Every signal, what it measures and how reliable it is.
Statuses: ✅ validated out of sample · 🟡 measured once: gap observed in the past, without correction for the number of conditions tested (part of it is luck) · ⚪ in research, not proven · ❌ tested without edge. Only 🟡 and ✅ appear on the Today screen.
See the full lab catalogue (358 conditions) →
Decision · 3
10-session low with options panic
❌ SETUP−0.18R per episode on average (90% CI −0.28 to −0.07) · 729 episodes 2008-2026
SPY, QQQ, VOO, IVV · entry at the next open · exit 3R or 10 sessions · costs 2 bp a side · median −1.03R · stop hit 72% of the time · worst streak 20 losses · Negative mean result of −0.18R per episode, below an arbitrary entry (−0.07R): does worse than a random entry on both halves. · Measured with the VIX in place of the real rule's options volatility (known only since 09/2024): an approximation of the rule, not the exact rule. The interval is computed on the independent signal runs, far fewer than the episodes.
Entry at the open after the signal · stop in session below the low · 3R target or exit at session 10. Not proven at this entry: shown as « watch », never as a trade.
10-session low with options panic, Nasdaq also at its low
⚪ SETUP−0.18R per episode on average (90% CI −0.29 to −0.07) · 658 episodes 2008-2026
SPY, QQQ, VOO, IVV · entry at the next open · exit 3R or 10 sessions · costs 2 bp a side · median −1.03R · stop hit 72% of the time · worst streak 23 losses · Negative mean result of −0.18R per episode, below an arbitrary entry (−0.07R): no proven edge over a random entry. · Measured with the VIX in place of the real rule's options volatility (known only since 09/2024): an approximation of the rule, not the exact rule. The interval is computed on the independent signal runs, far fewer than the episodes.
Entry at the open after the signal · stop in session below the low · 3R target or exit at session 10. Not proven at this entry: shown as « watch », never as a trade.
Index low not confirmed by small caps or semiconductors
⚪ SETUP−0.31R per episode on average (90% CI −1.16 to +0.26) · 255 episodes 2011-2026
SPY, QQQ, VOO, IVV · entry at the next open · exit 3R or 10 sessions · costs 2 bp a side · median −1.03R · stop hit 64% of the time · worst streak 14 losses · Negative mean result of −0.31R per episode, below an arbitrary entry (−0.07R): no proven edge over a random entry. · Measured with the VIX in place of the real rule's options volatility (known only since 09/2024): an approximation of the rule, not the exact rule. The interval is computed on the independent signal runs, far fewer than the episodes.
Entry at the open after the signal · stop in session below the low · 3R target or exit at session 10. Not proven at this entry: shown as « watch », never as a trade.
Measured context · 24
unusual off-exchange volume
🟡dark volume anomaly (FINRA, z30 and z90 ≥ 2) in a buy location: 10-day low closed in the top of the range, discount 0.5-0.786, trend pullback (> MA200 + discount); entry the next day
Measure and method (lab notes, reference F19)
🟡 (113 assets 2022-2026, vs the same location without anomaly: absorbed low +0.29R vs −0.05R, 439 trades; discount +0.11R vs −0.12R, 648; trend pullback +0.52R vs −0.04R, 78; positive gain on the 2 halves; 6 🟡 out of 26 tests, 1-2 expected by chance). 2026 check (19 Tempest assets): 50% of F19 days are S10b days; F19 outside S10b −0.34R (n 11) → not a standalone agent, S10b size bonus: S10b with F19 +1.46R vs +0.29R without (n 11 / 57, to confirm) · Confirmed on 600 assets (19/10, : discount +0.18R, deep discount +0.19R, absorbed low +0.19R, trend pullback +0.40R, the 2 halves positive. Daily radar: liquid universe, score min(z30, z90) of total dark volume · regimes 21/10/2026 : positive gain in every SPX trend, every VIX bucket and every episode (+0.10R to +0.74R) → holds everywhere · F19 = stress signal (21/10/2026, : stressed market (SPX < MA200 or VIX ≥ 20) → +0.33R absolute (1,695 trades, 307 dates, gain +0.36R vs the location without anomaly), on the 113 original assets (+0.44R) and the 424 never-seen ones (+0.29R), 4 years out of 5 (2026 −0.49R, n 115); calm market → −0.04R absolute (gain +0.11R) → trade it alone only under stress; stress rule chosen after the regime audit → to confirm live
off-exchange blocks
🟡day's block notional ≥ 3× its 30-session median, robust z30 ≥ 2, ≥ 2 blocks; entry at the next day's close
Measure and method (lab notes, reference F28)
🟡 fragile (20/10, 519 tickers, 05 → 10/2026): +0.21 to +0.29R vs the location alone, but alone −0.36R in the 2nd half; with F19: +0.20R, 2 halves positive; same-day entry ⚪ / ❌ · example XLF 29-30/09/2026 (C049): blocks 7.1× and 11.1× the median at the 10 d low below −1σ month, low 52.82 on 01/10 on a double discount, +4.4% by 21/10
fast drop + off-exchange burst
🟡drop ≥ 3σ (20-day realized vol) over 3, 5 or 10 sessions + at least one day with z30 ≥ 3 of the dark share (FINRA / total volume) over the last 3 or 5 sessions; entry at the next day's close
Measure and method (lab notes, reference F35)
🟡 candidate 21/10/2026 : 6 neighbouring cells on 113 assets (+0.37 to +0.78R vs −0.17R for the drop alone, 25-48 trades); replicates on 424 never-seen assets (+0.12 to +0.31R vs −0.03R; F19 +0.07R), 2 halves positive; small samples, 0-3 trades in 2026 → live tracking · regimes 21/10/2026: zero gain when VIX 15-20 (−0.01R, n 94) and during the 2022-2024 recovery (0.00R, n 109), +0.41 to +0.63R elsewhere → regime-dependent
ETF pullback + off-exchange burst
🟡close ≤ 0.92 × 63-session high, 30-day robust z of FINRA off-exchange volume ≥ 2 on D to D−2 (since 03/11: bursts on expiry Fridays, their eve and month-ends excluded; cleaned rule +1.48 / +2.71 pt, 133 cases, does not fire on 30/10), SPX GEX (SqueezeMetrics, proxy) < 0 or bottom 10% over 252 d; variant + 5-day short ratio ≥ 80th percentile
Measure and method (lab notes, reference F41)
🟡 t+1 entry (06/11, : r20 +1.2 / +3.7 pt (original rule; the « cleaned » version of 03/11 has no script); v2 ✅ but 12 pooled ETFs → p and episodes optimistic, and rule tweaked after the signal → 🟡 · before 06/11: ✅ 12-ETF panel 28/10/2026 , 2021-26): r20 +1.1 / +3.4 pt on 2 halves (88 / 69 cases), placebo +0.4 / +2.1; r5 near zero in the 1st half (+0.1); 60 d vol after +6.7 pt (« more volatile » claim confirmed) · short ratio variant ⚪ (r20 +2.5 / +4.8, negative placebo, 22 / 18 cases) · SMH alone ⚪ (5 / 10 cases, 1st half negative) · SMH's own GEX: after recharge · Fired on 30/10/2026 on QQQ and SMH (QQQ 0.905 of the 63 d high, off-exchange z30 +3.3, index GEX rank 6%; SMH 0.817, +2.5); F24 proxy narrowly missed (GEX z −1.76, QQQ dark buying ×3.17, SMH ×2.21); GEX OI convention (calls +, puts −, 1-120 d) QQQ −$6.5bn / 1% vs −7.3 on 16/10, SMH −0.33 vs −0.85
GEX at a low then DPI at a low
🟡GEX in the bottom 10% of 252 sessions on one day from t−5 to t−1, GEX back up since, and DIX in the bottom 5% at t, below its level on the day of the GEX low (02/11/2026: GEX 6% on 30/10, DIX 1.2% → present)
Measure and method (lab notes, reference F57)
🟡 t+1 entry (leading since 06/11): +3.4 / +3.5 pt at 20 d (+4.35% at t), 19 episodes (< 30: ✅ impossible), p < 0.001 · before 06/11: 🟡 fragile 03/11/2026 , SPX 2011-26, n 20): +4.35% at 20 d, 85% up vs +1.31% / 68% after a decline; halves +3.88% / +4.83%; excluding 2020 +3.50%; 36 settings all above; low GEX alone already does +2.11% (70%) → the DIX mostly adds consistency; high vol (27%) · v2 04/11: 🟡 (19 episodes, p < 0.001, t+1 entry: +3.4 / +3.5 pt at 20 d instead of +4.35%)
high DPI + negative GEX
🟡DIX in the top 10% of its 252 sessions and GEX < 0 on the same day (rule-based version of the cyan dots on the @TailThatWagsDog GEX-DIX chart, placed by hand)
Measure and method (lab notes, reference F62)
🟡 03/11/2026 , SPX 2011-26, 32 episodes): +20 d +2.55% / +2.57% vs base +0.75 / +1.24% on the 2 halves, next-day entry same; +60 d 91% / 67% up. Their orange dots (low DIX near the high): no effect, not kept
crowded index puts
🟡Put share of SPXW premium (Skylit, daily) in its top 10% of 120 sessions, two days in a row; variants net premium, SPXW + QQQ the same day
Measure and method (lab notes, reference F74)
04/11/2026 , 2023-26): SPY +5 d +1.27% / +1.01% vs +0.38 / +0.40% (11 + 11); next day ≈ base → 🟡 lead (single period, 3 variants); on 03/11 rank 86%: did not fire · v2 04/11: 🟡 (19 episodes, p 0.003, t+1 entry
APMEM vol easing (a)
🟡22-day vol forecast from the APMEM model (Engle-Gallo, Parkinson variance); 12-day exponential average crossing above the 40-day (2 d of confirmation) = SHORT « volatility picking up again »; below = LONG; spike z ≥ 2.5 = LONG « recovery after a crash »; minimum 50 sessions between 2 signals; two variants on the PANDORA S&P chart (docs/SIGNAUX.md): G33 (a) apLong / apShort = rolling refit 756 / 21, γ fixed 0.332, horizon 21 (pandora/chart/layers.py); G33 (b) volRegLong / volRegShort = yearly walk-forward fit since 1990, horizon 22, the one measured here and the one in the brief (pandora/reports/market_stats.py)
Measure and method (lab notes, reference G33a+)
🟡 28/10/2026 , parameters re-estimated each year on past data, 2000-26, 39 LONG / 40 SHORT): S&P at 3 months +1.3 pt after LONG vs −0.8 pt after SHORT, spread in the right direction on the 2 halves; as an exit rule ❌: Sharpe 0.32 vs 0.41 buy-and-hold, CAGR 5.8% vs 8.3% (worst drawdown −48% vs −55%); matches 5 / 7 dates from the site; SHORT since 21/10/2026
APMEM vol tightening (a)
🟡22-day vol forecast from the APMEM model (Engle-Gallo, Parkinson variance); 12-day exponential average crossing above the 40-day (2 d of confirmation) = SHORT « volatility picking up again »; below = LONG; spike z ≥ 2.5 = LONG « recovery after a crash »; minimum 50 sessions between 2 signals; two variants on the PANDORA S&P chart (docs/SIGNAUX.md): G33 (a) apLong / apShort = rolling refit 756 / 21, γ fixed 0.332, horizon 21 (pandora/chart/layers.py); G33 (b) volRegLong / volRegShort = yearly walk-forward fit since 1990, horizon 22, the one measured here and the one in the brief (pandora/reports/market_stats.py)
Measure and method (lab notes, reference G33a-)
🟡 28/10/2026 , parameters re-estimated each year on past data, 2000-26, 39 LONG / 40 SHORT): S&P at 3 months +1.3 pt after LONG vs −0.8 pt after SHORT, spread in the right direction on the 2 halves; as an exit rule ❌: Sharpe 0.32 vs 0.41 buy-and-hold, CAGR 5.8% vs 8.3% (worst drawdown −48% vs −55%); matches 5 / 7 dates from the site; SHORT since 21/10/2026
vol regime calming (b)
🟡22-day vol forecast from the APMEM model (Engle-Gallo, Parkinson variance); 12-day exponential average crossing above the 40-day (2 d of confirmation) = SHORT « volatility picking up again »; below = LONG; spike z ≥ 2.5 = LONG « recovery after a crash »; minimum 50 sessions between 2 signals; two variants on the PANDORA S&P chart (docs/SIGNAUX.md): G33 (a) apLong / apShort = rolling refit 756 / 21, γ fixed 0.332, horizon 21 (pandora/chart/layers.py); G33 (b) volRegLong / volRegShort = yearly walk-forward fit since 1990, horizon 22, the one measured here and the one in the brief (pandora/reports/market_stats.py)
Measure and method (lab notes, reference G33b+)
🟡 28/10/2026 , parameters re-estimated each year on past data, 2000-26, 39 LONG / 40 SHORT): S&P at 3 months +1.3 pt after LONG vs −0.8 pt after SHORT, spread in the right direction on the 2 halves; as an exit rule ❌: Sharpe 0.32 vs 0.41 buy-and-hold, CAGR 5.8% vs 8.3% (worst drawdown −48% vs −55%); matches 5 / 7 dates from the site; SHORT since 21/10/2026
vol regime rising (b)
🟡22-day vol forecast from the APMEM model (Engle-Gallo, Parkinson variance); 12-day exponential average crossing above the 40-day (2 d of confirmation) = SHORT « volatility picking up again »; below = LONG; spike z ≥ 2.5 = LONG « recovery after a crash »; minimum 50 sessions between 2 signals; two variants on the PANDORA S&P chart (docs/SIGNAUX.md): G33 (a) apLong / apShort = rolling refit 756 / 21, γ fixed 0.332, horizon 21 (pandora/chart/layers.py); G33 (b) volRegLong / volRegShort = yearly walk-forward fit since 1990, horizon 22, the one measured here and the one in the brief (pandora/reports/market_stats.py)
Measure and method (lab notes, reference G33b-)
🟡 28/10/2026 , parameters re-estimated each year on past data, 2000-26, 39 LONG / 40 SHORT): S&P at 3 months +1.3 pt after LONG vs −0.8 pt after SHORT, spread in the right direction on the 2 halves; as an exit rule ❌: Sharpe 0.32 vs 0.41 buy-and-hold, CAGR 5.8% vs 8.3% (worst drawdown −48% vs −55%); matches 5 / 7 dates from the site; SHORT since 21/10/2026
APMEM SPY long
🟡(Léo 10/11: « why does he have a signal and you don't ») his page marvelous-meringue-47da32.netlify.app embeds 752 sessions (11/2023 → 11/2026: APMEM vol, regime, signals, flags); S&P APMEM vol fitted to his series (correlation 0.93, frozen scale); LONG « vol rollover » = down day, ≥ 4% below the 1-year high, vol receding, 1-year vol rank ≥ 60%, vol ≤ 93% of its 20-day peak, from cash; SHORT « momentum exhaustion » = ≤ 2.5% from the high, vol rank ≤ 30%, 20 d ≤ +1%, 5 d < 0; 1-day / 1-week / 1-month forecasts = reversion toward 13.75% at 0.922 / day (his 3 published sets reproduced to the hundredth)
Measure and method (lab notes, reference G36+)
🟡 10/11/2026 , grid fitted on his window only): his dates found to the day 04/04/2024 short, 30/07/2024 long, 05/11/2026 long; 3 / 9 within ±3 sessions, same regime 67% of days; outside his window long / cash vs buy-and-hold: 2000-11 Sharpe 0.09 vs 0.14, worst drawdown −55% vs −55%; 2012-23 Sharpe 0.66 vs 0.82, worst drawdown −27% vs −34% → cuts the worst drawdown, not the return. His top / bottom watch flags (Signal Sigma options) not reproducible; Cboe put / call substitute ⚪
APMEM SPY cash
🟡(Léo 10/11: « why does he have a signal and you don't ») his page marvelous-meringue-47da32.netlify.app embeds 752 sessions (11/2023 → 11/2026: APMEM vol, regime, signals, flags); S&P APMEM vol fitted to his series (correlation 0.93, frozen scale); LONG « vol rollover » = down day, ≥ 4% below the 1-year high, vol receding, 1-year vol rank ≥ 60%, vol ≤ 93% of its 20-day peak, from cash; SHORT « momentum exhaustion » = ≤ 2.5% from the high, vol rank ≤ 30%, 20 d ≤ +1%, 5 d < 0; 1-day / 1-week / 1-month forecasts = reversion toward 13.75% at 0.922 / day (his 3 published sets reproduced to the hundredth)
Measure and method (lab notes, reference G36-)
🟡 10/11/2026 , grid fitted on his window only): his dates found to the day 04/04/2024 short, 30/07/2024 long, 05/11/2026 long; 3 / 9 within ±3 sessions, same regime 67% of days; outside his window long / cash vs buy-and-hold: 2000-11 Sharpe 0.09 vs 0.14, worst drawdown −55% vs −55%; 2012-23 Sharpe 0.66 vs 0.82, worst drawdown −27% vs −34% → cuts the worst drawdown, not the return. His top / bottom watch flags (Signal Sigma options) not reproducible; Cboe put / call substitute ⚪
battery A low
✅(close < MA200 or ≥ 10% below the 252-day high) AND SKEW in its bottom 10% over 252 d AND SMH − QQQ over 5 d ≥ +2 pt
Measure and method (lab notes, reference L1)
✅ 22/10/2026 , 13 ETFs, test 2020-26): +2.9 pt at 20 d vs normal
Nasdaq turbulence
🟡semis lagging QQQ · VIX9D > VIX · VVIX in its top 20% · HYG / IEF in its bottom 20% · SPX GEX < 0 or bottom 20%
Measure and method (lab notes, reference L10)
🟡 t+1 entry (06/11, : QQQ ≤ −2% open → t+1 close 6.6% vs 3.4%; v2 ✅ (47 episodes 21 sessions apart) → ✅ candidate (Léo review); signal of big sessions in both directions · before 06/11: 🟡 24/10/2026 : next session ≤ −2% 10.3% and ≥ +2% 11.2%; 47% near the lows → big sessions in both directions, not a top signal
VIX capitulation + buy confluence
🟡(Léo 10/11: « if we have this signal, this signal and this signal, what does it give ») P14 and P19 each lit within the last 5 sessions, one of the two on the same day; entry at the t+1 open; 10 sessions between episodes
Measure and method (lab notes, reference L13)
🟡 10/11/2026 , PANDORA chart 2021-09 → 2026-11, 225 combinations tested, Benjamini-Hochberg): 18 episodes, +21 d +2.59 pt above the base, 89% up, q 0.04; only pair that beats its members (P19 alone +2.11 pt); short history (5 years), gap vs P19 alone not tested; whole families (2 or 3 bullish signals within 3 or 5 sessions) ⚪
round-level break absorbed
🟡S&P closes under a multiple of 100 it held 5 sessions (R6 round100) and DIX ≥ 0.45 the same evening; long close t → t+20; 10-session cooldown
Measure and method (lab notes, reference L14)
🟡 18/11/2026 (PANDORA , SPX 2011-26, 0 credits): 43 cases, 20 sessions +2.91 % vs +0.98 %, up 77 % vs 67 %, p 0.001 two-sided (Bonferroni 0.0025 over 20 crosses); 2011-20 +4.68 % (n 7) and 2021-26 +2.56 % (n 36); 15 variants DIX 0.43-0.47 × held 4-6 all +1.9 to +3.4 %; not fired 16/11/2026 (DIX 0.409); on the S&P chart since 18/11/2026 (Léo's go), not in the brief
battery B low
✅10-session low AND RSP / SPY over 20 d in its bottom 10% AND monthly OPEX week
Measure and method (lab notes, reference L2)
✅ 22/10/2026 : +5.0 pt at 20 d in the test
battery C distribution
✅close ≥ 98% of the 252-day high AND RSP / SPY over 20 d in its bottom 10%
Measure and method (lab notes, reference L3)
✅ 22/10/2026 : −2.0 pt at 20 d, −5.1 pt at 60 d, 13 / 13 assets; on SPY alone since 2021: holds at 60 d only
B1 capitulation (day 1)
🟡≥ 3 red days out of 10 with DIX ≥ 85th percentile 63 d and GEX SMI < −20
Measure and method (lab notes, reference P13)
🟡 (1st day of episode, 06/11) · before: ⚪ weak · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 9 cases, +2.82%, 78% · 06/11 (PANDORA , SPX 2011-26, D+1 close entry, 21 sessions, 10 bp): every marked day (209 d, 28 episodes) +0.61 pt vs base, p 0.21 → ⚪; 1st day of episode +1.05 pt, halves +0.78 / +1.34, p 0.070 → 🟡 (28 episodes < 30) → chart: 1st day of episode only (PR #85)
VIX curve capitulation
🟡VIX9D / VIX3M > 1.05, signal on the first easing below 95% of the episode peak
Measure and method (lab notes, reference P14)
🟡 · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 30 cases, +3.35%, 77% · re-measured on 10/11/2026 over the full history (PANDORA , entry at the next open, 21 sessions): 2011-26 153 cases, +2.01% against +0.81% all sessions (circular permutation p 0.013, 90% CI +1.3 / +2.7); 2011 → 2021-09: 115 cases, +1.61% against +1.04% (p 0.18); 2021-10 → 2026-10: 38 cases (not 30), +3.23%; 108 intervals out of 153 are shorter than 21 sessions (clustered signals); conditional « recent stress » benchmark (ratio > 1.05 in the last 10 sessions): +1.84% on 2011-26, so the trigger adds ≈ +0.2 pt to the stress state: to test against this benchmark before any ✅
buy confluence
🟡≥ 2 independent buy families on the same day or within the 2 prior sessions (capitulation, off-exchange, dealers, exhaustion)
Measure and method (lab notes, reference P19)
🟡 · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): capitulation + off-exchange 23 cases, +4.45%, 87% (best combination, PASS)
DPI divergence ▲
🟡S&P ≤ −0.5% in the session and DPI (DIX) ≥ +1 pt
Measure and method (lab notes, reference P3)
🟡 · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 93 cases, +1.46%, 70% up
GEX crossover ▲
🟡very negative GEX then a sharp rebound, or 5-day average crossing back above zero
Measure and method (lab notes, reference P8)
🟡 · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 25 cases, +3.12%, 68%
Research · 23
Show the 23 research signals (shown nowhere else but the Lab)
extreme accumulation
⚪SqueezeMetrics GEX z-score over 126 sessions ≤ −2 and DIX rank among its last 60 sessions ≥ 0.95 (top 5 %), the same day; a new mark only when F24 was off the 5 prior sessions; known at close t (pandora/strategies/f24_extreme_accumulation.py)
Measure and method (lab notes, reference F24)
Definition corrected 19/11/2026 to match the code (old text: « GEX ≤ −2σ from its 126-session mean and the ETF's « bought » dark volume (FINRA short volume) ≥ 2× its 63-session median; their version: total GEX of QQQ / SMH; ours: the index's SqueezeMetrics GEX (proxy) ») · ⚪ promising, small n (proxy 2022 → 15/10/2026, measured from the next day's close): QQQ 3 episodes +6.6% at 10 d, 3 out of 3 up (vs +0.65%); SMH 6 episodes +4.5%, 83% (vs +0.96%); SPY 4 episodes +1.6% / +5.3% at 20 d; extreme GEX without dark buying ≈ base; widened version (−1.5σ, 1.5×) diluted → the extremes matter. Fired on 16/10/2026: index GEX −2.19σ, QQQ 2.84×, SMH 3.13× (SPY 1.69×). F19 family (dark accumulation at a buy location). Generalized (engine , any asset with data): native SPX (SqueezeMetrics GEX + SPY FINRA dark) 4 episodes +1.5% / +5.3% at 20 d; SPX with the DIX as dark buying (2011 →) 13 episodes +1.6% vs +0.5%; 538 assets with index GEX as proxy (2022 →): +1.44% at 10 d vs +0.48%, 61% up, but clustered on 52 dates: the first signals of a decline fail (Feb-Mar 2026), the later ones pay → to diagnose. SMH case 16/10/2026 (registry C041): dark buying 3.13×, dark prints ≥ $5M stacked on the 0.618 / 0.66 / 0.706 fibs of the discount ($592M), tracking open. Live (19/10): SPX in the PANDORA Telegram brief (state IDLE / ARMED / CONFIRMED); per ticker, never mix Skylit GEX and open-interest GEX in the same z-score (guide 44) · Live 16/10/2026: failed at +10 sessions (30/10: QQQ −2.7%, SMH −3.2% below its invalidation 529.53, SOXX −4.4%); +20 on 13/11
Mag 7 off-exchange burst
⚪average of 30-session robust z of total FINRA off-exchange volume of AAPL, MSFT, NVDA, GOOGL, AMZN, META, TSLA ≥ 2, QQQ closed at a 10-session low within the last 3 sessions; measured from the close of D+1 to D+11
Measure and method (lab notes, reference F39)
⚪ promising 21/10/2026 , 2021-2026): +4.61% at 10 d vs +0.70% at the same location without a signal, gain +3.9 pt, 6 episodes out of 6 up, 2 halves positive, p 0.02; 6 episodes < 8 required, all with VIX ≥ 20 (consistent with F19 under stress) → live tracking, to recount · fired on 30/10/2026 with VIX at 19.4, below the 20 of every measured episode; D+1 to D+11 result to recount
unusual SPY off-exchange volume
⚪total SPY off-exchange volume, robust z ≥ 2 vs 30 sessions, anywhere / at the low / near the high; variant excluding expiry and month-end
Measure and method (lab notes, reference F61)
⚪ 03/11/2026 , 2018-26): no direction (negative 2018-22, positive 2022-26); 29% of spikes fall on an expiration day (8% of sessions) → mechanical effect; high short share at the low ❌
leveraged ETFs: heavy outflows
⚪ProShares, 18 leveraged index ETFs (long + inverse), 1-month flow ÷ AUM, 2-year rank; HIGH ≥ 90%, LOW ≤ 10%
Measure and method (lab notes, reference F70+)
03/11/2026 , 2008-26): strong inflows ⚪, not bullish (+2.13 pt then −1.13 pt at 60 d; since 2018 drop ≥ 10% within 60 d 30% vs 14%); strong outflows = bullish (+2.77 / +1.85 pt, after a pullback +5.65 / +2.89 pt, breakage 10% / 0% vs 26% / 24%) → 🟡 contrarian. As of 02/11: inflows at rank 93% (HIGH episode on 22/10), net longs − inverses 96%
leveraged ETFs: heavy inflows
⚪ProShares, 18 leveraged index ETFs (long + inverse), 1-month flow ÷ AUM, 2-year rank; HIGH ≥ 90%, LOW ≤ 10%
Measure and method (lab notes, reference F70-)
03/11/2026 , 2008-26): strong inflows ⚪, not bullish (+2.13 pt then −1.13 pt at 60 d; since 2018 drop ≥ 10% within 60 d 30% vs 14%); strong outflows = bullish (+2.77 / +1.85 pt, after a pullback +5.65 / +2.89 pt, breakage 10% / 0% vs 26% / 24%) → 🟡 contrarian. As of 02/11: inflows at rank 93% (HIGH episode on 22/10), net longs − inverses 96%
1-day vol event premium
⚪Tempest 1-day vol ÷ 30-day vol ≥ 1.25 at the close (or top 10% over 120 d)
Measure and method (lab notes, reference F75)
⚪ t+1 entry (06/11, : +0.53 / −0.10 pt (2nd half flips sign at the open), 2 years → v2 ⚪ · before 06/11: 04/11/2026 , 2024-26): Nasdaq-100: next day +0.76% / +0.24% vs +0.09%, up 70% / 61% (23 + 23) → 🟡 lead; S&P ⚪. 03/11: 1.31 (S&P) and 1.56 (NDX, rank 99%) · v2 04/11: 🟡 (35 episodes, p 0.019, only 2 years → not ✅)
Unusual off-exchange activity (record 3 → 30 months or ≥ × 2 the 6-month average), read long or short
⚪(Léo 09/11: « biggest dp activity » over 3, 6, 12, 18, 24, 30 months, long or short, with the day's GEX, all PANDORA assets) FINRA off-exchange volume × close at the high of the previous 63 / 126 / 252 / 378 / 504 / 630 sessions; the longest window beaten is shown; read by location: LONG = down day and ≥ 5% below the 1-year high (discount), SHORT = within 2% of the high (premium), otherwise context; day's GEX = rank of the S&P GEX over 252 sessions (low ≤ 20% or < 0, high ≥ 80%); 1st day of the episode (10 sessions)
Measure and method (lab notes, reference F84)
⚪ 09/11/2026 , 16 tickers SPY QQQ VOO IVV SMH IWM DIA XLK XLF + Mag 7, 2018-26, t+1 open entry, Holm over 24 tests): 0 🟡. LONG + low GEX: +5 d +1.2 to +2.1 pt vs context, 62-70% up (6 months: 37 ep., +2.52%, 68%) but 1st half negative everywhere; SHORT + high GEX: +5 d −1.0 to −2.3 pt vs context (18 months: 24 ep., −1.49%, halves in the right direction) but price still rises in more than one case out of two, p 0.34-0.39; 6-month record alone , 183 ep.): +0.15 pt; a down day ≥ 1%: +5 d +1.12% (61% up) → a read to watch, not a trade. 05/11/2026: QQQ 30-month record (LONG + low GEX), SPY 6 months, VOO 30 months, XLK 30 months · multiple , 10/11): ≥ × 2 the 6-month average at a discount + low GEX 114 ep., +5 d +1.47% (+1.01 pt vs context, 58%); ≥ × 3 at a premium + high GEX 34 ep., +21 d −2.50%, halves in the right direction, p 0.35; 12 tests, 0 🟡
inverted VIX futures curve
⚪settlement of the 3rd monthly VIX contract − 1st (Cboe) < 0 at the close; « negative for equities » state until back to ≥ 0; long proxy VIX3M − VIX
Measure and method (lab notes, reference G34)
⚪ t+1 entry (06/11, : Sharpe 0.69 vs 0.75 buy-and-hold, worst drawdown −24.5%; as a 21 d signal: opposite sign on the 2 halves (❌); as protection ⚪ · before 06/11: 28/10/2026 , futures 2013-26 + proxy 2008-26): as a bearish signal ❌ — S&P +1.9 pt at 21 d vs normal on inverted days (2 halves, 2 datasets), median episodes of 2 sessions, S&P down during the episode only 32% of the time; as protection 🟡 — stepping out during inversion: same Sharpe (0.76 vs 0.75) with a worst drawdown of −22% vs −34%, but −3 pt of return per year; 3 halves out of 4. Curve positive as of 28/10
put/call fear
⚪10-day average of total put/call ≥ 0.99 = fear; ≤ 0.82 = complacency
Measure and method (lab notes, reference L12+)
⚪ 24/10/2026 : fear → S&P +1.2 / +1.6%; complacency +0.4 / −0.4%
put/call complacency
⚪10-day average of total put/call ≥ 0.99 = fear; ≤ 0.82 = complacency
Measure and method (lab notes, reference L12-)
⚪ 24/10/2026 : fear → S&P +1.2 / +1.6%; complacency +0.4 / −0.4%
gamma × curve ▲
⚪SqueezeMetrics aggregate GEX proxy: exit from an extreme state, VIX curve normalizing
Measure and method (lab notes, reference P1)
⚪ weak · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 31 cases, +1.33%, 58% up
price · GEX ▲
⚪lower low on the S&P and higher low on GEX (▲) or the reverse at the top (▼), confirmed 20 sessions later
Measure and method (lab notes, reference P11)
⚪ (low n) · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 6 + 6 cases
GEX quiet decline
❌fast GEX drop while the VIX does not move
Measure and method (lab notes, reference P12)
❌ as an alert · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 17 cases, no decline afterwards
washout
⚪≥ 4 of the 6 capitulation channels at the same time (DIX, GEX, VIX curve, breadth, credit, drawdown)
Measure and method (lab notes, reference P15)
⚪ (low n) · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 5 cases, +6.83%, all up
VIX renormalisation
⚪exit from a lasting VIX curve inversion
Measure and method (lab notes, reference P16)
⚪ (low n) · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 5 cases, +4.81%
credit veto
⚪S&P at a 6-month high while HYG / LQD lose > 2% over 63 d
Measure and method (lab notes, reference P17)
⚪ (low n), contradicted · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 7 cases, S&P +2.7% afterwards
gamma × curve ▼
❌same proxy, each qualified sell observation
Measure and method (lab notes, reference P2)
❌ as a sell · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 44 cases, the S&P rose 1.1% afterwards
DPI divergence ▼
❌S&P up and DPI down
Measure and method (lab notes, reference P4)
❌ as a sell · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 109 cases, S&P +1.2% afterwards
DPI absorption ▲
⚪new price low, off-exchange accumulation line (cumulative DIX − 252-day median) above its low
Measure and method (lab notes, reference P5)
⚪ weak · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 28 cases, +1.15%, 61%
DPI distribution ▼
❌new price high, accumulation line stalling
Measure and method (lab notes, reference P6)
❌ as a sell · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 69 cases, S&P up afterwards
accumulation regime
⚪DIX in its top quintile and GEX falling over 5 d (start of episode)
Measure and method (lab notes, reference P7)
⚪ weak · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 49 cases, +0.87%, 63%
GEX crossover ▼
❌first day of negative GEX after a positive phase
Measure and method (lab notes, reference P9)
❌ as a sell · PANDORA 5-year audit (PANDORA , 2021-10 → 2026-10, S&P at 21 sessions, sells sign-flipped): 40 cases, S&P +2.6% afterwards
term inversion at the high (short)
—Personal decision-support tool, not investment advice. Past results do not guarantee future results; a trade can lose more than the planned risk on an opening gap. Glossary