Hansel
← The village

No. 04 · The exchange

AI Investing

An AI reads filings like Buffett and values quality companies. It buys only when the price is right. This page logs every trade and judgment daily.

Status
Open
Account
Paper money, at Alpaca
Benchmark
SPY total return
Updated
Snapshot of Oct 23 close
The exchange in the gingerbread village

+7.47%

AI portfolio since Sep 28, paper money

−0.74%

S&P 500 (SPY) same period

+8.21 pp

Ahead of SPY

−0.85%

Worst drop, Oct 22 to Oct 23

44%

In cash, waiting for prices

19 days

Trading days live

Learnings: 8 logged, 6 fixed See all11 trades, each tied to its judgment

Exhibit 1Explore · pick a range, tap a day

Ahead of the S&P 500 since the first trade

Account value vs SPY total return, both set to 100 at the Sep 28 close.

  • AI portfolio
  • SPY total return

AI portfolio +7.47% versus SPY −0.74% since Sep 28, across 19 trading days. The chart needs JavaScript to draw.

How to readGreen fill means ahead of SPY, red means behind. The lower strip shows the cash share.

NotePaper money. Not investment advice.

SourceKernel nav table, Alpaca paper fills.

Daily log

Tap a day to mark it on the chart
Oct 23AI −0.85% · SPY −0.52%
  • No trades today.

  • Judged 11 companies; 0 cleared the 8% bar.

Oct 22AI +2.39% · SPY −1.18%
  • New judgments: 16. None cleared 8%. Best: CME at 6.5% a year.

Oct 21AI −0.18% · SPY +0.53%
  • New judgments: 8. None cleared 8%. Best: MSCI at 6.6% a year.

Oct 20AI +1.35% · SPY +0.34%
  • New judgments: 5. None cleared 8%. Best: MMM at 4.0% a year.

Oct 19AI +0.04% · SPY +0.42%
  • New judgments: 1. None cleared 8%. Best: COST at 3.1% a year.

Oct 16AI −0.66% · SPY −1.11%
  • New judgments: 3. None cleared 8%. Best: MRSH at 6.4% a year.

Oct 15AI +2.03% · SPY +0.28%
  • New judgments: 5. None cleared 8%. Best: CVS at 4.0% a year.

  • Learning

    The emergency stop could report done while orders were still live.

Oct 14AI +0.79% · SPY −1.08%
  • New judgments: 1. None cleared 8%. Best: PG at 5.3% a year.

Oct 13AI −0.07% · SPY −0.01%
  • New judgments: 1. None cleared 8%. Best: SYK at 5.9% a year.

  • Learning

    Book and broker share counts could drift apart unnoticed.

  • Learning

    Book fills record about 5 bps worse than the broker's real fills.

  • Learning

    The paper broker never pays dividends.

Oct 12AI +0.33% · SPY −0.06%
  • New judgments: 5. None cleared 8%. Best: CI at 7.0% a year.

Oct 9AI +0.29% · SPY +0.60%
  • New judgments: 2. None cleared 8%. Best: EOG at 4.0% a year.

  • Error fix

    Sold back 29 BKNG shares bought in error on Oct 8. Fixed Oct 15 (v0.25.8).

  • Learning

    Bought twice the intended BKNG position, then trimmed it the next day.

Oct 8AI +1.11% · SPY −0.42%
  • New judgments: 2. None cleared 8%. Best: GD at 5.1% a year.

  • Bought

    Bought BKNG, 63 shares. The price had fallen under its buy-up-to price of $156.44 ().

  • Dividend

    INTU dividend: $1.38 a share on 34 shares, $46.92. The book credits it on the ex-date. The paper broker pays none.

Oct 7AI +0.54% · SPY −0.24%
  • New judgments: 9. None cleared 8%. Best: MCK at 7.0% a year.

Oct 6AI +0.08% · SPY +0.55%
  • New judgments: 3. None cleared 8%. Best: SYK at 6.3% a year.

Oct 5AI +0.45% · SPY +0.67%
  • New judgments: 4. None cleared 8%. Best: DRI at 6.5% a year.

  • Part 2 of 2

    Bought RMD, part 2 of 2 of the Oct 2 order: 43 shares. .

Oct 2AI −0.65% · SPY +0.74%
  • New judgments: 4. None cleared 8%. Best: ORLY at 5.0% a year.

  • Bought

    Bought ADBE, 39 shares. at 9.0% a year, over the 8% bar.

  • Added

    Added 17 INTU shares. This is the one top-up its allows.

  • Part 1 of 2

    Bought RMD, part 1 of 2: 42 shares. One order is capped at 10% of the account. .

  • Learning

    ADBE's buy and an INTU top-up went out Oct 2, days late.

  • Learning

    RMD's Oct 2 order filled at $223.53, above its $222.10 buy-up-to price.

Oct 1AI +0.34% · SPY +0.18%
  • New judgments: 5. 1 cleared 8%: CI. CI cleared 8%, but conviction was 2 of 5. Buying needs 3.

Sep 30AI −0.07% · SPY −0.21%
  • New judgments: 3. None cleared 8%. Best: CBOE at 6.0% a year.

  • Added

    Added 48 PAYX shares. This is the one top-up its allows.

  • Error fix

    Sold back 42 RMD shares bought in error on Sep 29. Fixed Oct 15 (v0.25.8).

  • Learning

    Bought RMD on Sep 29, sold all of it the next morning, rebought Oct 2.

Sep 29AI +0.02% · SPY −0.18%
  • New judgments: 141. 4 cleared 8%: ADBE, INTU, PAYX, RMD.

  • Bought

    Bought INTU, 17 shares. at 9.0% a year, over the 8% bar.

  • Bought

    Bought PAYX, 47 shares. at 8.5% a year, over the 8% bar.

  • Bought

    Bought RMD, 42 shares. at 8.0% a year, over the 8% bar.

Prices judged at the close. Orders go out next morning, ET.Download the data (JSON)
Exhibit 2Explore · tap a holding to re-test its judgment

Five holdings, almost half in cash

Each holding with its judgment, buy-up-to price and exit plan. Oct 23 close.

    Cash Weight44.2% Waits for a price that clears 8%. The best judgment today is FICO at 7.5% a year (Sep 29). Cash earns nothing in this paper account.

    How to readThe rail runs from the AI's low to high value. Left of the dashed line clears 8%.

    NoteOpen holdings are marked at the last close, so their returns are estimates. A sale makes them final.

    SourceAlpaca paper positions, kernel lots table.

    Exhibit 3Explore · hover a dot, or find a stock

    Most judgments miss the 8% bar

    Each dot is one AI valuation, as expected yearly return at that day's price.

    • 8% buy bar

    241 valuations across 149 stocks; 6 cleared the 8% bar. The chart needs JavaScript to draw.

    A judgment is a standing order. If the price falls, it can clear 8% later. BKNG did: judged Sep 29 at 7.5%, bought Oct 8 under $156.44.

    500 largest US stocks
    500
    Pass the five quality gates
    143
    Valued by the AI
    149
    Cleared 8% at least once
    6
    Held now
    5

    149 valued is more than 143 passing today. Some valued stocks have since failed a gate.

    How to readDots above the dashed line cleared the buy bar.

    NoteSep 29 is the first full pass over all survivors.

    SourceKernel v4 ledger, live arm.

    Learnings

    What went wrong, and whether it is fixed

    8 logged. 6 fixed, 2 watching, 0 open. Every error stays on the record.

    Fixed
    Changed in code. The chip names the version and date.
    Watching
    Known and measured daily. Not changed yet.
    Open
    Not fixed. Still affects the numbers above.
    DateWhat happenedLessonTradesStatus
    Sep 30Bought RMD on Sep 29, sold all of it the next morning, rebought Oct 2.A missing price was filled with the prior close. That put RMD exactly on the 8% line. The real price next morning was higher, so it sold. Cost about $115.Fixed Oct 15 (v0.25.8)
    Oct 2ADBE's buy and an INTU top-up went out Oct 2, days late.The morning step skipped orders with no usable quote, without a warning. Cause inferred from code, since quotes were not logged.Fixed Oct 14 (v0.25.5)
    Oct 2RMD's Oct 2 order filled at $223.53, above its $222.10 buy-up-to price.Buy limits had no cap at the 8% price. That slice earned 7.93% a year, just under the bar.Fixed Oct 16 (v0.28.1)
    Oct 9Bought twice the intended BKNG position, then trimmed it the next day.A 5% target in the book became 10% for the broker account. One extra round trip.Fixed Oct 15 (v0.25.8)
    Oct 13Book and broker share counts could drift apart unnoticed.Any share mismatch now blocks new orders until the two agree.No tradeFixed Oct 13 (v0.25.3)
    Oct 13Book fills record about 5 bps worse than the broker's real fills.On fills alone, broker cash runs about $40 ahead of the book. Small, tracked daily.No tradeKnown, monitored
    Oct 13The paper broker never pays dividends.The book credits each dividend on its ex-date, like SPY total return. So book cash runs ahead of the broker by every dividend.No tradeHandled in the book, gap tracked
    Oct 15The emergency stop could report done while orders were still live.Stop now holds until broker and book are both flat, confirmed.No tradeFixed Oct 15 (v0.25.7)

    How it works

    1. 1. Universe

      Top 500 US companies by market value, rebuilt every morning.

    2. 2. Five quality gates

      Return on capital of 20% or more. Earnings cover interest six times. Retained earnings add value. A $10M profit floor, and little dilution. About 140 pass.

    3. 3. AI valuation

      When a survivor files news, the AI values it from the filing.

    4. 4. Buy rule

      The AI estimates a yearly return at today's price. It buys at 8% or more.

    5. 5. Re-tested daily

      Every judgment is re-tested each morning. A lower price means a higher return.

    6. 6. Conviction sets size

      Conviction 3 buys 5% of the account. 4 or 5 buys 10%. One top-up per filing. No stock above 20%.

    7. 7. When it sells

      Not because a price rose. It sells if the business breaks, or the price passes its sell line.

    8. 8. Timing

      Prices are judged at the close. Orders go out the next morning, ET.

    9. 9. Two shadow AIs

      Two more AIs judge the same filings. One has no web search, one sees no names. They never trade.

    The cost of the rule: long stretches in cash while prices stay high. Today 44% waits in cash.

    Backtest

    Rule changes

    • Oct 16Buy limits are capped at the 8% price.v0.28.1
    • Oct 15A missing price no longer stands in for a real one.v0.25.8
    • Oct 15Order size follows the book, not the broker account.v0.25.8
    • Oct 15The emergency stop holds until broker and book are flat.v0.25.7
    • Oct 14An order with no usable quote now raises a warning.v0.25.5
    • Oct 13Any share mismatch blocks new orders until fixed.v0.25.3

    Method: returns are account value including cash, from the Sep 28 close. SPY is total return with dividends. The book also credits each dividend on its ex-date. The page never annualizes a period under a year. Paper money at Alpaca: simulated fills at real market prices. No commission on paper fills. The paper broker pays no dividends (see Learnings). Times are in ET. Not investment advice. Download the data (JSON).