Eric's Agent System

The Financial Copilot.

An agent that hunts the whole market on a schedule, must name its edge before it's allowed to recommend anything, and keeps a public score on itself. This page is how it works.

08 · The Hunt

It doesn't wait for ideas. It goes looking.

Earnings-forecast revisions, insider buying, institutional positioning, short interest and the actual text of SEC filings — read across every listed US stock, every week. Before it looks at a single name it reads the tape, and one of the four postures it can take is cash is the position. Survivors face the hard gate: the agent must name the edge.

6,757 listed US stocks 929 143 54 44 3 and some weeks: zero
Stage 1 · five lenses

Five questions, asked in parallel.

One screen can only find one kind of opportunity. This runs five over the same universe — and every candidate carries the lens that found it, all the way to the end.

What it catches: a business whose earnings forecasts are rising faster than its share price. It gets cheaper while going up, so it appears on no bargain list anywhere.
Stage 2 · triage

Name the edge, or pass.

Not a score. Not a rank. A specific, sayable reason this is mispriced — and no reason means no survival. It also reads what the system concluded in prior weeks, so it can't quietly re-litigate a call it already made.

What it catches: the 786 names that looked interesting on a screen and had nothing behind them.
Stage 3 · quick evaluation

Judged by the lens that found it.

A compounder is not graded like a falling knife. Four outcomes — and one of them is I can't tell yet, which sends the name out for more data instead of to the bin.

What it catches: the difference between a bad idea and a thin write-up. Missing information is a stated verdict here, never a silent gap.
Stage 4 · the critic

A fresh reader tries to kill it.

Thirty-four standing questions, asked by something that did not write the thesis. Every idea must state the single best argument against itself before it can pass.

What it catches: eight "independent" ideas that are one bet wearing eight tickers.
Stage 5 · full evaluation

Thirteen sections. Primary sources.

Reads the actual 10-Q. Pulls exact tagged figures straight out of SEC filings. States plainly what it could not find.

What it catches: the things no data feed sells — segment margins, debt maturity schedules, retention rates, backlog. Zero obligation to trade any of them.
Stage 1, up close

Price tells you a stock moved. It never tells you why.

Cross price direction with what analysts are doing to their earnings forecasts and four situations separate out — four situations that look identical on a chart.

Price falling
Price rising
Estimates rising
Disagreement
Analysts and the tape hold opposite views. One is wrong, usually inside a quarter.
candidate
Confirmed strength
Forecasts rising faster than the price. The multiple compresses while the stock climbs.
candidate
Estimates falling
No floor
The drawdown isn't a discount. The numbers are falling with it.
disqualifier
Quiet deterioration
Forecasts coming down while the price holds up. The market hasn't noticed.
exit signal
Confirmed strength is not a momentum trade.

Momentum buys the price move. This buys the gap between the price move and the earnings move — the stock is climbing and getting cheaper at the same time.

Quiet deterioration is the one that costs money.

Analysts are cutting their forecasts and the share price hasn't moved yet. Nothing looks wrong on the chart, so nobody sells — and the repricing happens all at once, later. If you already own it, this is the signal to get out.

Some weeks nothing meets the criteria.

It reports either way. Every research product gets paid whether or not there is anything worth buying, so every research product finds something. This one reports what it found and what it rejected — and when nothing clears the bar that week, it says so. That is a finding, not a blank page.

runs unattended · sunday 11:00 AM PT · output is a file the agent reads, not a tip
09 · The Mirror

It writes down what it rejected.

452

verdicts logged — passes, kills, and the not-yet-sures. Each carries the lens that found it and the reason it gave, so next week's hunt has to say what changed before it's allowed to disagree.

writeback.py recent --ticker EME
lens      CONFIRMED_STRENGTH
verdict   DEEPEN
thesis    estimates rising faster than the multiple is
          expanding — the only name in the batch
          where that is true.
Four positions closed and scored so far. Not enough to grade the system on — which is exactly why the log exists.

Most research shows you its winners. This keeps the losers in the same file.

Follow along.

Pick one or both. One email field, no spam, unsubscribe anytime.

selecting at least one list enables the button
Live Output

This Is What It Looks Like Running.

Tail of the agent status log — real format, unedited.

[06:05] Financial Agent · morning-briefing → COMPLETE
tail -f /home/oc/.openclaw/shared/status/AGENT_STATUS.md
[04:30:01] insider_cluster.sh shared/financial/INSIDER_CLUSTERS.md
[04:31:03] sec_edgar_feed.sh shared/financial/SEC_FILINGS.md
[04:32:06] macro_calendar.sh shared/context/MACRO_CALENDAR.md
[04:33:04] thirteen_f.sh shared/financial/13F_TRACKER.md (QUIET)
[04:34:03] congress_trades.sh shared/financial/CONGRESS_TRADES.md
 
Financial Agent | 06:05 PT | morning-briefing | COMPLETE
Briefing → shared/briefings/am.md
Portfolio: all positions graded
Watchlist: 3 names in entry range
Macro: calendar clear through week
 
Financial Agent | 06:35 PT | market-open-alert | COMPLETE
1 watchlist name hit entry zone → surfaced for approval
 
Marketing Agent | 07:00 PT | news-scan | COMPLETE
Top-3 stories tagged · content angles drafted
 
Orchestrator | 07:30 PT | summary | SENT
Telegram: "Portfolio graded. 1 approval needed.
Focus today: watchlist execution."