💰 Practice money
$77,772
−$22,228 (-22.23%) since 2026-05-08 · Today: +$128
Locked in: −$22,927 · On paper (open): +$699
The things it looks at most when deciding what to bet on:
It does better in some market moods than others — and it knows it:
50 bets in the latest run, 7 of them experiments to learn new things.
So far its experiments beat the market 45 / 100 vs 43 / 100 for its confident bets — measuring that gap is how it learns what actually works.
It has made 271 trades so far — 23 still open and 248 finished. On the finished ones it has lost $22,927 for real; its open positions are +$699 on paper.
Showing the 30 most recent.
Exact sell prices settle at the broker, so closed trades show how they ended, not a per-trade dollar figure.
Every day it writes down what it is learning, then later checks whether it was right. So far it has confirmed 131 ideas that genuinely beat the market and flagged 297 that didn't pan out — and it is dialing those back.
It also caught itself 30005 times trying to trade stocks too illiquid to fill safely — its plan is to stop recommending names that keep failing that check.
The notes below are the bot's own raw jottings — technical, but it's all here if you want to dig in.
# Learnings Playbook
Curated standing trading-rules document, distilled by the Sunday weekend
routine from `memory/learnings.jsonl`. Last rewritten: 2026-09-20 (weekend
prep for week of 2026-09-22, post-FOMC clean week — next binary GDP Thu
9/24, then NFP Fri 10/2). Group by category, drop expired, merge
duplicates, keep ≤200 lines.
## Process
- Always read `learnings.summary` (this file) and `learnings.recent` from
the bundle before writing. Side with a recent lesson over default unless
today's data clearly invalidates it (record regime change as new lesson).
- Append 1–3 lessons every run (wins AND losses). Selection bias toward
only-corrections drifts the bot from validated approaches.
- **COVERAGE REQUIREMENT** — every run issuing an options OR direct-short
pick MUST write at least one lesson covering that horizon / direction
with a falsifiable predicate. Weekend brief 2026-09-20 wrote options
gate (XLF 12-week escalation rule) coverage predicate.
- **Caveat staleness as FIRST visible block** when context is older than
the run date by 4+ hours.
- For pre-event runs (FOMC/CPI/NFP/GDP/PPI/Retail), explicitly defer NEW
entries to **post-event range confirmation**. Tight stops in the 30 min
before release get shredded by 1.5–2× ATR bars.
- **Pre-event breakout-conditional triggers**: gate on VIX/sector
confirmation through ALL multi-event windows; never chase-at-reference
pre-event.
- **CLEAN-week post-stack rule (ACTIVE 2026-09-22)**: the trading week
immediately after a multi-event stack settles T+0 favors *pullback-entry*
(SMA20 + XLK green + rvol < 1.2× + RSI < 72) over *breakout-chase* from
Friday-close reference. This week qualifies (post-FOMC T+2 clean).
- **Weekend brief checklist**: conditional triggers only, this-week-vs-
next-week macro split, gap-risk + 52w-extremes inventory, calibration
+ sizing rule in footer, non-Tech diversifier when concentration ≥ 80%.
- **Futures aggressive-mode override**: futures-* labels bias toward
action, not watch. ≥ 3 directional rows per action plan; ≤ 1 watch row;
direct medium shorts allowed on liquid futures (NQ/ES/YM/RTY/CL/GC) with
hard stop when at least one confirmation present.
- **Trader-signal noise filter**: the trader-live log spams the same
`MET/HAE/VLO/TMO/ILMN blocked at max_per_position/max_gross` lesson every
30s during trading hours. If a trader gate lesson recurs ≥ 20 times in a
window, remove the ticker from the advisor's recommendation list rather
than re-appending the lesson.
- **Cool-streak sizing lock (BINDING 2026-09-20, provisional-lift gated)**:
when 14d attribution hit-rate < 40% floor, posture = STAND ASIDE binding.
All new picks must be watch-with-triggers or defined-risk options only.
NO direct-long stack conviction until 3-session ≥ 60% hit-rate recovery.
Attribution recovered to 54.55% n=33 this weekend — sizing lock stays
1/6 non-floor / 1/3 floor for the week, promoted to provisional 1/3
non-floor / 2/3 floor upon 3-session ≥ 60% intraweek confirmation.
## Regime
- **VIX gates (binary, daily close)**: < 17.0 = deep full-unlock (also
requires IWM 1w lead vs SPY ≥ +1pp — see IWM gate); 17.0–17.5 = fragile
full-unlock (sizing cap 1/3 floor / 1/6 non-floor); 17.5–19.0 = dead-zone,
equity direct-short window LOCKED + long-bias default; > 19.0 = tactical
short ARMED via defined-risk put debit spreads on triple-confirm; > 22.0
= FULL tactical-short unlock (defined-risk only). Equity direct shorts at
conf ≥ medium PROHIBITED regardless of bucket calibration (VIX-gate
binding).
- **Current regime 2026-09-20 (weekend brief)**: VIX 14.81 (**-6.50% 1w**
post-FOMC crush), SPY 1w -0.34%, QQQ +0.92%, IWM **-1.66%**, DXY 100.22
(+1.11% 1w strengthening). Regime = **neutral, posture fragile-partial-
unlock** — VIX deep zone BUT IWM lead-gap -1.32pp fails +1pp gate for
the 4th consecutive week. Sizing cap **1/6 non-floor / 1/3 floor**
binding, provisional-lift on 3-session ≥ 60% confirmation.
- **IWM lead-gap gate** (validated 2026-08-09, re-broken 8/30 → 9/20):
deep-full-unlock requires VIX < 17 AND (IWM 1w % − SPY 1w %) ≥ +1pp.
11-week trajectory: -0.5 / -1.0 / -0.9 / -1.09 / +0.05 / +0.55 / +0.77
/ -1.87 / -0.02 / -0.98 / **-1.32**. Now 4 consecutive weeks failing.
Graduated gate active: 0.5–1.0pp = fragile-partial-unlock, >1.0pp gap
the RIGHT direction = deep-unlock. Current -1.32pp is deep-fail →
fragile posture binding.
- **Post-FOMC T+2 clean-window (ACTIVE 2026-09-22)**: FOMC 9/16 delivered,
vol crush -6.5% 1w confirmed. Next binary macro is GDP Q2 third revision
Thu 9/24 (surprise magnitude typically < 0.1pp — non-binary), then NFP
Fri 10/2. Pullback-entry preferred over breakout-chase per CLEAN-week
rule.
- **VIX one-day-spike bounce predicate** (validated 2026-06-18, 06-21):
when VIX close jumps ≥ 8% in one session into 17.5–19.0 dead-zone
WITHOUT a new catalyst in the following week, next 3–5 sessions favor
VIX crush back < 17.5. MYM long is preferred expression.
- **Post-FOMC/NFP T+1 chop 4/4 stop-out rule**: when post-event T+1 +
slate near 52w highs, trigger placement ≥ session high + 1.25% MINIMUM
+ second 5min K close + 1.5× volume; stop = SMA20 buffer NOT 0.5×ATR
tight. Applies to any post-print breakout re-entry.
- **Attribution 14d recovery 2026-09-20**: 54.55% n=33 mean_return +0.84%
= first above-floor reading in 4 weeks (28.4 → 38.7 → 23.9 → 54.55).
Long_bias 25/25 resolved 56% hit rate. Best MU long +9.93%. Sample thin
(n=33) — formal cool-streak lift requires 3-session ≥ 60% intraweek
confirmation before sizing relax to 1/3 non-floor baseline.
## Sector
- Cite top 2 / bottom 2 sectors from `sector_leadership` in exec summary —
rotation surfaces here.
- **Current rotation (2026-09-20)**: **XLV +1.83% top-1** (recovery from
bottom-1 -4.56% last week — 6.4pp WoW UNSTABLE), **XLK +1.03% #2**
(Tech continues); XLP -0.70%, XLE -1.27%. Bottom: **XLU -3.04% bottom-1**
(rate-sensitive weakness), XLF -2.43%, XLRE -2.05%. Rate-sensitive
triple bottom (XLU/XLF/XLRE) signals sticky-yield concern.
- **1w-rotation-flip instability rule** (added 2026-08-02, active
2026-09-20 for XLV): when a sector's 1w % swings > 8pp week-over-week,
treat current rotation as UNSTABLE — require 2nd week of confirmation
before shifting overweight allocation. XLV +6.4pp WoW is right at the
threshold; combined with prior week's crash (-4.56%), classify as
UNSTABLE for the coming week. Single-week entry allowed only for ELITE
ticker-memory names (IQV 65.62%, HALO 62.5%).
- **Sector-1w-crash-block rule**: when sector 1w < -3.0%, BLOCK direct
entry on any candidate in that sector regardless of name-level ticker-
memory strength. Require ≥ 2 consecutive sessions 1d green before
individual entry. Applies **XLU (-3.04%)** this week → block XLU
single-name entries. XLV NOT blocked (was crushed last week, recovered
this week — recovery gate satisfied).
- **Sector-ETF diversifier XLF sleeve** (validated 17 consecutive weeks
through 2026-09-06, DOUBLE-BLOCKED 2026-09-13 through 2026-09-20):
XLF as validated non-Tech ETF sleeve when Financials 1w flat-to-
positive AND +0.4% 1m intact. **DOUBLE-BLOCKED 10th consecutive brief**
this weekend: XLF 1w -2.43% fails flat-to-positive AND 1m -2.82% fails
+0.4%. Oct 17 58/60 debit spread deferred pending BOTH gates restore.
**12-week escalation rule (added 2026-09-20)**: if DOUBLE-BLOCK persists
through a 12th consecutive brief, sleeve requires formal review —
Financials 1m -2.82% is now BELOW its own ETF SMA200.
- **Options-medium bucket sanctioned alternate**: with XLF sleeve
DOUBLE-BLOCKED, QQQ Oct 17 730/740 defined-risk debit spread is the
sanctioned this-week options expression (VIX 14.81 < 17 permits;
entry on pullback to 716–718 zone).
- **Validated non-Tech diversifier roster**: VRT, AKAM, NXT, GNRC, LNTH,
AA, CIEN, DY, XLF, XLI, XLV, NVT. AI-DC infrastructure non-Tech triple
= DY + GNRC + CIEN when concentration > 80% AND XLI top-5 1w.
- **Sector concentration escalation**: when top sector share ≥ 85%,
review MUST cite non-sector diversifier picks; ETF expression preferred
over single name when ≥ 5 single-name diversifier candidates fire
parabolic predicate. Current top-25 sector share **28% Healthcare +
28% Technology** — well under 85% escalation this week.
- **Non-refiner Energy expression rule (reinforced 2026-09-20)**: when
XLE top-half 1w AND refiner basket in HARD AVOID, express Energy
exposure via CVX (integrated, 75% ticker-memory n=4 thin) or APA/PR/
OVV/EOG (E&P) at low conf with RSI < 65 sweet-spot, NOT the refiner
basket. XLE is #4 this week (-1.27%) — Energy exposure NOT strongly
indicated this week.
## Setup
- **High-volume breakouts** (> 1.5× 20d avg) with RSI < 70 follow through
better than RSI-extended breakouts.
- **RSI sweet spot 55–72**: cleanest entry filter when parabolic basket
fires simultaneously elsewhere.
- **Ticker-memory ≥ 60% priority** (highest-signal single feature):
IQV 65.62% n=32 mean_excess_5d +0.83% (**ELITE**, XLV recovery),
HALO 62.5% n=16 mean_excess_5d +2.14% (**ELITE**, biotech),
FTNT 64.29% n=42 mean_excess_5d +0.42% (elite cybersec),
DINO 64.1% n=39 (elite BUT HARD AVOID refiner cluster),
VLO 62.5% n=40 (elite BUT HARD AVOID),
PSX 58.33% n=24, MPC 74.14% n=58 mean +2.12% (elite BUT HARD AVOID),
PBF 53.57% n=56 (elite BUT HARD AVOID),
MU 52.98% n=151 mean_excess_5d +1.90% (**ELITE + largest sample**),
CVX 75.0% n=4 (thin, tie-breaker only).
Use ticker-memory rank as tie-breaker between otherwise similar setups.
- **Ticker-memory weak HARD AVOID / weak-watch**: ROIV 0% n=11 (weakened
from prior), P (Everpure) 0% n=4, ILMN 21.74% n=23, USO 22.22% n=9,
VEEV 26.67% n=15, DXCM 11.11% n=9, SWKS 33.33% n=3 (thin), TWLO 35.71%
n=14, EBAY 0% n=3, VOYA 37.5% n=8 (weakened from prior). Skip
regardless of catalyst OR use only as HARD AVOID reference.
- **Mean-reversion needs trigger + level + catalyst**, not RSI alone.
Recipe: RSI < 40 + price 0–5% ABOVE 50-day SMA (reclaim) + fundamental
catalyst.
- **Post-earnings soft-forward-guide fade** (BOTH directions validated):
when software/AI-adjacent Q2 print delivers headline beat BUT forward-
guidance framing raises AI-monetization questions AND stock opens
down > 8% AND sector-peer trades same-direction, DOWNGRADE the swing
pick to WATCH ONLY regardless of pre-print thesis strength.
- **Storage / memory / refiner parabolic predicate** = HARD avoid filter:
RSI > 75 + sector-relative-momentum 1m > 40% (or 3m > 80%) + analyst
target < spot. Tactical shorts ONLY on triple-confirm.
- **Refiner parabolic 5-way** (validated 19 consecutive weeks through
2026-09-20): PBF/DINO/MPC/VLO/PSX simultaneously firing RSI 60–90 +
52w-high T0-3 + analyst upside NEGATIVE (-7% to -14%). HARD AVOID
posture. Direct shorts prohibited. Wait signal: ≥ 10% pullback +
RSI < 60 + weekly close reclaims prior swing high with volume.
- **SWKS post-M&A parabolic** (2nd consecutive week HARD AVOID): RSI 70+
+ day-1 52w-high + analyst upside -22.99% (post Skyworks/Qorvo $22B
merger announcement spike overshoot). Ticker-memory 33% n=3 thin. HARD
AVOID direct chase at reference. Direct short prohibited (VIX-gate).
- **HALO post-parabolic pullback** (new priority 2026-09-20): day-0 52w-
high RSI 69.72 + CEO insider buy 60k shares 9/11 + XLV top-1 → pullback
entry to $104–106 SMA20 preferred over chase at $112 reference; elite
ticker-memory 62.5% n=16 mean_excess_5d +2.14%.
- **CL structural short thesis WATCH ONLY**: attribution 14d shows CL
11/11 stop-out streak. HARD AVOID any CL entries without 3-session
structural rebreak.
- **GC persistent-extreme-rvol distribution predicate**: rule active
EXCEPT when DXY < 100 AND weakening. Currently DXY 100.22 (+1.11% 1w
strengthening) → predicate REMAINS SUSPENDED until DXY reclaims below
100 AND weakens 1w. GC 14d attribution mean_r +2.20 continues holding.
- **Friday-weekend forced-trim discipline** BINDING when Fri = T=0 event
print day OR next week is multi-event stack. This week: NOT active
(next Fri 9/26 is empty macro; GDP Thu is third-revision non-binary).
- **Cool-streak-recovery-pattern (new 2026-09-20)**: when 14d attribution
hit-rate recovers above 40% floor after 3+ consecutive weeks below,
DO NOT immediately relax sizing lock — require 3-session intraweek
≥ 60% confirmation. Sample n=33 recovery reading is noisier than the
n=95 floor-fail reading; false-positive recovery risk is meaningful.
## Risk
- Every pick MUST include confidence (Low/Medium/High) and ≥ 1
invalidation level. No exceptions.
- **Cool-streak sizing lock 2026-09-20 (BINDING with provisional-lift
gate)**: 1/6 non-floor / 1/3 floor sizing HOLDS through Mon 9/22.
Provisional-lift to 1/3 non-floor / 2/3 floor triggers upon 3
consecutive intraweek sessions ≥ 60% hit-rate. Full lift requires 14d
attribution to stay above 40% for a second consecutive weekend brief.
- **No multi-event stack this week** (post-FOMC clean). GDP Thu 9/24
third revision is non-binary; NFP Fri 10/2 is next-week not this-week.
T-2-close rule NOT active for the week.
- **Trim-winners operational rule**: at +0.4×ATR take 1/3, +0.7×ATR
take 2/3, last 1/3 trail-stop. MU last-week +9.9% → trim 1/3 at Mon
reference $1015; T-2-close applies before Tue 9/23 MU Q1 print AMC.
- **Equity direct shorts prohibited at conf ≥ medium (BINDING)**:
VIX 14.81 < 17.5 dead-zone → short window LOCKED regardless of
calibration. Short bias expressed ONLY as defined-risk debit spreads
gated on VIX > 19 daily close + triple-confirm. Futures aggressive-mode
overrides this for liquid contracts only.
- **HIGH/long calibration floor-fail (BINDING 2026-09-20)**: high/long
18.75% n=16 (well below 40% floor, WORST direction bucket). Do NOT
stack size into any "high-conviction" single-name long. H/M-conviction
net-long expression sanctioned via ELITE ticker-memory names at
MEDIUM confidence only.
- **No persistent unhedged hold across MU 9/23 print**: T-2 close Mon
EOD for any MU exposure that would otherwise carry into Tue AMC. The
trim-1/3-at-reference rule applies to last-week's +9.9% locked gain.
## Attribution
- **14d equity attribution 2026-09-20**: **54.55% n=33 mean +0.84%**
(first above-floor since 3-week binding cool-streak lock). Long_bias
25/25 resolved at 56%. Best MU long +9.93% (2026-09-13 elite ticker-
memory setup), worst OC watch -3.71% (2026-09-09). Sample n=33 is
thin — provisional recovery, formal lift requires 2nd consecutive
weekend > 40% + 3-session intraweek > 60%.
- 14d futures attribution: continues holding above 50% floor. GC leader
mean_r +2.17. NQ/ES/YM day-trade cluster performing per aggressive-
mode plan.
- **CL attribution FAILED 2026-08-14 through 2026-09-20**: 14 days
11/11 stop-out streak. HARD AVOID any CL entries without 3-session
structural rebreak.
## Calibration discipline
- **Monotonicity NON-monotonic 2026-09-20** (90d n=2796, low 40.74% /
medium 37.50% / high 55.37%). Brier 0.2517 (worse than coin-flip 0.25
baseline by 0.0017). Useful separator: `high` vs `not-high`; low and
medium are effectively identical.
- **Direction warnings (BINDING)**:
- HIGH/LONG 18.75% n=16 (WORST in table, well below 40% floor).
- MEDIUM/LONG 36.32% n=793 (below 40% floor). Reduce single-name long
weight; prefer ELITE ticker-memory or ETF options.
- LOW/LONG 39.99% n=1353 (below 40% floor). Watch entries need
explicit trigger + level; no "monitor broadly."
- **Positive-edge buckets (validated)**:
- HIGH/WATCH 57.06% n=347 mean_excess_5d +1.03% (STRONGEST bucket).
- MEDIUM/SHORT 50.0% n=74 mean_excess_5d +0.75% (VIX-gate override
prohibits direct short entry in current regime, so expressed as
HARD AVOID cluster labels only).
- LOW/SHORT 50.0% n=100 mean_excess_5d -0.28%.
- **Options bucket calibration by conf**: medium/options 36.36% n=44
mean_excess_5d -0.20% (sanctioned bucket but thinner-edge than
historical +0.20%), low/options 38.46% n=117 mean_excess_5d -2.16%
(auto-withhold binding BUT overridden by XLF-diversifier + QQQ
post-FOMC defined-risk sleeve when VIX-gate satisfied).
- **Issue picks as conditional 'watch'** when they fail triple-
confirmation rather than 'buy/short' at reference.
## Pattern playbook
- **Dual-trigger event-day brief**: two mutually-exclusive conditional
orders, dovish-long + hawkish-short (short leg = defined-risk put
debit spread only). NOT active this week (no binary catalysts).
- **Pre-event futures-midday brief** (≥ 1h before release): defer NEW
entries to post-event range confirmation. Manage existing only.
- **Defined-risk options across catalyst-stack windows**: WATCH-ONLY
through catalyst windows. Not active this week.
- **Weekend prep with sector-flip risk overlay**: when a sector's 1w
reverses > 8pp from prior-week ranking, mark current-week rotation as
UNSTABLE. XLV this week +6.4pp WoW (from -4.56% to +1.83%) → sector
UNSTABLE, ELITE-only single-week entry allowed for IQV/HALO; TMO/DHR
require 2nd-week confirmation.
- **Post-parabolic no-chase-at-reference** (validated 7+ consecutive
weeks through 2026-09-20 with MU +9.93% locking prior weekend pick):
when Q2-beat or news print drives session-gain > +8% AND stock
at/near 52w-high AND RSI > 65, DO NOT chase at close reference. Wait
for pullback to SMA20 zone + 5min stabilize + sector ETF green + rvol
< 1.5×. Existing longs trim 1/3 at reference + 0.5×ATR.
- **Post-FOMC/Retail T+1 chop widen-stops rule**: when post-print T+1 +
slate near 52w highs, breakout stop = SMA20 buffer NOT 0.5×ATR tight.
- **Cool-streak posture recovery (new 2026-09-20)**: when 14d attribution
hit-rate recovers above 40% floor after 3+ consecutive weeks below
floor, default brief posture shifts from STAND-ASIDE to *watch-with-
triggers PLUS 1–2 MEDIUM-conf ELITE ticker-memory single-name entries
as provisional partial-lift*. Full new-entry conviction returns after
3-session > 60% intraweek confirmation.
- **12-week ETF sleeve escalation rule (new 2026-09-20)**: any
historically-validated options sleeve DOUBLE-BLOCKED for 12+
consecutive briefs requires formal sleeve-review — treat as
regime-broken until re-confirmed by 2 consecutive gates-restored
briefs. Currently applies to XLF sleeve at 10th brief; escalation
triggers at 12th.
---
This file is rewritten by the Sunday weekend routine after distilling the
recent `memory/learnings.jsonl` tail. Keep ≤200 lines, group by category,
drop expired entries, merge duplicates, demote what recent runs have
invalidated.
# Morning Review — Thu 2026-10-09 *Pre-market, 07:30 ET. Data: 2026-10-08 close + overnight. CPI Tue 2026-10-14 (T-3).* ## Today's Action Plan | # | Ticker | Action | Trigger | Confidence | Stop | |---|--------|--------|---------|------------|------| | 1 | **SMH** | Watch (buy on pullback) | Pullback to 596–600 (SMA20 upper) + 5min reversal + XLK not new day-low + rvol ≥ 1.0× | Medium | 584 | | 2 | **MRVL** | Watch (buy on pullback) | Pullback to 265–268 (SMA20 zone) + 5min hold + rvol ≥ 1.0× + XLK not new day-low | Medium | 254 | | 3 | **FTNT** | Watch (breakout) | Break > 192.50 on 1.25× rvol + XLK 1d green + no post-parabolic reversal 5min | Low | 180 | | 4 | **HAE** | Watch (pullback only — no chase) | Pullback to 112–114 (SMA20) + 5min reversal + XLV green + rvol ≥ 1.0× | Low | 106 | | 5 | **P / PTC / DDOG / MRNA** | Avoid | Parabolic cluster (RSI 86–90, day-0/1 52w high). Weak-ticker-memory P 0% n=4 HARD AVOID direct. No chase at reference | High | — | *Pre-CPI defer rule binding — no new breakout-chase entries within 3 sessions of a high-importance print.* ## Executive summary - **Regime label is "risk_on" but posture is stand-aside.** VIX 15.25 is in the deep-unlock bucket (<17), BUT the week's sector leadership is XLE +4.05%, XLP +3.85%, XLU +3.50%, XLY +2.67% with **XLK -0.015% bottom-tier** and XLI -0.14%. Defensive rotation (energy + staples + utilities leading) with Tech lagging is a late-cycle fingerprint, not a textbook risk-on tape. Treat today's bundle as conflicted, not green-light. - **IWM lead-gap gate FAILS 5th+ consecutive week.** IWM 1w -0.52% vs SPY +1.30% = **-1.82pp** gap the wrong direction. Deep-unlock requires VIX <17 AND IWM lead ≥+1pp. Fragile-partial-unlock binding: **1/6 non-floor / 1/3 floor sizing cap**. - **Pre-CPI (T-3).** Tue 2026-10-14 is the first high-importance macro since mid-Sep FOMC crush. Pre-event defer rule: no fresh breakout-chase entries; manage existing only. Options bias to **defined-risk only** through Oct 14. - **Attribution degraded — reviewed_count=0.** The 14d parse found no graded picks from recent markdown archives. Treat today's conviction floor as lower than usual, lean further on structured picks_history calibration (54%+ recovery n=33 weekend reading, provisional-lift not yet confirmed intraweek). - **Calibration monotonicity NON-monotonic 90d n=2586.** low/medium/high = 40.77 / 37.88 / 50.00. Brier 0.2535 worse than 0.25 coin-flip. Only high-conf-watch (57% n=347) carries real signal. Medium/long 37.1% n=645 BELOW 40% floor binding — tighten single-name long entries; prefer ELITE ticker-memory (HAE 58%, MRVL 52.83%, SMH 67.5%, FTNT 63.83%) over bundle-top momentum. ## Swing setups ### SMH — VanEck Semiconductor ETF (Medium) - **Entry:** pullback to 596–600 (SMA20 upper = 576.95, 50-day ~ 577, but current 607.27 so wait for closer-to-SMA retest). Primary buy zone = 596–600 on 5min reversal + rvol ≥ 1.0×. - **Reference:** 607.27 close (day-76 from 52w high 671.83). - **Stop:** 584 (below SMA50 577 + ~1pt buffer; -3.8% from 600). - **Target:** 640 (prior Aug consolidation top), then 665. - **Why:** SMH ELITE ticker-memory **67.5% n=40, mean_excess_5d +1.72%** = highest-signal single feature in today's bundle. Non-parabolic (day-76 from 52w high, RSI 64.96 sweet-spot). Semiconductor correlation note: SMH and ENTG at 0.87 — treat as one slot in the semi complex. - **Invalidation:** XLK breaks below 1d low on 1.5× volume AND SMH < 584 OR SMH 5min close < 596 after pullback. ### MRVL — Marvell Technology (Medium) - **Entry:** pullback to 265–268 (SMA20 zone, 50-day = 234.65 well below). - **Reference:** 274.66 close (day-87 from 52w high 329.88, deeper pullback setup). - **Stop:** 254 (below SMA20; -5% from 268). - **Target:** 305 (prior-high reclaim), then 320. - **Why:** MRVL ticker-memory **52.83% n=53 mean_excess_5d +2.88%** = strongest validated single-name mean. RSI 68 upper sweet-spot boundary. 4/4 earnings beats avg +1.4% (decelerating but structural). Analyst upside +23%. Insider net-buying. - **Invalidation:** SMH < 584 OR MRVL 5min < 263 after pullback OR gap-down > 2% at open with no fade. ### FTNT — Fortinet (Low) - **Entry:** break > 192.50 on 1.25× rvol + 5min hold + XLK 1d green. - **Reference:** 189.10 close (day-2 from 52w high 192.41). - **Stop:** 180 (below 20-day; -6.5% from 192.50). - **Target:** 205, then 215. - **Why:** FTNT ticker-memory **63.83% n=47 mean_excess_5d +0.49%** ELITE cybersec. 4/4 beats avg +19.8% accelerating. BUT day-2 from 52w high + RSI 76 = post-parabolic region → confidence Low, breakout trigger must gap ≥1.25% above session high per post-FOMC T+1 chop rule. - **Invalidation:** session high rejected + 5min close < 189. ### HAE — Haemonetics (Low, pullback-only) - **Entry:** pullback to 112–114 SMA20 + 5min reversal + XLV 1d green + rvol ≥ 1.0×. - **Reference:** 119.47 close (day-0 at 52w high 121.00, RSI 70.71). - **Stop:** 106 (below 100-day equivalent; -7% from 114). - **Target:** 130, then 140 (analyst mean 118 just reached — upside slim at reference). - **Why:** HAE ticker-memory **58.06% n=31 mean_excess_5d +2.17%** ELITE healthcare. 4/4 beats accelerating. XLV +1.18% 1w (not top-tier but not blocked). BUT **day-0 52w high + RSI 70.7 = post-parabolic-no-chase binding** — pullback entry only, never chase at reference. Analyst target already met. - **Invalidation:** XLV turns 1w negative OR HAE 5min close < 110 after pullback. ### HARD-AVOID cluster (High conviction avoid) Parabolic names in today's top-25 — do not initiate direct longs regardless of catalyst. Direct shorts prohibited at VIX 15.25 < 17.5 dead-zone (VIX-gate binding, calibration floor binding on direct shorts at conf ≥ medium). - **P (Everpure)** — RSI 89.6, day-1 52w high, ticker-memory **0% n=4** (worst in table). HARD AVOID direct. Watch-only re-entry criteria: pullback ≥10% + RSI < 60 + weekly reclaim. - **PTC** — RSI 88.99, day-243 50-day base-break (not fresh 52w high but RSI parabolic). HARD AVOID chase. - **DDOG** — RSI 86.83, day-46 from 52w high, ticker-memory 45.45% n=22 mean_excess_5d **-3.69%** (consistent underperformer). HARD AVOID. - **MRNA** — RSI 69.96, day-7 from 52w high + 3m +189%. Ticker-memory 50% n=42 BUT mean_excess_5d skewed by a single +130% spike; downside asymmetry extreme. Analyst upside -38.58% (target 121 vs spot 197). AVOID direct. - **PLTR / DT / SNPS / NTAP / OPCH / MAT / FN** — all firing RSI > 76 with day-0 to day-3 52w-high proximity. AVOID chase at reference; re-enter only via pullback-to-SMA20 + volume-reclaim. ## Long-term ideas ### HUM — Humana (Low, watch) - **Entry:** reclaim weekly close > 395 + XLV weekly green. - **Reference:** 387.12 close (day-19 from 52w high 428.88, RSI 50.4 mid-range). - **Stop:** 360 (below SMA200 at 296 but above recent swing; -7% from 390). - **Target:** 430 (52w high retest), then 465. - **Why:** Managed-care Medicare Advantage long-term thesis. 4/4 beats avg +6.3%. Analyst upside +11.3%. RSI 50.4 = non-extended, mid-range accumulation structure. Insider net-buying. Defensive-sector (XLV healthcare) with sector 1w +1.18% constructive but not top-tier. - **Invalidation:** XLV 1w turns ≤ -1.5% OR HUM weekly close < 365. ### ACN — Accenture (Low, watch) - **Entry:** pullback to 200–204 (near SMA200 194.44 but above) + 5min hold + rvol ≥ 1.0×. - **Reference:** 208.36 close (day-184 from 52w high 291.09 — deep drawdown long-term reaccumulation). - **Stop:** 192 (below SMA200; -7.5% from 208). - **Target:** 240, then 275. - **Why:** IT services long-term reset after 52w-high puke. Value score 0.418 top-of-bundle. RSI 65.35 constructive but non-parabolic. Analyst upside +7.3%. 3/4 beats accelerating. Day-184 pullback = structural (not parabolic). - **Invalidation:** weekly close < 192 OR XLK 1w drops below -3%. ## Day-trade watchlist (levels only — no time-of-day calls) Pre-CPI (T-3) defer rule binding. No fresh breakout-chase entries unless the setup is a pullback-reclaim with triple-confirm (SMA20 + rvol + sector green). Manage prior positions only. - **SMH** (ETF proxy for semi complex): watch 596–600 pullback zone for 5min reversal. Breakdown level 586 would signal failed support; wait for 5min close + rvol confirmation before shorting via QQQ puts (direct equity shorts prohibited VIX-gate). - **QQQ**: 1d -1.34% weakness. Watch 736–740 for pullback entry on OPEX week; below 730 is sector-wide short-side signal (expressed via defined-risk put debit spreads only). - **IWM**: 1w -0.52% underperforming SPY by 1.82pp. Watch 274 (prior-week support) — a close < 274 would confirm small-cap breakdown; expressed only as IWM Oct 24 272/266 put debit spread (not direct short). - **MRVL**: 265–268 is the pullback bid zone; a 5min reversal + rvol ≥ 1.0× is actionable. - **FTNT**: 192.50 break level; day-trade entry requires second 5min close above + 1.25× rvol per post-parabolic chop rule. Deferred intraday chases: PLTR, DT, SNPS, DDOG, OPCH, HAE at reference, MRNA — all parabolic + pre-CPI. ## Options ideas **Sanctioned bucket**: medium/options 36.36% n=44 mean_excess_5d -0.20% is thin-edge; must gate on defined-risk and VIX < 17 (satisfied at 15.25). ### QQQ Oct 24 740/750 call debit spread (Low, watch) - **Entry:** pullback to 736–738 on QQQ + rvol ≥ 1.0× + VIX < 17 daily close + no CPI gap-down Mon Oct 13 ADD. - **Reference:** QQQ 747.58 close. - **Max debit:** 3.30 (33% of 10-wide spread cap). - **Target:** 740/750 fully in-the-money at Oct 24 expiry. - **Stop:** debit halved OR QQQ weekly close < 730. - **Why:** QQQ post-FOMC-crush continuation setup conditional on surviving Oct 14 CPI without a > 1% gap-down. Defined-risk expression per VIX-gate; XLF-sleeve DOUBLE-BLOCKED 10+ consecutive briefs so QQQ is the sanctioned alternate. - **Caveat:** this is a conditional. Enter only AFTER CPI (post-event confirmation). Pre-event entries are prohibited per pre-event defer rule. ### XLF Oct 24 54/56 call debit spread — DOUBLE-BLOCKED 11th consecutive brief XLF 1w +1.44% (just above flat-to-positive gate) BUT **1m -4.96%** fails the +0.4% 1m gate. Sleeve stays DOUBLE-BLOCKED. **Approaching the 12-week escalation threshold** (added 2026-09-20) — one more brief in this state and the sleeve requires formal regime-review before re-arm. ### Hedge expression (defensive) If late-cycle defensive sector rotation (XLE/XLP/XLU top + XLK bottom) persists through midday, consider QQQ Oct 24 740/730 **put debit spread** at max debit 3.30 — defined-risk long-vol hedge against the IWM lead-fail + pre-CPI vol run-up. Entry gated on QQQ 5min close < 744 AND VIX > 15.5 intraday. Not initiated at open; waiting on confirmation. ## Risks / invalidation - **CPI Tue Oct 14 (T-3)** — any surprise > ±0.15pp vs consensus will drive a ≥1% SPX move. Pre-event defer binding; no new breakout-chase entries. - **Defensive sector rotation signal.** XLE/XLP/XLU/XLY are the top-4 weekly performers with XLK/XLI at the bottom. This is **late-cycle rotation signature** — a tactical defensive shift inside a nominally risk-on vol regime. If XLK fails to recover above 1w flat by end of Fri Oct 10, treat the week as **regime-flip-pending** and tighten exposure. - **IWM lead-gap -1.82pp = 5th+ consecutive week** failing +1pp gate → fragile-partial-unlock sizing cap (1/6 non-floor / 1/3 floor) binds all new entries. - **Attribution block degraded** (reviewed_count=0). Lower conviction floor on single-name new entries. - **Calibration non-monotonic** — low/med/high = 40.77/37.88/50 is not a signal-bearing confidence ladder; defer to ELITE ticker-memory rank + predicate match over conviction label. - **High-conviction long bucket calibration floor-fail: 0.0% n=8** — do NOT issue a single-name HIGH/LONG today. (None issued above.) - **FOMC Oct 28 (T-13)** + **GDP Oct 29** + **ISM/NFP window Nov 2–6** — multi-event stack 3 weeks out; keep sizing locked. --- Disclaimer: This newsletter is for educational and informational purposes only. It is not financial advice, not a solicitation to buy or sell any security, and reflects algorithmic screens and model-written commentary that may be wrong. Do your own research. Past performance does not guarantee future results.
Armed to trade 6 stocks (from 2026-10-09 — markets have been closed since, so this is the most recent plan, not today's): HPE, FTNT, P (betting it falls), MRNA, DOCS, HUM. It only buys if a price trigger is hit.
The things it looks at most when deciding what to bet on:
The robot hasn't settled on what matters yet.
It hasn't changed its mind much yet.
It does better in some market moods than others — and it knows it:
Not enough days watched yet.
28 bets in the latest run, 8 of them experiments to learn new things.
So far its experiments beat the market 45 / 100 vs 37 / 100 for its confident bets — measuring that gap is how it learns what actually works.
It has made 305 trades so far — 69 still open and 236 finished. On the finished ones it has lost $19,122 for real; its open positions are −$1,709 on paper.
Showing the 30 most recent.
Exact sell prices settle at the broker, so closed trades show how they ended, not a per-trade dollar figure.
Every day it writes down what it is learning, then later checks whether it was right. So far it has confirmed 8 ideas that genuinely beat the market and flagged 8 that didn't pan out — and it is dialing those back.
The notes below are the bot's own raw jottings — technical, but it's all here if you want to dig in.
Armed to trade 6 stocks (from 2026-10-09 — markets have been closed since, so this is the most recent plan, not today's): MRVL, DDOG, AMD, FTNT (betting it falls), CEG, SBUX. It only buys if a price trigger is hit.
The things it looks at most when deciding what to bet on:
The robot hasn't settled on what matters yet.
It hasn't changed its mind much yet.
It does better in some market moods than others — and it knows it:
Not enough days watched yet.
27 bets in the latest run, 6 of them experiments to learn new things.
So far its experiments beat the market 41 / 100 vs 40 / 100 for its confident bets — measuring that gap is how it learns what actually works.
It has made 368 trades so far — 75 still open and 293 finished. On the finished ones it has lost $15,264 for real; its open positions are +$749 on paper.
Showing the 30 most recent.
Exact sell prices settle at the broker, so closed trades show how they ended, not a per-trade dollar figure.
Every day it writes down what it is learning, then later checks whether it was right. So far it has confirmed 4 ideas that genuinely beat the market and flagged 8 that didn't pan out — and it is dialing those back.
It also caught itself 3360 times trying to trade stocks too illiquid to fill safely — its plan is to stop recommending names that keep failing that check.
The notes below are the bot's own raw jottings — technical, but it's all here if you want to dig in.
Armed to trade 6 stocks (from 2026-10-09 — markets have been closed since, so this is the most recent plan, not today's): MRNA, FTNT, NTAP, MRVL, P (betting it falls), HSIC. It only buys if a price trigger is hit.
The things it looks at most when deciding what to bet on:
The robot hasn't settled on what matters yet.
It hasn't changed its mind much yet.
It does better in some market moods than others — and it knows it:
Not enough days watched yet.
0 bets in the latest run, 0 of them experiments to learn new things.
No bets placed in the latest run.
Not holding any stocks right now — waiting for one of its triggers to hit.
It hasn't placed any trades yet.
No open trades right now.
No closed trades yet.
Every day it writes down what it is learning, then later checks whether it was right. So far it has confirmed 0 ideas that genuinely beat the market and flagged 0 that didn't pan out — and it is dialing those back.
The notes below are the bot's own raw jottings — technical, but it's all here if you want to dig in.
Nothing logged yet.
Nothing logged yet.
Nothing armed to trade at the moment.