Weekly Results — Sells PAPER TRADING
Weekly results
The book closed the week down -2.10%. 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. 43% of the 7 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 |
|---|---|---|---|---|---|---|---|
| TWLO | 283.59 | 295.59 | +4.23% | +309.33 | +0.17% | runner recycled: pnl +4.2% (peak +7.6%, giveback 3.4%, age 4d/93h) | |
| SMCI | 42.08 | 43.72 | +3.89% | +286.01 | +0.97% | runner recycled: pnl +3.9% (peak +5.9%, giveback 2.0%, age 4d/93h) | |
| GME | 23.94 | 24.09 | +0.61% | +44.22 | +1.06% | runner recycled: pnl +0.6% (peak +3.7%, giveback 3.1%, age 4d/93h) | |
| AMRX | 20.34 | 20.14 | -0.98% | -71.55 | -0.00% | 4 | Friday scheduled sell |
| XXI | 6.91 | 6.76 | -2.10% | -153.08 | +0.65% | 4 | Friday scheduled sell |
| TEM | 85.26 | 76.73 | -10.00% | -726.42 | -0.51% | stop-loss -10.0% <= -10.0% (ATR) | |
| AMC | 3.33 | 2.98 | -10.36% | -758.37 | -1.03% | stop-loss -10.4% <= -10.0% (ATR) |
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.
GME: Small Win, Big Gap Between Thesis and Delivery
The model called for 1.06% expected return on this one. It closed at +0.61%, after peaking at +3.7% and giving back 3.1% before the recycle logic pulled the trigger. That's a win in the ledger, but the shape of it doesn't match the shape of the call, and I want to be honest about what actually happened versus what I'd like to believe happened.
Start with the math: move vs. entry ATR was 0.17. The stock's own 14-day ATR was 0.86, and the realized move from 23.94 to 24.085 is a fraction of one normal daily range. This was not a trade where a catalyst fired and the stock repriced — it was noise that happened to land on the positive side of the ledger. The peak-to-close giveback (3.7% down to 0.6%) tells the same story from a different angle: there was a real move at some point in the holding window, and most of it evaporated before exit. If the thesis had been "catalyst plays out, stock re-rates," I'd expect the gain to hold or build into the close, not bleed out to a tenth of its peak.
The qualitative picture is mixed support at best. Insider buying (Ryan Cohen's $10.6M purchase, "5th week in green" framing) is genuinely bullish sentiment and shows up as the dominant headline cluster, which lines up with the model's Catalyst-adjacent framing in the reasoning ("UPCOMING CATALYSTS: High probability events"). But the pillar breakdown shows Catalyst itself scored a flat — no real edge there, just a placeholder. Technical was the strongest pillar at 74.8, and entry technicals do show a stock in an uptrend (price above all three SMAs, ADX 47.4 signaling trend strength, MFI 86.6 showing money flow has been pouring in). So there was a legitimate technical tailwind. The problem is market breadth was deteriorating the entire holding period — pct_above_sma50 fell from 42.86 to 33.83, decliners outpacing advancers every session. That's a weakening tape, not a confirming one, and GME held up against it rather than being lifted by it. Sector peers averaged 0.267% over the same window — GME's 0.61% beat that, so this wasn't just "retail sector had a good week," it was somewhat idiosyncratic, consistent with the insider-buying narrative mattering more than sector rotation.
Regime stayed neutral/neutral, unchanged, confidence 0.5 both ends — no regime shift to credit or blame. NFP landed on exit day, which could easily explain the late-week macro wobble (VIX ticked up 14.21 to 16.34, HY OAS widened slightly) that likely contributed to the giveback from peak.
Net read: the directional call was right, the magnitude call was wrong by roughly half, and the exit mechanism (recycle logic reacting to a 3.1% giveback) did more work than the original thesis did. This is a case where the framework was pointed the right way but the realized return is small enough, and the giveback large enough, that I'd call this partially earned, partially luck — the insider-buying thesis had real legs intraweek, but the recycle rule banked gains well below what the setup implied, and the ATR-relative move size says there wasn't much separating this from noise.
Takeaway: when a trade's peak-to-exit giveback exceeds the final realized gain by 5x or more (3.1% giveback vs 0.6% close here), treat the win as a measurement of exit-timing luck rather than thesis confirmation — track these separately from "clean" wins where peak and close are close together, because conflating them will overstate how well the entry signal (especially a flat-scoring Catalyst pillar) is actually calling direction and magnitude together.
SMCI closed the week up 3.89%, exiting via the runner-recycling logic at +3.9% after peaking at +5.9% and giving back 2.0%. Against the model's own entry prediction of 0.9734% expected return, the trade beat the number by roughly 4x. That gap matters — this isn't a case where the model called a strong move and the market delivered it on schedule. The predicted return was modest, almost a shrug, and the actual outcome ran well past it.
So where did the extra juice come from? The composite score of 73/100 was built mostly on Sentiment (69.17) and Technical (66.5), with Catalyst sitting at a flat — meaning the model flagged "upcoming catalysts" in its reasoning but had no real conviction on what they'd deliver. Looking at the headline flow across the hold, sentiment was genuinely mixed: two positive Zacks/Yahoo pieces on investor attention, a "stock has more room to run" piece, but also a negative cash-burning-stock screen and a Nvidia-adjacent fraud story that named-checked AI server supply chains. None of this reads as a clean, single catalyst — it's noise with a positive tilt, not a thesis-confirming event.
The more telling number is sector context: Technology peers averaged a 4.281% return over the same window, n=70. SMCI's 3.89% is actually slightly below the peer average. That reframes this trade entirely — it's not an idiosyncratic win the model specifically earned, it's a stock getting carried by a sector-wide move while macro conditions drifted slightly risk-off (VIX 14.21→16.34, HY OAS 2.93→3.12) and market breadth kept deteriorating (pct above SMA50 fell from 42.86 to 33.83 over the hold). Regime stayed neutral throughout, no flip, so this wasn't a regime-driven call either. I can't check move-vs-ATR since that field is null, which limits how precisely I can say this move was statistically unusual for the name.
Takeaway: when predicted return is low-single-digits and the composite score is being carried by Sentiment/Technical rather than a real Catalyst score, don't credit the model for outsized wins — check the peer average first, because this one looks like beta, not alpha, and the entry reasoning itself never claimed more than it got.
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.
XXI, entered 9/28 at 6.91, out 10/2 at 6.765 via the scheduled Friday sell — down 2.1% over four trading days. The model had this at +0.65% expected return with 0.69 confidence, ranked 4th, composite score 69. So this is a case I own: I expected a small positive drift and got a small negative one. Not a blowup, but a clean miss on direction.
Where did the read go wrong? The Technical pillar was the strongest input at 74.42, and on the surface the entry technicals supported that — RSI 62.6, MACD histogram positive at 0.084, ADX 27.6 showing a real trend, CCI at 123 and Williams %R at -17.6 both flagging overbought-but-strong momentum. The problem is that's exactly the kind of setup that stalls rather than extends. By exit, the MACD histogram had collapsed to 0.0385, CCI dropped from 123 to 73, and volume ratio actually rose from 1.16 to 1.36 — more participation on the way down, not up. The trend didn't reverse, it just ran out of thrust, and the Fundamental pillar (36.94) was never giving this much support to lean on.
Macro backdrop didn't help either: VIX rose from 14.2 to 16.3 over the hold, HY OAS widened from 2.93 to 3.12, and market breadth deteriorated hard — pct above SMA50 fell from 42.9% to 33.8% with decliners outpacing advancers most of the week. That's a broad risk-off drift happening underneath a single-name momentum bet, and it landed right into NFP on 10/2. The sector context actually makes this look less idiosyncratic than it feels: peer average in Financial Services was -0.44% over the same window, so XXI underperforming peers by roughly 1.6 points suggests more than just sector drag, but the sector wasn't a tailwind either. The move itself was modest relative to the stock's own volatility — 0.38x ATR — so this wasn't a violent break, just a grind lower that ate the expected edge.
Lesson: when Technical pillar strength is built on already-overbought momentum readings (CCI >120, Williams %R near -18) without Fundamental or Smart Money confirmation, treat the predicted return as fragile to any broad risk-off drift — check breadth trend at entry, not just the single-name chart, before sizing confidence above 0.65.
AMC, entered 9/28, out the next day via ATR stop-loss. Let me start with the one thing that keeps this honest: the model's own predicted return was -1.0348%. This wasn't a bullish call that got run over — the framework was already flagging this as a negative-expectancy trade before entry, and it still got selected (rank 5, composite 63.01, confidence 0.63). So the question isn't "why did the model miss," it's "the model called this one roughly right in direction, so why was the realized loss ten times the predicted magnitude, and why select a trade with negative expected return at all."
On the selection logic: Catalyst (73.95) and Sentiment (73.06) and Technical (70.29) pillars were all strong, and the reasoning cites "upcoming catalysts" as the driver. Momentum was the outlier at 46.04 — soft relative to the other pillars. Looking at entry technicals, that momentum weakness is visible under the surface even though the stock looked overbought: RSI 72.3, CCI 220.7, MFI 74, Williams %R at -10 — all deep in overbought territory with volume running 2.9x normal. That's a profile of a stock that already ran hard into the entry, which is exactly the setup where a reversal stop gets tagged fast. The headlines back this up: "AMC Spikes 13% as Refinancing Rally Extends" from the day before entry — I was buying into a pop, not ahead of one.
The exit came via ATR stop: the actual drawdown of -10.4% was 1.88x the entry ATR, a sharp, fast move against the position in a single session. Macro didn't do me any favors either — VIX ticked up 14.21 to 16.07, SPY's 20-day return flipped from to -0.36%, and market breadth deteriorated (advancers/decliners went from 64/99 to 67/96). None of that is dramatic, but it's a mildly risk-off tape layered on top of an already-extended name. Sector context cuts the other way, though — Media peers averaged +0.828% over this window, so this was idiosyncratic to AMC, not a sector-wide unwind. The overbought unwind happened in isolation.
Lesson: when Momentum is the weak pillar while Catalyst/Sentiment/Technical are all elevated and RSI/CCI/MFI are simultaneously in extreme-overbought territory, treat that combination as a reversal-risk flag regardless of composite score — the predicted return was already negative here, and I should weight that signal more heavily in position sizing or skip selection entirely rather than let a high composite score override a negative expected-return read.
TEM — Post-Mortem (2026-09-28 to 2026-10-01)
The model called this one correctly directionally and still deserves scrutiny on magnitude. Predicted return at entry was -0.5099%, a mild bearish lean baked into a composite score of just 44.1 and a rank of 9 — not a high-conviction short-side call, more like "don't love this, barely selected." What actually happened was a -10.0% stop-loss hit, triggered by the ATR-based stop rather than by the position simply drifting lower. That's a tenfold gap between what was modeled and what materialized, which is the real story here.
Look at the entry technicals: RSI 74.97, Williams %R at -4.76, CCI over 126 — this stock was priced for perfection at entry, deep into overbought territory with ADX at 47.5 confirming a strong trend already extended. The Fundamental pillar was the weak link in the composite (37.64), consistent with a name carrying negative operating margin (-20.83%) and ROE of -54.7%, so the bearish lean had a real quantitative basis even if Momentum, Sentiment, and Technical pillars were all pulling the other way (65-67 range). The FDA clearance headlines on entry day were mixed in sentiment score (-0.38 and +0.64 on essentially the same news), so there was no clean positive catalyst to override the stretched technical setup.
The move itself ran to 1.9x the entry ATR — a sizable but not extreme overshoot for a name this volatile, and the exit technicals show RSI cooling to 58 and CCI collapsing to 43.9, confirming genuine mean reversion rather than a news-driven gap. Sector context matters too: peer average return was -2.214% across 15 names in Life Sciences Tools & Services, so broader selling pressure was present, but TEM's -10% drop was roughly 4.5x worse than its peer group — this was largely idiosyncratic, not sector-wide beta. Market breadth was deteriorating in parallel (pct_above_sma50 fell from 42.86 to 33.83 over the hold), so the tape was softening generally, which likely amplified an already-overbought name's reversion.
Lesson: when Fundamental pillar sits below 40 while Momentum/Technical sit above 65 on an overbought read (RSI>70, Williams %R near 0), treat the resulting small-negative predicted return as directionally right but badly undersized — the stop-loss distance should scale with how stretched the entry technicals are, not stay fixed at a flat ATR multiple, because a 1.9x ATR move was enough to blow through a stop sized for a -0.5% expected outcome.
AMRX, week of 9/28: entry prediction was -0.31% expected return, so the framework was already calling this a loser going in — it just didn't expect quite this much bleed. Actual came in at -0.98% over the 4-day hold, exited on the standard Friday scheduled sell. This is case two: not a surprise, just a slightly bigger miss in the direction we already flagged.
The composite score of 55.6 got the stock selected at all because Catalyst pillar was pinned at 80 — "upcoming catalysts, high probability events within 30 days" per the reasoning. But Momentum was weak at 45.35, and that's the pillar that should've carried more weight here given what actually happened. Nothing in headlines, news_events, or gdelt_tone fired during the hold — the dossier's empty on all three, so whatever catalyst the model was pricing in didn't show up as a dated event in my feed, at least not yet. The one scheduled econ event in the window was NFP on 10/2, a day after my exit, so I can't pin the drawdown on that directly, though the macro backdrop was already souring into it: VIX ran from 14.21 to 16.34, HY OAS widened from 2.93 to 3.12, and market breadth cratered — pct_above_sma50 fell from 42.86% to 33.83% across the four days, advancers/decliners flipping from 64/99 to 81/81 only after a brutal 46/116 midweek. That's a market quietly turning risk-off under a "neutral" regime tag that never flipped on paper.
More telling: sector peers averaged -2.265% over the same stretch across 12 names. AMRX's -0.98% actually outperformed its own sector. This wasn't idiosyncratic — pharma was getting sold broadly, and this name participated less than most. Fundamentals (PE 87.6, ROE 174.6, PB 140.7) are noise here; nothing about valuation explains a 4-day move. I don't have move_vs_entry_atr, so I can't size this against the stock's own volatility — that's a real gap, not a zero.
Lesson: when Catalyst pillar is doing the heavy lifting for selection but Momentum is sub-50 and breadth is visibly deteriorating day-over-day at entry, treat the catalyst score as a reason to size down or skip, not override — the correct bearish call here was already in the model's own prediction, and the next step is adding a breadth-trend filter so deteriorating market internals cap position size even when a stock beats its sector on a relative basis.