Equity committee

A team of nine software agents runs a $100,000 practice portfolio of undervalued US companies. Six follow fixed rules, each with one job, and together they make every decision. Three are AI agents built on Claude: they review each decision, explain why each company was picked and cast their own vote on risk.

When an AI agent disagrees, the rule-based decision stands and the disagreement is recorded. Two of the rule-based agents work every night. The other four work only on the night before the monthly trade, the one morning a month anything is bought or sold.

No real money. Positions and prices are simulated. The portfolio started at $100,000.

Portfolio value

updated nightly; the dashed line is the S&P 500 (SPY) over the same period

book
buy-and-hold SPY

portfolio valuecash plus the value of everything held

$132,819

+$32,881 · +32.90% since inception

vs SPY

+27.90%

portfolio +32.90% · SPY +5.01% · comparison only

fall from peakworst fall from a peak

0.00%

buying halts at −25%

holdingscompanies held right now

as of 2026-10-11

Holdings

shares move with prices between monthly trades

11 companies. The largest is VLO at 13.1% of the portfolio, then MCK at 12.8%. These shares shift as prices move, until the next monthly trade.

VLO13.1%19.24%

MCK12.8%6.92%

FCX9.4%2.29%

ADP9.4%4.58%

DIS9.2%0.99%

GD8.9%9.50%

CRM8.9%8.03%

TGT8.8%6.87%

PCAR6.6%12.19%

NKE6.6%2.02%

TXT6.1%4.82%

Holdings with unrealised profit and loss
ticker	weight	entry	last	change	value
VLO	13.1%	363.76	433.75	up 19.24%	$17,327
MCK	12.8%	878.07	938.84	up 6.92%	$17,025
FCX	9.4%	75.42	73.69	down 2.29%	$12,539
ADP	9.4%	283.89	270.88	down 4.58%	$12,503
DIS	9.2%	106.99	108.05	up 0.99%	$12,267
GD	8.9%	366.03	331.27	down 9.50%	$11,891
CRM	8.9%	249.15	229.13	down 8.03%	$11,886
TGT	8.8%	143.76	153.64	up 6.87%	$11,680
PCAR	6.6%	123.04	108.04	down 12.19%	$8,789
NKE	6.6%	35.42	34.71	down 2.02%	$8,754
TXT	6.1%	91.54	87.13	down 4.82%	$8,158

How the agents decide

from the nightly screen to the monthly trade

every night, 21:30 New York · rule-based agents

01market monitorchecks whether the US stock market is rising or falling, using the S&P 500 and its long-run average
02analystscreens about 500 US companies for cheap, healthy businesses, using only figures that were public that day

These two keep the daily report current with the market's direction and a watchlist of candidates. They never place a trade.

the night before the monthly trade · rule-based agents

03strategy testerre-runs the fixed strategy on recent years of market history to check it still behaves
04reviewerstops the process if those results miss fixed limits on return, losses or the number of companies
05portfolio managerdecides how much goes into each company, within set limits per holding and per sector
06risk monitorvotes to approve or reject the finished plan

if the risk monitor rejects

First and second rejection: back to the portfolio manager for a new plan.
Third: back to the strategy tester, with stricter settings.
Fourth: the process stops and nothing trades that month.

if approved

The plan waits for a person to approve it. Nothing is bought or sold overnight.

first trading day of the month, 08:30 New York

The approved plan is carried out at simulated prices, paying realistic trading costs. This is the only step that changes the portfolio.

AI agents, built on Claude · after the decision

AI reviewerwrites a second opinion on the night's plan
AI analystwrites a short case for each company picked
AI risk reviewercasts its own vote to approve or reject

These three read the finished plan and respond in writing. They cannot change a holding, block a trade or stop the process. If the AI risk reviewer disagrees with the risk monitor, the rule-based vote stands and the difference is recorded.

A worked example

the night of 17 July 2026, one company

21:30, 17 July 2026
the market monitor checks which way prices are heading
The S&P 500 (SPY) closed at 743.29, above its long-run averages, so the market was rated as rising, with 0.719 confidence. This sets how much of the portfolio can be invested, before any company is looked at.
then
the analyst screens about 500 companies
Each one is scored on how cheap it is against what it earns, and on the quality of the business. Expand Energy came out near the top: it earned 14.8% relative to its price, generated free cash worth 9.8% of it, traded at 8.2 times earnings, and scored 9 out of 9 on a standard nine-point financial health test. It had also fallen 16% over the previous year, which is what made it cheap.
then
the strategy tester replays recent history
This step replays the whole rulebook over past data to check it still behaves. It returned 22.4% a year, with a return-per-unit-of-risk score of 1.21 and a worst fall of 19%.
then
the reviewer checks the results against fixed limits
Return, worst loss and the number of eligible companies all had to clear fixed limits. They did, so the cycle continued. Had any of them missed, the night would have ended here with no proposal.
then
the portfolio manager sizes the position
Expand Energy was one of the five strongest picks, and each of those got $11,142.90. The other four got $8,571.40 each. No single company may exceed its limit, and no sector may exceed 30%.
then
the risk monitor votes, and raises a flag
It approved the plan and recorded a flag: Basic Materials made up 20% of the portfolio across two companies that tend to move together. That was inside the limit, so the flag was only recorded. A rejection would have sent the proposal back to be rebuilt.
then
the AI agents could not respond
All three AI agents returned UNAVAILABLE that night because the account that pays for them was out of credit. They sit outside the decision path, so the process finished and the trade went ahead unchanged.
then
a person approves it
Nothing is bought or sold overnight. The plan waited for approval before any trade was allowed.
11:38, 18 July 2026
the trade is made
126.35 shares of Expand Energy at $88.19, a position of $11,142.90, part of $90,000.10 put to work across nine companies. Prices and fills were simulated. The portfolio stood at $99,937.04 at the close.

Readiness

every check must pass before real money is considered

These are the conditions this system has to meet before it would be trusted with real money. It has not met them. Until every one passes, the verdict stays NO-GO and the portfolio stays simulated.

verdictNO-GO

passedenough historyHas 0 hours of decisions on record.
not passedresearch nightsMissed 1 research night.
missing: 2026-08-02
passedlosses inside the limitDrawdown is inside the −25% buying halt.
book 130558.05 · drawdown 0%
not passedadvisory reviewsAdvisory reviews are missing or could not be read.
llm_review_memo: UNAVAILABLE, llm_risk_vote: UNAVAILABLE