PACER · Portfolio Agent for Cadence Experimentation and Reporting
Does deciding more often make you richer, or just busier? And can an AI with a research budget beat rules that never flinch? PACER runs identical stakes through different decision cadences, and one judgment model with live web access, then logs every signal, every trade, and every thesis. The verdict will be data, not vibes.
Sign in Access is invite only. Viewers see performance; trade directions stay private.
Every night, PACER pulls end-of-day market data, validates it ruthlessly (partial data means no trades, ever), computes each model's signals with zero discretion, and writes tomorrow's trade tickets. Today, a human executes them at the market open; nothing trades itself. Fills reconcile back through an audited email rail, and the books are rebuilt from the raw trade log every single night. Even the days nothing happens are recorded, because the boring days are a key part of the dataset.
The baseline
Buys the index on day one and never touches it again. Every other model exists to justify doing more than nothing.
The mechanicals · one frozen rulebook, three calendars
M1, M2, and M3 run identical rules over the same menu: industry sector ETFs (technology, health care, financials, energy, industrials) plus gold and long treasuries. The only variable is how often each one is allowed to decide.
Looks at the market daily and determines a weekly adjustment Sunday night. Trades Monday morning, then watches and learns throughout the week.
Re-evaluates every single close with the same rules as M1. Same brain, forty times the caffeine.
Same rules as M1, but rebalances once a month. Ignores the panic brake; volatility is someone else's concern.
The judgment
AI models running the show - total control, with a choice of models.
A frontier AI with live web research and an unbounded universe of US-listed instruments, on a leash: hard position caps, instrument screens, and a rule that its thesis goes on the record before any trade exists. Pick its brain at kickoff: Claude Fable, Opus, or Sonnet, and a campaign can race several brains side by side, one sleeve each.
The ruleset freezes at launch and its hash is stamped into every decision record. Nobody quietly improves the strategy in month seven.
Every evaluation stores the exact prices it saw. When data vendors restate history (they do), the decisions stay reproducible.
AI proposals pass position limits, leverage and OTC screens, and price checks. Anything invalid becomes a logged HOLD, never a hand-patched trade.
A named experiment: pick a roster of models, pick the AI brains (one sleeve per LLM), set the stake, choose real money or paper, and race them in parallel over the same market days.
Simulated fills at the closing price, no brokerage required. Test whether Sonnet beats Opus before a dollar moves.
Sign in, see exactly which trades are due, check them off as you fill them, and copy a pre-composed reconciliation email. The badge only clears when fills actually land.
Every AI call logs its tokens and its dollars, hold weeks included. The scorecard shows what each campaign's judgment actually cost.
Every record mirrors nightly to plain CSV. The whole experiment is reconstructable without any vendor, database, or goodwill.
Pre-registered hypotheses are the anti-hindsight device: written down before the first trade, graded at twelve months, no edits in between.
Today, campaigns are kicked off by the operator. A hosted run-your-own-campaign tier is on the roadmap.
Invite only. A signed-in viewer sees charts, scorecards, and signal history; tickets, trades, and briefs stay with the operator.
Already have an account on HouseLedgerAI.com, AICashFlow.ReboundMan.com, or SSA.ReboundMan.com? That same sign-in works here; the finance apps share one login.
PACER is a personal research experiment in decision cadence, not an investment product. Nothing here is investment advice. All trades are executed manually by the account owner; the system is decision support and record-keeping. Hypotheses are pre-registered and results will be reported against them, wins and losses alike.