Weekly Results — Sells PAPER TRADING
Weekly results
The book closed the week down -1.71%. I would rather own that plainly than explain it away; the question I ask is whether the loss came from the process behaving or misbehaving. 38% of the 8 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 |
|---|---|---|---|---|---|---|---|
| SEI | 70.12 | 76.74 | +9.44% | +693.46 | +0.79% | runner recycled: pnl +9.4% (peak +11.5%, giveback 2.1%, age 8d/189h) | |
| CLS | 352.91 | 381.10 | +7.99% | +584.87 | -0.16% | runner recycled: pnl +8.0% (peak +10.9%, giveback 2.9%, age 7d/164h) | |
| ANET | 205.97 | 211.08 | +2.48% | +181.21 | +0.04% | runner recycled: pnl +2.5% (peak +4.8%, giveback 2.3%, age 9d/212h) | |
| LRCX | 331.95 | 319.79 | -3.66% | -267.52 | -0.66% | 3 | Friday scheduled sell |
| ABCL | 12.97 | 12.42 | -4.24% | -311.30 | +0.20% | 3 | Friday scheduled sell |
| AVT | 103.49 | 98.67 | -4.66% | -337.40 | -1.14% | 1 | Friday scheduled sell |
| BAND | 61.38 | 55.05 | -10.31% | -753.27 | +1.64% | 2 | Broker-side exit filled mid-week (reconciled) |
| TWST | 203.00 | 181.28 | -10.70% | -781.92 | +0.73% | 0 | Broker-side exit filled mid-week (reconciled) |
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.
CLS, entry Sept 28 at 352.91, exit Oct 5 at 381.10: +7.99%. Model predicted -0.1627% at entry. This is case two — the model expected roughly nothing, maybe a touch negative, and the trade ran hard in the opposite direction. I'm not going to dress that up as a correct call. It wasn't. It was selected (rank 7, confidence 0.6185, composite 61.9) on catalyst and technical pillars, not because the return forecast itself was bullish — the predicted_return field was negative. So the sizing/selection logic and the return forecast were talking past each other here, and that's worth flagging on its own: the system bought a name its own forecast didn't like, and got bailed out.
What actually moved it: the headline stream from Oct 2-5 is thick with AI-infrastructure-expansion coverage — the "$3b AI Expansion" piece (sentiment score 0.82, run twice across sources), a ROE-screen inclusion at 0.86, and a peer read-through from Jabil's strong AI-infrastructure quarter. That's a sector narrative, not an idiosyncratic CLS print, and the sector_peer_return data backs that up: 21 peers in Electrical Equipment averaged +4.49% over the same window. CLS at +8% roughly doubled its peer group, so there's an idiosyncratic kicker on top of a real sector tailwind, not just beta. Market breadth was actually deteriorating most of the week (pct_above_sma50 fell from 42.9 to 33.8 before ticking back to 35.3), and the regime stayed "neutral" with no flip, so this wasn't a market-wide risk-on move lifting everything — it's a name-and-sector-specific AI story overriding a soft tape. No move_vs_entry_atr figure is available, so I can't size the run against CLS's own volatility, which limits how confident I can be about whether this was an ordinary swing or a genuine outlier.
Takeaway: when predicted_return and the pillar-based selection score disagree in sign, treat the predicted_return as the thing to distrust less and the selection score as the thing to audit — this trade made money despite the forecast, not because of it, and that gap shouldn't be quietly forgotten just because it closed green.
SEI — Post-Trade Analysis
First thing to reconcile: the model called for a 0.79% expected return on this name, and the actual exit print was +9.44%. That's not a case of the framework nailing its number — it's a case of the trade massively outrunning the thesis. The entry reasoning cites "upcoming catalysts" and a composite score of 60.9/100, driven mostly by Sentiment (71.98) and Technical (64.69) pillars, with Momentum and Catalyst sitting closer to neutral (52.91 and flat). That's a modest, unexciting setup — not the kind of profile that typically produces a near-double-digit move in eight days. So the question isn't "did the model call this right," it's "what actually happened that the entry read didn't capture."
Looking at what's available, nothing in the dossier points to a single dominant catalyst. The headline set is mostly noise unrelated to the company — comics previews, anime news, a salmon lawsuit — with the one directly relevant item, "SEI Group Acquires Greenhome Specialties, Inc.," scored neutral by the sentiment model, not flagged as a mover. Earnings surprise is null, options flow is empty, news_events is empty. The NFP print on 10/2 sits inside the holding window, and macro did shift during the hold — VIX up from 14.21 to 15.52, HY OAS widening from 2.93 to 3.12, yield curve steepening — but that's a modestly risk-off drift, not a backdrop that explains a stock ripping. Market breadth was actually weak through most of the hold (pct_above_sma50 fell from 42.86 to a low of 33.83 before recovering to 37.59 by exit), so this wasn't broad tape strength lifting everything either.
The sector comparison is the most interesting data point: Energy peers averaged +4.68% over the same stretch (n=49), so there was sector-wide tailwind, but SEI's +9.44% move roughly doubled the peer average — meaning at least half of this looks idiosyncratic to the name, not just sector rotation. I can't quantify the move against SEI's own ATR since that field is null, so I can't say how extreme this was in vol-adjusted terms, only that it beat the sector average by a wide margin.
Net read: the entry score correctly flagged a name with positive sentiment and reasonable technicals in a strengthening sector, which is a real signal — but the predicted magnitude (0.79%) was nowhere close to what played out, and the exit mechanism ("runner recycled" after peak +11.5%, giving back 2.1%) tells me the position was left to run past the original thesis rather than being validated by it. This is a win the framework caught the direction on but badly undersized, with actual catalyst strength possibly outside the data I have on file.
Takeaway: when predicted return is small (sub-1%) but the stock is in a sector showing strong relative performance, treat the predicted return number as low-confidence on magnitude — the directional call (sentiment + sector tailwind) may be worth trusting more than the point estimate, and position sizing or runner-management rules should account for that gap rather than assuming the entry model's return forecast is calibrated.
ANET closed out at +2.48% on a trade the model priced for a scant +4.41 basis points of expected return. That's not a case of the framework nailing a thesis — the predicted return was functionally zero, a coin-flip with a slight positive tilt, and what we got was 56x that number. This is luck wearing a win's clothing, and I want to be honest about that before I go looking for reasons to feel good about it.
The composite score of 51.19 tells the same story in different units: barely above the selection threshold. The pillar breakdown shows Catalyst (74.01) and Sentiment (65.8) doing the heavy lifting while Momentum (44.01) actively dragged the score down — the model was betting on an upcoming event, not on a stock already in motion. That's consistent with the reasoning field, which cites "upcoming catalysts" and nothing else as the selection driver. No earnings surprise is on file, no news_events fired, and the headlines during the hold are lukewarm at best — one is literally titled "Registers a Bigger Fall Than the Market," which is not what you want to see mid-hold on a position you're long. The Ciena and Nokia networking-stock pieces suggest some sector-wide interest in comms infrastructure, and the peer data backs that up hard: sector_peer_return shows the five Communications peers averaging +11.5% over the same window. ANET's +2.48% is a fraction of that — this looks like a name getting dragged along by sector tailwinds rather than winning on its own merits.
Macro backdrop was unremarkable — VIX ticked up slightly (14.21 to 15.01), yield curve steepened a bit, regime stayed "neutral" the whole hold with no flip. FOMC landed on the exit date itself, which could easily explain timing noise in the final print, but I have nothing in the dossier connecting that event to ANET specifically. Market breadth was choppy and net negative (advancers trailing decliners on 5 of 8 days), so this wasn't a rising-tide-lifts-all-boats environment either. The exit itself — "runner recycled" after a peak of +4.8% and a 2.3% giveback — tells me the position actually ran further intraday than the final number shows, then got trimmed on a trailing mechanism, not on any fresh signal.
Takeaway: when predicted_return sits this close to zero (4.4bps) and the position still gets selected and sized, the win needs to be judged against sector beta, not skill — here the peer set outran the name by nearly 5x, which argues for tightening the bar on "upcoming catalyst" selections that lack real momentum or sentiment edge, rather than crediting the model for a read it never actually made.
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.
AVT, entry 10/8, out 10/9 on the Friday scheduled sell: -0.047%. Basically flat, dressed up as a loss by rounding. Worth noting upfront because the framing matters here — the model's own predicted return was -1.1385%, meaningfully negative, and what actually printed was a razor-thin decline. So this isn't a case of the model getting fooled; it's a case of the model calling a loss and the market barely obliging. The direction was right. The magnitude was nowhere close.
Look at what fed that -1.14% call: Momentum was weak (49.07), Fundamental mediocre (55.14), and yet Catalyst scored 80 and Smart Money reasoning flagged "institutional accumulation detected" even though the Smart Money pillar itself is null — a real tension in the inputs. Confidence sat at 0.664 and the stock ranked 6th with a composite of 66.4, landing in the 81.67th percentile cross-sectionally, which means relative to the rest of the selected universe this was viewed as a stronger name despite the negative predicted return. That's an unusual combination: picked near the top of the pack, but with a bearish point estimate. The model was essentially saying "best of a mediocre bunch, still expected to lose a little."
On the macro side, nothing was screaming trouble — VIX at 15.08, yield curve positive at 0.51, HY OAS tame at 3.09, regime "neutral" at both entry and exit with no flip. Sector peers averaged a flat return over the same window, so this wasn't idiosyncratic badness or sector rotation; AVT just moved roughly in line with a non-move sector. I don't have ATR on file (move_vs_entry_atr is null), so I can't size this decline against the stock's own volatility — that's a real gap, not a zero.
No headlines, no news events, no econ events, no earnings surprise logged for this window — the dossier is thin here, and I'd rather say that plainly than manufacture a catalyst. This looks like a one-day scheduled exit on a name that didn't move much, full stop.
Lesson: when predicted return and actual return agree on sign but diverge by an order of magnitude (-1.14% called vs. -0.05% realized), treat it as a calibration flag on magnitude, not a directional failure — track whether high-Catalyst/null-Smart-Money combinations like this one systematically overstate expected downside, and stop grading these trades as "misses" when the sign call was correct and peer/sector context was flat.
TWST entered at 203.00 and exited same-day at 181.28, a -10.7% hit on a trade the model expected to return +0.73%. This is case one: the read was wrong, and wrong by a lot, not a near-miss on a bearish call. Worth owning squarely rather than hunting for an alibi.
The setup looked clean on paper. Composite score 68.94, xsec_pct in the 88th percentile, ADX at 42.5 signaling a strong trend, volume ratio 2.8x normal — all technical ingredients for a momentum continuation. The reasoning leaned on "smart money accumulation" and upcoming catalysts. But the one headline I have on file from entry day cuts directly against that thesis: Cathie Wood's ARK unloaded $18M of TWST shares that same morning. The model's Smart Money pillar is null — meaning the "institutional accumulation detected" claim in the reasoning wasn't actually backed by a populated score, and the real flow that day was selling, not buying. That's a real tension in the dossier, not a stretch.
Quantitatively, the move was large relative to the stock's own volatility — 1.49x the entry ATR of 14.62, confirming this wasn't noise, it was a genuine dislocation. Sector peers averaged 0.0% that day, so this was idiosyncratic to TWST, not a biotech-wide rotation. Macro was quiet (VIX 15.52, regime neutral, no flip during the hold), which rules out a market-wide shock as the driver. Fundamentals were already shaky going in — negative operating margin, negative ROE/ROA, negative free cash flow — so there was no cushion.
The exit itself ("broker-side exit filled mid-week, reconciled") suggests this wasn't even a clean model-driven stop — it reads like a forced or delayed fill, which may have compounded slippage beyond whatever the intraday low actually was.
Lesson: when the Smart Money pillar is null, don't let the text reasoning assert "institutional accumulation" as if it were scored — treat a null pillar as missing evidence, and cross-check it against any same-day headline flow before sizing the trade on that claim.
ABCL, entry Oct 6 to exit Oct 8 via the Friday scheduled sell, lost 4.24% over 3 holding days against a predicted return of +0.2036%. The model expected a small gain and got a real loss — not catastrophic, but a clean miss on direction, and the magnitude isn't trivial relative to a near-flat prediction.
The entry case leaned on catalyst and technical pillars — composite 78.95, Technical pillar 78.23, a cross-sectional percentile, and reasoning citing upcoming catalysts plus "institutional accumulation detected." Smart Money and ML pillars were null, so that accumulation claim wasn't corroborated by an actual scored signal in this dossier — it's asserted, not quantified. Momentum (53.65) and Fundamental (52.92) were mediocre, which in hindsight should have tempered the technical read rather than being outweighted by it.
What actually happened in the hold: RSI dropped from 56.9 to 46.5, MACD histogram went from -0.047 to -0.221 (deepening bearish divergence), ADX fell from 31.9 to 29.0 (trend losing strength, not building), and Williams %R collapsed from -57.9 to -92.7 — deep oversold, no bounce by exit. Volume ratio also decayed from 2.15x to 1.73x, so the entry's volume spike didn't sustain. The move itself was 0.68x ATR(14) — within normal daily noise for this name, so this wasn't an outlier shock, just a steady grind against the position.
Context: an FOMC event landed squarely inside the hold (Oct 7), and market breadth flipped that same day — advancers/decliners went from 116/46 to 62/100 before recovering Oct 8. Regime stayed "neutral" at 0.5 confidence both ends, no flip, so this isn't a regime-change story. But sector peers averaged -1.48% over the same window across 14 names — ABCL's -4.24% is worse than sector, so this is partly idiosyncratic, not purely sector drag. Headlines, news_events, and gdelt_tone are all empty, so I have no qualitative catalyst to point to — no single news driver stands out in what's on file.
Lesson: when Smart Money and ML pillars are null, don't let the Catalyst/Technical pillars carry a 98th-percentile selection on their own — require at least one corroborating quantitative signal before sizing conviction that high, especially heading into a scheduled macro event like FOMC during the hold window.
LRCX, entry 10/6, out 10/9 on the scheduled Friday sell: -3.66%. The model wasn't looking for a winner here — predicted return was -0.6574%, so this is case two: a bearish-to-flat call that came in bearish, just more bearish than the setup implied. The real question isn't "why did the model miss," it's "was the magnitude in line with what the signals suggested," and the answer is no — the loss ran roughly 5.6x the size the model's own expected return implied.
Breaking down the composite: confidence was 63.59, with Fundamental (76.81) and Catalyst (69.21) pillars doing the heavy lifting, while Momentum sat weaker at 52.87 — that's the tell. Momentum pillar being the laggard in an otherwise strong-scoring name is consistent with a stock that's extended (RSI 68.6, Williams %R at -12.2, CCI over 154) and vulnerable to mean reversion even with good fundamentals underneath. The entry technicals show price well above SMA50 and SMA200, classic late-cycle-in-the-uptrend positioning.
Context-wise, the FOMC date landed inside the holding window (10/7), and macro was otherwise calm — VIX ticked down from 15.52 to 15.08, yield curve and HY OAS barely moved, regime stayed "neutral" both ends, no flip. So this isn't a macro-regime story. Sector peers averaged +0.584% over the same window — LRCX was a clear underperformer relative to its 61-peer Tech set, meaning this reads idiosyncratic, not sector-wide drag. Headlines were net positive to neutral (DRAM wins, decade-return pieces), so sentiment tone doesn't explain the drop either — if anything it conflicts with price action. Move vs entry ATR was 0.55, meaning the decline stayed within less than one ATR, a modest statistical move, not a breakdown.
Lesson: when the Momentum pillar is the visible laggard against strong Fundamental/Catalyst scores, treat the composite's magnitude estimate as unreliable — the directional call can be right while sizing is off by multiples; weight position size to the weakest pillar, not the composite average.
BAND — Post-Mortem
The model called this one at +1.64% expected return, built on a 70.9 composite with catalyst and smart-money accumulation flagged as the drivers. It sold off -10.3% in two trading days instead. That's a clean miss, not a case where the setup already warned me — so let's actually dig into why the read was wrong.
Start with the move size: -10.3% against an entry ATR of 3.94 is a 1.61x ATR move in two days. That's a real break, not noise — the stock moved more than its own typical daily range would suggest for a two-day hold, which tells me whatever hit it wasn't priced in at entry.
The technicals confirm the reversal mechanically rather than explain it. RSI dropped from 56.9 to 42.7, MACD histogram flipped from +0.50 to -0.59, CCI collapsed from +65 to -59, and Williams %R went from -34 to -96 — oversold territory by exit. Volume ratio nearly doubled (0.41 to 0.90), meaning this was distribution on real participation, not a thin-volume air pocket. The "smart money accumulation" thesis that helped select this name doesn't show up anywhere in the technicals during the hold — nothing in the dossier shows what reversed it.
What I can rule out: the regime didn't flip (stayed neutral, confidence 0.5 both days), VIX actually eased (15.52 to 15.08), and market breadth improved into the exit date. So this wasn't a broad risk-off wave. Sector peers averaged -0.285% over the same window — BAND's loss was roughly 36x that, meaning this was overwhelmingly idiosyncratic, not sector-wide. The FOMC event sat right in the middle of the hold (Oct 7), which is a plausible volatility trigger given the timing, but I have no headline or news event logged to confirm it was the mechanism. Fundamentals were already shaky going in — negative operating margin, negative EPS growth, current ratio under 1 — but those were known at entry and didn't stop the model from selecting it on catalyst/smart-money grounds.
Lesson: when Fundamental pillar score is the weakest leg (32.86 here, well below Momentum/Sentiment/Technical) but gets overridden by catalyst and smart-money flags I can't independently verify in the data, that's exactly the setup prone to violent idiosyncratic reversals — I should weight pillar dispersion (strong technicals/sentiment propping up a weak fundamental score) as a red flag for position sizing, not just composite score.