05-08 → 2026-06-11)
Next-day rank IC+0.0337 (t-stat 2.32)
Top-3 vs bottom-3 sectors, next day+0.115%/day · 54% win
On big overnight moves+0.238%/day
On quiet nights≈ 0 (correctly silent)
Overnight global moves (NASDAQ, crude, USDINR…) translated into expected NSE sector tilts via rolling 250-day sensitivities. Signal scales with overnight magnitude — the signature of a structural effect, not curve-fitting.
See today's compass →
🔥 Sector Ignition Scanner
TOP PHASE VALIDATED — LIVE
Replayed history11,837 sector-days (2023-05-30 → 2026-06-05)
IGNITING → 2-day alpha+0.162% (n=893)
IGNITING → 3-day alpha+0.256%
IGNITING hit rate57%
Baseline (all sector-days)+0.055% · 51%
WARMING phase+0.049% — no edge, context only
Futures OI Pressure overlay+0.486%/2d spread · 67% win (151d)
Point-in-time replay: each historical day computed exactly as the scanner would have seen it that evening. The IGNITING claim validates at ~3× baseline; we disclose that WARMING does not. The OI Pressure overlay (sector futures long/short buildup) is validated separately and shown on each card without altering the phase score.
See live ignition scores →
⚖️ Sector Ranking Model
GATED — NOT SHIPPED
Candidate models tested7 (3/5/10-day horizons)
Promotion gatespread ≥ 0.3%/5d · win ≥ 53% · IC ≥ 0.03
ResultAll rejected — no robust edge found
StatusAuto-retries weekly with fresh data
EOD price/RRG/delivery features cannot reliably rank sectors 3-10 days ahead — so we refuse to ship that model. If a trending-rotation regime returns and a candidate passes the gate out-of-sample, it promotes automatically and its numbers appear here.
🤖 AI Stock Picks
RECORDING — DAY 3 OF 30
Lists being recorded nightlyAvoid list (short), F&O high-conf bullish, Top-10 Picks (long), F&O high-conf bearish
Recording since2026-06-10
Why restartedLegacy archive never stored per-call conviction
Every call the site displays is now archived before its outcome exists and scored vs NIFTY at 1/3/5 days. Verified accuracy appears here automatically once a list has 30 recorded days and 100 scored outcomes — no number gets published before it is earned.
See today's picks →
Questions
Why publish failures?
Because a scoreboard that only shows wins is marketing, not measurement. Knowing that WARMING phases carry no edge, or that 5-day sector ranking isn't reliably predictable from EOD data, is exactly what keeps you from over-trading weak signals.
Why are these numbers small — fractions of a percent?
They are per-day or per-event edges across hundreds of observations, not single-trade returns. A +0.16% average 2-day edge over 893 events with a 57% hit rate is a real, compounding statistical advantage — anything promising "5% per trade" should worry you.
Will these numbers change?
Yes — they re-validate weekly on fresh data and we show the current figures, whatever they are. That is the point of a live scoreboard.
Page generated 2026-06-13 13:15 IST. SpikeDesk is not SEBI-registered; nothing here is investment advice. Backtested results do not guarantee future performance.