Weekly Results — Sells PAPER TRADING
Weekly results
The book closed the week up +0.66%. I will take it, but a green week is not a verdict on the process — one week is noise, and I read it as such. 44% of the 9 names worked; I care more about whether the winners were big enough to pay for the losers than about the hit rate alone. Either way, the realised dispersion feeds straight back into next week’s priors.
| Ticker | Buy | Sell | Return | P&L $ | Predicted | Days | Exit reason |
|---|---|---|---|---|---|---|---|
| SMTC | 154.82 | 180.15 | +16.36% | +1243.34 | +0.41% | runner recycled: pnl +16.4% (peak +18.7%, giveback 2.3%, age 4d/97h) | |
| SPCX | 147.98 | 151.99 | +2.71% | +205.53 | +0.87% | 4 | Friday scheduled sell |
| DELL | 521.52 | 529.87 | +1.60% | +124.84 | -0.59% | runner recycled: pnl +1.6% (peak +8.6%, giveback 7.0%, age 7d/168h) | |
| AVT | 93.02 | 93.20 | +0.19% | +14.50 | +0.16% | 4 | Friday scheduled sell |
| GRDN | 44.90 | 44.47 | -0.96% | -72.87 | +0.26% | 4 | Friday scheduled sell |
| SLDE | 25.60 | 25.11 | -1.91% | -144.10 | -0.16% | 4 | Friday scheduled sell |
| ADM | 87.87 | 86.10 | -2.01% | -155.66 | +0.57% | 4 | Friday scheduled sell |
| CRGY | 14.83 | 14.10 | -4.94% | -372.63 | +0.13% | 4 | Friday scheduled sell |
| AVAH | 14.38 | 13.64 | -5.11% | -394.83 | -0.11% | 4 | Friday scheduled sell |
Predicted is the expectation I carried in before the trade; the calibration behind it is mine and stays proprietary.
Why the winners won
Every position below made money this week. Here is what actually drove it -- and whether the framework's own entry thesis is what earned it, or whether this one worked out despite a shakier read.
No narrative available for this trade (generation was disabled, errored, or has not run yet) — the mechanical facts above are real and on file regardless.
SMTC — Post-Trade Analysis
The model's entry prediction was 0.41% expected return. The trade closed at +16.4%. That's not a case of the framework nailing its thesis — the magnitude gap is roughly 40x. Whatever the model saw at entry, it wasn't pricing in a move of this size, so the first thing to say plainly: this win doesn't validate the entry read. It validates something else, and I need to figure out what.
Looking at the composite, the selection logic leaned on "UPCOMING CATALYSTS: High probability events within 30 days" with a 69.6 composite score, Technical at 81.13 and Sentiment at 78.31 doing the heavy lifting, Catalyst pillar sitting at a flat (uninformative), and Smart Money/ML both null — so two of seven pillars simply weren't contributing signal. The prediction of 0.41% suggests the model itself didn't have strong conviction on magnitude even though it ranked SMTC #1 and gave it a 98.75 cross-sectional percentile. That's a selection call, not a return forecast, and the two diverged hard here.
The headlines tell the real story: "Semtech (SMTC) Shares Skyrocket, What You Need To Know" on 9/16, paired with "Does Semtech (SMTC) Have the Potential to Rally 36.99% as Wall Street Analysts Expect?" and a broad-tape mention in "12 Information Technology Stocks Moving In Wednesday's Intraday Session" (score 0.87). This reads like an idiosyncratic gap-up event — something specific to SMTC hit the wires around 9/16, and the stock re-rated fast. The sector context confirms it's not a sector-wide beta ride: peer average return in Semiconductors was 6.35% over the same window, while SMTC did 16.4% — roughly 2.5x the peer group. That's idiosyncratic outperformance, not sector rotation lifting all boats.
Macro doesn't explain it either. Regime stayed "neutral" through the hold with confidence sitting at a coinflip 0.5, no flip recorded, VIX ticked up to 17.71 but nothing dramatic, and there was an FOMC event on 9/16 sitting right in the middle of the move — plausible catalyst for market-wide vol, but it doesn't square with a single name tripling its peer group's return. The 9/18 headline ("Semtech: Probably Too Much Optimism," score -0.17) arriving right as the position got recycled is worth noting too — that's the market starting to question the move on the same day the runner logic took the exit. Good timing on the exit mechanism, whether or not that was foresight.
Deep stats don't help calibrate this one — move_vs_entry_atr is null, so I can't quantify how many multiples of normal daily range this represents, which is a real gap given the size of the move. The exit itself, a "runner recycled" with peak +18.7% and only 2.3% giveback, suggests the trade management captured most of the run rather than getting stopped early or riding it down — that part of the process worked regardless of the entry thesis.
Net read: this looks like the model correctly identified a name with catalyst exposure and elevated sentiment/technical scores, got the direction right, and then got bailed out by a magnitude of move the entry prediction never called. The selection process (rank #1, high percentile) earns partial credit. The 0.41% predicted return does not.
Takeaway: when predicted_return is small (near-zero to low-single-digit) but the pillars show strong Sentiment/Technical with a live catalyst flag, don't treat the size of the eventual win as confirmation the return model is well-calibrated — it isn't, by a factor of 40x here. Track how often "Catalyst: (flat/uninformative)" cases still produce outsized idiosyncratic moves; if this pattern repeats, the fix is to widen the predicted-return distribution for high-percentile, low-conviction-magnitude picks rather than crediting the framework with a correct magnitude call it never made.
# DELL: A Small Win That Doesn't Vindicate the Entry Read
The headline number is +1.6%. That's real money, but the exit reason tells the actual story: this was a runner recycled at pnl +1.6% after peaking at +8.6% and giving back 7.0% of the move over 168 hours. In other words, the trade briefly delivered a genuinely strong result and then I let more than three-quarters of the gain evaporate before flattening it. That's a meaningfully different trade than "predicted a small loss, banked a modest win" — it's "predicted a small loss, got a big move I didn't anticipate, and then rode it back down to barely-positive before exiting."
Start with the entry read: -0.5857% predicted return, composite score 63.36, confidence 0.6336. The pillar breakdown shows Technical (68.38) and Sentiment (63.64) were the strongest components, with Catalyst sitting at a flat — no real edge there per the model's own scoring at entry. The reasoning field cites "upcoming catalysts" and the composite score, but nothing quantitative in the dossier justified an +8.6% peak move. So when I ask "did the framework call this correctly," the honest answer is no — it modeled a slightly negative expectancy and got a sharp positive spike that it wasn't positioned to explain or capture.
What actually drove the move is fairly visible in the headlines cluster on September 13-14: RBC initiating coverage with an Outperform rating pushed DELL to a record high, plus positive Cramer commentary and Nscale AI-supply-chain coverage. That's a real, idiosyncratic catalyst — but it landed on September 13-14, essentially at the tail end of a nearly week-long hold that started September 7. The stock's real move was the *late* week AI-narrative reprice, not something priced in during the entry window when CPI (September 11) and mixed macro conditions (VIX 14.32 falling to null by exit, HY OAS stable at 2.65, unemployment flat at 4.1%) dominated.
Sector context makes this look worse for the model's read, not better: Technology peers averaged +4.629% return over the same window across 81 names. DELL's peak of +8.6% modestly outpaced that sector tailwind, but the final +1.6% print actually undershot it. Broader tech was moving; DELL got a stock-specific news catalyst on top of that; and the exit process gave back most of the alpha generated by both. Regime stayed neutral throughout (confidence no flip) — this wasn't a regime-driven trade, it was a name-specific news event inside a already-firm sector tape.
**Takeaway:** don't credit the entry model for this one — it predicted a loss and got an unrelated analyst-upgrade/AI-narrative spike instead, which means the win is closer to luck than correct calling. The real lesson is in the exit: a trailing/recycling rule that let peak +8.6% decay to +1.6% (a 7-point giveback) on a single stock-specific news catalyst is the bigger process issue here — next time a runner peaks that far above the sector's own average move, tighten the giveback tolerance rather than let it round-trip most of the way back to breakeven.
AVT closed the week up 0.19% against a model prediction of 0.1634% expected return. On the surface that looks like a correctly called trade — the number came in positive, roughly in the neighborhood the model expected, ship it. But I want to be precise about what "roughly in the neighborhood" means here, because the gap between prediction and outcome is actually enormous in relative terms. A predicted return of 0.1634 and a realized return of 0.0019 are off by nearly two orders of magnitude if I read the predicted_return field at face value as a percentage. Even granting some unit ambiguity in how predicted_return is scaled versus return_pct, this was not a case where the model's edge showed up cleanly — this was a trade that essentially treaded water for four trading days and got closed out on the standing Friday rule, not because any thesis played out.
The entry case leaned heavily on the Catalyst (80.0) and Technical (77.67) pillars, with a cross-sectional percentile of 88.75 — a genuinely strong-looking setup on paper. But I have no technicals or sentiment payloads populated for entry or exit, no headlines, no news events, and no earnings surprise on file, so I can't verify what catalyst was actually supposed to fire. That's a real gap: the model's own reasoning cites "upcoming catalysts" as the driver, and I have nothing in the dossier confirming one occurred during the hold.
What I can say with more confidence is that this was not a stock-specific win. Sector peers in Electrical Equipment averaged a 2.058% return over the same window — AVT badly lagged its own peer group. That reframes the small gain: it wasn't AVT participating in a sector move, and it wasn't AVT outperforming on an idiosyncratic catalyst either. The FOMC event landed mid-hold on 9/16, and macro conditions between entry and exit show VIX rising into 17.71 and market breadth deteriorating sharply on the 16th (advancers 54 vs decliners 109) before partially recovering by the 17th — consistent with a choppy, FOMC-driven week where the regime classification never flipped (stayed "neutral," confidence both ends). AVT essentially rode that chop to a flat-to-slightly-positive finish.
Given move_vs_entry_atr is null, I can't even confirm whether 0.19% represents a meaningful move relative to AVT's own volatility — it may well be inside normal noise for this name. Combined with underperforming its peer group and a predicted return that wasn't remotely matched in magnitude, this reads much more like a lucky flat exit than a confirmed thesis.
Takeaway: when predicted_return and realized return_pct diverge by this much in scale even though both are positive, don't count it as a validated call — treat "small win, badly lagging peers, no confirmed catalyst" as functionally the same signal quality as a small loss, and discount confidence in the Catalyst/Technical pillar combination until I can see the actual catalyst materialize in the news_events field rather than infer it from the score.
Why the losers lost
Every position below lost money this week. I owe a real explanation for each one -- whether it was a call I got wrong or a bearish read that played out as expected -- not a table cell.
CRGY, week of 9/14: bought at 14.83, sold at 14.10 on the scheduled Friday exit, down 4.94% over four trading days. The model wasn't calling for a big win here — 0.13% predicted return is essentially a coin-flip position dressed up with a 59.17 composite score, not a high-conviction bet. So this isn't a case of the framework seeing a clean winner and getting blindsided; it's a marginal, catalyst-driven pick that lost roughly 38x its own predicted edge. That ratio is the real story, not the miss itself.
Look at where the composite came from: Sentiment pillar at 80.15 and cross-sectional percentile at 96.25 did the heavy lifting, while Momentum sat at a mediocre 48.18. That's a red flag in hindsight — the stock was flagged mostly on sentiment and an "upcoming catalyst" framing, not on price action already confirming strength. The single headline in the file is positive (0.75 score) but it's about stocks to own "for decades," not a CRGY-specific near-term driver — thin gruel to hang 96th-percentile selection on.
Macro backdrop didn't help. FOMC landed mid-hold on 9/16, and between entry and exit VIX printed 17.71 (unknown at entry, so we can't say implied vol spiked, but realized conditions shifted), the 10y2y curve moved to and breadth cratered on the 16th — pct_above_sma50 dropped from 52.63 to 45.86, decliners outnumbering advancers 109-54 that day. That's a market-wide risk-off pulse right in the middle of the hold, even though the regime label stayed "neutral" throughout (no regime flip recorded).
The sector context makes this look more idiosyncratic than macro-driven, though: Energy peers averaged +1.895% the same week across 55 names, while CRGY fell nearly 5%. That's a wide idiosyncratic gap the dossier doesn't explain — no earnings surprise, no news events logged, no options flow data. ATR-relative move isn't available, so I can't size this against the stock's own volatility.
Lesson: when Sentiment and catalyst-probability pillars are carrying a pick while Momentum lags and peer/sector data isn't confirming direction, treat the predicted-return magnitude as unreliable regardless of composite score — cap position sizing to the confidence in the weakest pillar, not the average.
GRDN post-mortem — small loss, but the setup was already stretched
The model expected a modest gain here — 0.2575% predicted return, ranked 7th, composite 55.7 — and got a small loss instead, -0.96% over four trading days. That's not a directional disaster, but it's worth being honest about why the read was off, because the entry technicals were flashing conditions that should have made me more skeptical of a fresh long.
At entry, RSI-14 sat at 77.8, MFI at 86.3, Williams %R at -2.5, and CCI-20 at 190.8 — every overbought oscillator I track was pinned near its ceiling, with price sitting right at the upper Bollinger band (44.9 vs. 45.23 upper). Volume ratio was 3.3x normal, which the Momentum pillar (58.75) and Technical pillar (72.74) clearly rewarded, but that combination — a volume spike into an already-extended tape — is as often exhaustion as it is continuation. By exit, RSI had cooled to 63.6, MFI to 75.5, volume ratio back to 1.2x — the froth simply drained out, and price gave back a bit of the premium it was trading at. The move itself was small relative to the stock's own volatility: only 0.31x ATR, so this wasn't a violent reversal, just a mean-reversion drift against an overbought entry.
Context didn't help but wasn't the main driver either. FOMC landed mid-hold (9/16), and macro shifted from a null VIX read at entry to 17.71 at exit alongside a steepening 10y2y — a regime shakeout, though the regime label itself stayed "neutral" throughout, no flip. Sector peers actually averaged +0.98% over the same window, so this was idiosyncratic underperformance, not a Health Care rotation dragging GRDN down. No headlines, no news events, no earnings surprise on file — the dossier is quiet on catalysts, which fits: this looks like a technical unwind, not a story-driven selloff.
Lesson: when RSI, MFI, and CCI are all simultaneously at extreme highs alongside a 3x+ volume spike, treat that as a mean-reversion warning rather than a momentum confirmation — I should downweight or skip entries where three or more overbought oscillators converge at once, regardless of what the Technical pillar score says.
ADM, entry 9/14, out 9/18 on the scheduled Friday sell: -2.01% over four trading days. The model expected +0.57%, so this is case one — a miss, not a correctly-called loss. Worth being honest about how thin the conviction was to begin with: confidence 0.551, composite 55.1, rank 8. This wasn't a high-conviction long: it got in on Catalyst pillar strength (80.0) while Momentum was weak (39.39) and Sentiment barely above neutral (51.79). The reasoning field is explicit — the trade was selected on "upcoming catalysts within 30 days," not on trend or technical confirmation. That's a bet on an event that either didn't materialize as bullish or got overwhelmed by something else.
The something else is visible in the macro/regime data even though it's sparse. Entry-day VIX is null, but by exit VIX had printed 17.71 with an FOMC event landing squarely inside the hold window on 9/16. Market breadth cratered that same day — pct_above_sma50 dropped from 52.63 to 45.86, decliners spiking to 109 vs advancers — before partially recovering by the 17th. Regime label stayed "neutral" both ends, so no clean regime flip to point to, but the breadth data shows a real risk-off pulse mid-week that lines up with the FOMC date. That's a plausible transmission mechanism even if I can't prove causality here.
What stings more is the sector comparison: peers in Food Products averaged +1.468% over the same window while ADM fell 2%. That's not sector-wide drag, that's idiosyncratic — something specific to ADM (or its factor exposure) went wrong that a macro story alone doesn't explain. Headlines in the file are all late (9/18, exit day) and mildly positive/dividend-flavored — nothing there predicts a decline, and nothing bearish company-specific shows up in the dossier at all. Fundamentals (revenue growth -0.82%, thin gross margin at 6.27) hint at a low-quality setup that catalyst optimism papered over, but that's a stretch given no news_events flagged bearish. I don't have a smoking gun; the ATR-normalized move is null so I can't even size this against the stock's own typical range.
Lesson: when Catalyst pillar score dominates while Momentum and Technical pillars are mediocre (39.39 and 59.37 here), treat that as a lower-conviction setup regardless of composite score, and check sector peer dispersion before entry — a stock diverging from peers into a scheduled macro event (FOMC) inside the hold window is exactly the profile that should get sized down or skipped until this framework proves it can separate catalyst-driven names from idiosyncratic underperformers.
SLDE, week of 9/14: model called it -0.1553% expected, tape delivered -1.91%. Both negative — this is not a "the model said buy and got burned" story, it's a "the model said this was a weak setup and it was" story, just with the magnitude running about 12x larger than the read implied.
Start with what the model actually saw. Composite 57.93 got it selected (rank 10, barely), but that score was carried almost entirely by Fundamental (82.08) and Sentiment (63.57) — the balance sheet and social read looked fine. Momentum was the weak link at 42.17, and that's the pillar that matters most for a 4-day hold. Smart Money and ML pillars are both null, so on this name the model was flying without two of its usual inputs — worth flagging as a calibration gap rather than pretending the score was fully informed. The reasoning field cites "upcoming catalysts" as the selection driver, but Catalyst itself scored a flat no headlines, no news_events, no earnings surprise logged — there's no identifiable catalyst that actually fired during the hold. So the -0.1553% expected return already priced in "this isn't a strong long," and the market simply confirmed that read with more force than the model's own confidence (0.58) suggested it should.
Context backs up idiosyncratic-but-sector-flavored weakness: SLDE's peers averaged -0.521% over the same week, so the stock underperformed its own sector by roughly 3.7x. That's not sector rotation dragging it down — Insurance was soft, but SLDE was softer. Market breadth also curdled mid-week: advancers/decliners flipped from 87/76 on entry day to 54/109 by 9/16, and pct_above_sma50 dropped from 52.63 to 45.86 the same day — that's the FOMC date, and it lines up with the macro shift entry-to-exit (VIX rose to 17.71, yield curve moved to HY OAS ticked up to 2.70). Regime stayed labeled "neutral" both ends with unchanged 0.5 confidence, so nothing structural flipped, but breadth clearly wobbled around the Fed event before recovering by exit day. No ATR figure is logged, so I can't size this move against the stock's own volatility — that's a real hole in the record, not a zero.
Lesson: when Momentum sits below and both Smart Money and ML pillars are null, treat the composite score as under-informed and cut position size or hold length regardless of what Fundamental/Sentiment say — the magnitude miss here (12x expected loss) traces to trusting a score built on incomplete pillars through a scheduled macro event (FOMC) rather than to any wrong qualitative call.
AVAH — Post-Mortem, week of 9/14
The model called this one negative going in: -0.1059% predicted return, selected anyway at rank 5 with a 0.6558 composite. That's an important framing detail — this wasn't a bullish thesis that blew up, it was a marginal/negative-expectancy trade that got selected on catalyst strength (Catalyst pillar 80.0, xsec_pct 86.25) despite the model already flagging soft forward returns. Actual result: -5.11% over 4 days, exiting Wednesday's price through the scheduled Friday sell. So the sign was right. The magnitude was not — realized loss ran roughly 48x the predicted return, and the move was 1.36x the entry-day ATR (0.5415), meaning this wasn't noise inside the stock's normal daily range, it was a real trend break.
Look at the technicals and the setup was already stretched: RSI 74.5, CCI 148, MFI 70 at entry — textbook overbought on a name that had run hard (SMA20 13.43 vs SMA200 8.52, a huge stack). MACD histogram was already negative (-0.0425) at entry, widening to -0.0597 by exit — momentum was rolling over into the trade, which is likely what the model's negative return read was picking up on even as the Catalyst/Fundamental pillars (80.0, 75.39) kept it selected. Volume ratio jumping from 0.76 to 1.55 into the exit says distribution, not accumulation, on the way down.
Macro/market context doesn't exonerate the sector call, either. There was an FOMC event mid-hold (9/16), VIX moved into the record (17.71 at exit vs null at entry, so we can't compare directly), and market breadth flipped hard — advancers/decliners went from 87/76 to 54/109 by 9/16 before recovering some by the 17th. Regime stayed "neutral" throughout (no flip), so this wasn't a macro regime shift dragging the stock down. More notably, sector peers (n=41) averaged +0.924% over the same window — AVAH lost while Health Care broadly gained. This was idiosyncratic, not sector-wide beta. The positive Yahoo "momentum stock" headline on 9/17 also aged poorly against the price action already underway.
Lesson: when the model's own predicted return is negative but the trade still gets selected on catalyst/fundamental strength, treat the overbought technical readout (RSI>70, negative MACD histogram, rising volume into weakness) as the higher-priority signal for sizing or skipping, not the catalyst score — here the pillars overrode a correctly bearish quant read, and the realized loss came in ~48x larger than what that read implied.