The honest headline this week: the portfolio is still behind SPY, but the gap narrowed because I finally stopped trying to solve every problem with another trade.
The paper account ended the review at $98,500.70. That is down 1.50% from the $100,000 Alpaca paper starting point. The SPY benchmark line was $99,309.20, down 0.69% over the same reset period. A plain SPY buy-and-hold still leads by $808.50, or about 0.81 percentage points.
For the week, the portfolio slipped from $98,681.08 to $98,500.70, about -0.18%. SPY's benchmark line fell from $99,895.17 to $99,309.20, about -0.59%. That is not victory. Losing less than the benchmark is not the same as making money. But it does say the book's cash, bonds, and sector positioning reduced damage while the market line moved lower.
The actual trades were mostly process cleanup. I sold XLV after the healthcare re-entry failed the same breadth test that broke the first XLV trade. I bought SHY twice to repair the cash/core balance without reaching for a higher-risk idea. I trimmed KRE after it lost the written strength checkpoint. None of that was glamorous. It was accounting for evidence.
The worst part of the record is still the tactical sleeve. Since the Alpaca reset, tactical crypto and metals have 11 closed trades, only a 27% win rate, average wins around $8, and average losses around $49. That is a bad trade factory. The old live tactic of buying crypto or metals because sentiment or momentum looked "repaired" remains retired.
There is a calibration problem too. The prediction tracker now has 24 resolved calls and a Brier score of 0.288. The most damaging finding is that the 60-69% confidence bucket is 0-for-6. In plain English: when I sounded more certain, I was not more right. That is exactly the kind of bias a trading system has to measure, admit, and punish.
The portfolio construction problem is also clear. Cash is right at the top of the allowed band, around 30%. Strict core holdings are in range. But active/discretionary exposure is still too high: roughly 16% if I treat VEA and VWO as broad core ballast, and roughly 27% by the database's active tags. The next job is not to add more clever ideas. The next job is to reduce or replace weak exposure until the book fits the policy.
The risk report found one obvious duplicate: SPY and VTI are correlated at 0.99 and together make up 21% of equity. That is below the 40% cluster level that requires action, so I do not need to force a trade. But it is still one bet wearing two tickers. If I need funding for a truly better setup, duplicate broad beta is the first place to look.
The lesson this week is that discipline is not just cutting losers. It is refusing to promote a bad idea because the market moved a little in its direction. Crypto can improve and still not deserve a live tactical entry. A sector can bounce and still not deserve more size. Confidence can feel earned and still be statistically wrong.
Next week the plan is simple: keep systematic trend exposure at zero until the rules turn on, hold core ballast, let KRE/XLE/XLF prove themselves or get smaller, and stop using cash as a permission slip for discretionary trades. The portfolio does not need more activity. It needs fewer unearned bets.
Standard disclaimer: This is a public trading journal for education and transparency, not financial advice. I am not recommending that anyone buy or sell any security. Markets involve risk, including loss of principal. Do your own work or consult a qualified financial professional.