Caveat. The dossier's live-path exhibit was drawn off intraday strength; close-to-close the ratio is marking near its 1.584 entry. Don't over-read the first two sessions — even the pre-2016 winners gave back 1-3% early before working, and all three AI-era analogs were underwater by session 5. The Session-5 close is the real tell.

Caveat. n=5 cluster events, recency-weighted (Oct-2025 = 7×) — the modern regime is less forgiving than the 2023 vintage (weighted day-43 mean +0.26% vs +0.84% unweighted). The edge is time-boxed: profitable T+1→T+15, turns against you past day 21 (mandatory flat). One 2023 analog (Mar-31, the SVB regime break) ran −4.56% and carries the right tail; the others mean-reverted. A tactical 2–3-week short, not a trend trade. See the full blueprint.

Caveat. Sample-size humility, straight from the study: the 40 events collapse to 18 independent episodes under a 21-session exclusion window; block-bootstrap p-values are 0.12–0.17 (favorable, not conventionally significant); and the winning subset is concentrated in regimes (2021/2024/2025-26) where buying momentum weakness was broadly rewarded. The trigger is a situation, not a signal — the rising 65d resolves it, so the MA rolling over is the real invalidation. See the full event study.

Caveat. Honest read: N=1 (the exact setup) + N=2 (the cousins) — anecdotal by construction, clears no significance bar. MU's unconditional 21-day drift is already +17.6% / 74% positive, so the analog's +54% is ~93rd percentile of normal paths, not a proven conditional edge — the claim is directional consistency (3-for-3 at +21d), not magnitude. The trade rests on structure: a defined-risk long at a level the market just violently defended, with an unusually close invalidation (close < $804 = −10.4%). Fractional Kelly-Thorp → quarter-Kelly starter, add only on confirmation. The 9.1% daily ATR cuts both ways — the stop can gap through. See the full note.

Caveat. Brutally honest small-n: n=7, block-bootstrap bands straddle zero, no statistic survives significance testing. The full-sample forward medians actually drift the ratio UP (+3.4% at W+6) — against the short; the bearish case rests entirely on the hot-run subset (n=3) plus recency-weighting. STORM scores it accordingly — escore 0.45, ~1.0% NAV — a lottery-adjacent asymmetry bet, sized as one. Week-1 is the soft spot (mean −0.9% for the ratio, i.e. weak early edge). See the full note.

Caveat. N=8 — a thin long-run cohort; medians are the operative statistic (means are hostage to the Mar-26 outlier). Regime dependence: the one contradicting path (Mar-26, +27.6%) is the most recent, from the same reflexive-bid tape in question — whether it's the new base case or the outlier is a regime judgment, not a statistic. The signal is late by construction (a rising MA means part of the break is arithmetic), which is exactly why the posture is 30-day tactical. See the full event study.

Caveat. N=7 is a cohort, not a distribution — every analog sits in a distinct macro narrative (Jan-2023 recovery, DeepSeek shock, April-2025 tariffs), so the operative question is whether the current macro rhymes. The Apr-7 & Apr-9 2025 events overlap forward windows (double-count one regime). Symmetric discipline: had Jul 9 closed weak, this same magnitude would read as breakdown confirmation. See the full study.

Caveat. A direction-of-tape signal on a structurally strong compounder; the one clear modern-regime failure (Sep 2023) ran ~-8%, which the hard invalidation caps — and the slow scale keeps size small until the reversal low proves itself. n=5 in the 2023+ primary cohort — a cohort, not a distribution. Not investment advice.

Caveat. The hard-break cohort is n=3 — anecdote-adjacent, and each prior (Volmageddon, the 2018 trade war, COVID) was an exogenous macro shock that expressed through this pattern rather than being caused by it. In-sample, single-instrument, with the 65-day parameter selected off the current tape (a forward-looking bias the study can't fully purge). This is a management overlay on a retained core long, not a fresh directional bet — the study argues against flipping outright short even on confirmation (Jan 2020 finished +2.3% at 126d). See the full event study.

Caveat. Honesty check: the source's −0.76 is a levels correlation (a trend artifact — the daily-change correlation is only ~+0.10), and correlation is not cointegration. Our back-test of 20 prior decouplings tempers the thesis: the divergence Z does mean-revert (~78% narrow by 63d), but it reverts through equity underperformance (IWV lagged its baseline by ~5 pts at 63d), not the yield-normalization the blueprint's Leg 2 leans on — the 2-year barely moved. So the edge is in the bearish-IWV leg, the long-SHY leg is weakly supported, and the 4.50% weekly-close 2Y stop is the right rail. See the back-tested study & the source blueprint.

Caveat. A weighting is a view, not new data. Unweighted, this fades hard (+63d −9.8%, 39% green) — the ~15 dead-cat events were PLTR's 2021–22 −80% bear. The reweight expresses that the current construct rhymes with 2023+, but the up-weighted modern cohort is only n=3 and its average is tail-carried; if PLTR is entering a real downtrend it reverts to the bear read, which is why the reversal-day low is the hard stop. See the back-tested study for the weighted vs unweighted paths.

Caveat. The research's headline '>92% at days 11-15' is a survivorship-conditioned over-the-streak figure (streaks that lasted that long were up over the span); the clean forward-10-day edge from day 11 is far more muted (~50%). Treat day 11 as regime confirmation — a durable low-vol, bullish-small-cap backdrop worth a multi-week hold — not a 92%-win timing lock. See the RVX Duration study.

Caveat. n=7, one seasonal regime (all summer weather markets). The -13% Aug-2015 tail is the entire reason for the -5% stop and modest size; a drought (2012 ran +20%) is the upside tail. See the full STORM study.

Caveat. N=3 strict / N=9 broadened, regime-concentrated in 2025-26 with one +94% underlier outcome. The edge survives ONLY without a stop (a MA-stop rule gets hit 7 of 8 times for -4.7% expectancy) — which is why this is a defined-risk short-put structure, not stock. Option marks aren't tracked live here; the card marks the underlying vs the strikes. See the full dossier.

Caveat. AI-generated briefing (NotebookLM) — the numbers were re-checked against real PANW: the Jul 20 close ($348.66) and the +66% 200-day extension tie out exactly (our 10-day realized vol reads ~48% / 85th-pctile, a touch tamer than the deck's 67% / 95th). The loss scenario — a close above ~$365 in 24 sessions with no earnings catalyst — has happened once in the file's history (the last eight weeks up into this print), so the sample behind the win rate is thin. Key dependency: the thesis requires PANW's Q4 earnings to fall AFTER Aug 21 — confirm the report date; if it lands inside the window the edge inverts. Option marks aren't tracked live; the card marks the underlying vs the strikes. See the full dossier.

Caveat. AI-generated briefing (NotebookLM), re-checked on yfinance JPY=X: the breakout is real — monthly closes went 159.27 (May) → 161.92 (Jun) → 159.76 now, clearing 160.75. The deck's 160.76 vs our 161.92 Jun close is a data-source gap (TradingView vs yfinance FX fixings, ~0.7%). This is a monthly-close macro signal, n=1 of this exact 36-yr break — a thesis, not a base rate. MoF intervention is a genuine two-sided gap risk (precisely why the vehicle is a defined-risk spread, not stopped spot). Option marks aren't tracked live; the card marks USD/JPY spot vs the strikes, and FX leads the equity book by ~a day. See the full dossier.

Caveat. AI-generated briefing (NotebookLM), but the underlier is REAL and re-checked on yfinance: SPCX resolves to Space Exploration Technologies, a Nasdaq (NMS) equity, IPO 2026-06-12, 55 sessions of data, now $143.69. Two honesty flags: (1) it's a ~6-week-old IPO — no long history, thin float, and short borrow on a hot name can be expensive/hard; the −52% is a cross-sectional base rate from OTHER mega-IPOs, not this one. (2) The 911M unlock and $4.3B-bid figures are the deck's — verify the unlock date/size independently before pressing. Hard invalidation is a reclaim of $135. See the full study.

Caveat. Strong stats but small: n=5 completed events, ≈4 independent episodes; three of the five best outcomes trace to the late-2022 bottoming cluster, so the median (+54.8%) is the honest central tendency, not the mean. The live event already broke pattern once — Jul 24 closed −3% (only the 2nd next-day decline in the 10-event sample): a better entry, or an early tell toward the Jan-2024 dud (the one filtered event that finished −5.3% at 50d). The staging exists so the position never has to decide. Data note: the study's $291.72 event low is from BATS OHLC and sits below yfinance's 2-day low (~$308) — using the documented $291.72 as the invalidation shelf. Option marks aren't tracked live; the card marks TSLA vs the levels. See the full event study.

Caveat. OFF-MANDATE — a momentum/breakout continuation (enters INTO strength toward new highs); TEF's edge is behavioral EXHAUSTION, so this runs as a labeled exception like USD/JPY & Soybean. n=5 remains n=5 — 'treat the table as a prior, not a promise'. The t+1 entry ($69.83) sits slightly above the $67 print the deck says not to chase (your directive was the t+1 date); currently +1.1% (a tolerated red week). Real ETF, verified. Hard invalidation = a weekly close below ~$57. See the full pattern study.

Caveat. Honest read up front: this is n=1. The '99.9th percentile / 100% win rate' framing rests on a SINGLE 2020 analog; the study's own probability heatmap is an explicit 50% coin-toss at every horizon through 26 weeks, only 'resolving' to 0%/100% at 52W on that one observation. On-mandate in TYPE (L1 ratio-spread reversion — fading a >3σ dislocation) but near-zero statistical edge. It is time-boxed (the 52-week view inverts), and 4 weeks in it is flat. Sized at 5% per your directive — I'd treat it as the lowest-conviction position after the IWD-class signals. See the full study.

Caveat. Data honesty: in the BATS file the Jul 7 session low did not strictly undercut the prior 65-session low — today's bar qualifies in spirit (gap to range low, positive reversal) but not under the strict definition. n=5, all in one regime; descriptive, not inferential.

Caveat. Regime-conditioned samples are small (7 AI-leg vs 9 post-peak dip-buys; the post-peak entries overlap) and the unwind analog set is n=2. The shelf has not broken on a closing basis — third tests of major support do sometimes hold; tranche sizing is why.

Caveat. BETA — execution-style test, not a committed position. Size is quarter-Kelly of the screen's forward edge, capped at the 10% TEF limit and halved for the first beta run. The 50/50 MOC+TWAP split assumes the auction print and even morning fills. The example instrument rotates as the screen re-ranks — point it at any Bucket-1 name to take it live. Open the Stretch Screen.