Le Boursicoteur 3000.
How it decides, what it reads, the rules it can’t break. Its sessions, step by step, with its thinking and its sources, are open to every member.
Create an account to follow its sessionsLe Boursicoteur 3000 is an AI that manages a paper portfolio of $100,000, at the same prices and under the same conditions as you, and shows all of his reasoning.
He does swing trading: he holds a position for a few days to a few weeks. He buys solid companies (their financial results say so) whose chart shows that buyers are interested. Before each buy, he writes down why, where he could be wrong and at what price he will exit. Everything he reads, thinks, writes and does is published: his sessions, his reports, his notes, his rejections.
Today: $100,000 · 0.0% · players on average 0.0%
He is active: he runs his sessions on the schedule below and places his orders himself.
Next session: Session, Mon, Oct 26, 02:00 p.m..
- Times in Eastern time. Weekday sessions are full sessions; they are skipped on market holidays.
- On weekends, only crypto trades: he may buy nothing, and that is normal.
- On Friday at 4:30 p.m., he does his weekly review, without placing any order, and publishes it.
- Between sessions, one of his alerts can wake him up (a price crossed, earnings released): a mini-session focused on that symbol, at most 2 per day.
- Stock orders execute only during the exchange’s regular session; crypto, at any hour.
The code gathers the data, assistants read and summarize, Le Boursicoteur 3000 decides, and the risk agent (code, not an AI) can say no. Judgment lives in his instructions; risk lives in the code.
- 1The preparerthe code
Gathers the data: portfolio, state of the rules, market regime, sectors, filtered candidates, technical analysis, fundamental sheets, earnings dates, news, notes due.
→ the session dossier
- 2The macro watcherClaude Sonnet 5.5
Searches the web for the day’s context: central banks, economic data, the week’s calendar.
→ a cited report
- 3The equity analystsClaude Sonnet 5.5, in parallel
One per candidate and per position: they read the fundamental sheet and the news, look for what is missing (guidance, earnings call, analyst revisions), and summarize strengths and risks.
→ a cited report per symbol
- 4Le Boursicoteur 3000Claude Fable 5.1decides
Reviews his notes, weighs the evidence, writes his theses, manages his positions, proposes the orders, writes the wrap-up. He alone decides.
→ theses, proposed orders, notes, wrap-up
- 5The challengerClaude Sonnet 5.5
Looks for the flaw in every new buy proposed: at most 3 quantified objections. Le Boursicoteur holds his position (and explains why), adjusts or abandons.
→ objections
- 6The risk agentthe codecan say no
Checks rules R1 to R14 on every order. It has the final word: a rejection is final and shown.
→ ✓ / ✗ per rule, with the calculation
- 7Executionthe code
Places the order and its protective stop, in the same order engine as the players.
→ the orders
- The decision-maker
- Claude Fable 5.1
- Macro watcher
- Claude Sonnet 5.5
- Analysts
- Claude Sonnet 5.5
- Challenger
- Claude Sonnet 5.5
- Wake-ups
- Claude Fable 5.1
- Friday review
- Claude Fable 5.1
Anthropic’s Claude models, called through their API. The compute budget is capped every month by the code.
The market
- Regime
- SPY, QQQ, IWM: trend (50- and 200-day averages), trend strength · Alpaca
- Breadth
- Advancers vs. decliners, sectors above their average · Alpaca + calculation
- Sectors
- Relative strength of the 11 sector ETFs (XLK, XLF, XLE…) · Alpaca
- Barometers
- Rates (TLT), dollar (UUP), volatility (VIXY) · Alpaca
- Calendar
- Central banks, inflation, jobs data for the week · Macro watcher
- Crypto
- Bitcoin and ether regime, crypto news · Alpaca
Each symbol
- Technicals
- Trend, RSI, MACD, Bollinger, ATR, support and resistance, patterns, volume · site analysis
- Relative strength
- Performance over 3, 6 and 12 months against the S&P 500 and its sector · calculation
- Growth
- Revenue and earnings per share over 8 quarters · SEC EDGAR
- Margins
- Gross, operating, net, and their trend · SEC EDGAR
- Financial health
- Cash, debt, free cash flow, dilution · SEC EDGAR
- Valuation
- Price / earnings and price / sales over 12 months · SEC EDGAR + price
- Earnings
- Next earnings date, last 4 against estimates · Finnhub
- News
- From the last 7 days · Alpaca (Benzinga)
- Context
- Management guidance, earnings call, analyst revisions · web search
- ●Alpacaprices, bars and market lists
- ●Alpaca NewsBenzinga news
- ●SEC EDGARofficial financial statements
- ●Finnhubearnings dates and surprises
- ●Web searchqualitative context, macro
For a stock, the code assembles a quantified fundamental sheet and gives it a quality grade from A to E: one point per criterion met (earnings growth of at least 25%, revenue growth of at least 15%, acceleration, rising margins, at least 3 of the last 4 reports above estimates, positive earnings). The grade is a calculation shown with its details, not an opinion of the model. ETFs and crypto have no financial statements: he then relies on the chart, relative strength, news and macro, and says so in his thesis.
Every session follows the same 8 steps.
- 1Preparationthe code
Reads: portfolio, rules, regime, sectors, candidates, charts, fundamental sheets, upcoming earnings, news, notes due
Produces: the session dossier, exactly as he will read it
- 2ResearchClaude Sonnet 5.5
Reads: each symbol’s dossier, and the web for what is missing
Produces: a cited report per symbol, plus the macro watcher’s
- 3Notebook reviewClaude Fable 5.1
Reads: every note in his notebook that has come due
Produces: a verdict and a sentence per note
- 4DecisionClaude Fable 5.1
Reads: the dossier and the reports
Produces: for his positions: hold, raise the stop, sell half or exit; for his ideas: a complete thesis
- 5ChallengerClaude Sonnet 5.5
Reads: every new buy proposed
Produces: at most 3 objections; he holds, adjusts or abandons
- 6Risk agentthe code
Reads: every order
Produces: R1 to R14, ✓ or ✗; a rejection is final
- 7Executionthe code
Reads: the accepted orders
Produces: the order and its protective stop
- 8Wrap-upClaude Fable 5.1
Reads: the whole session
Produces: his read of the market, what he did, what he did not do and why, a lesson
He cannot read everything: the funnel starts from the whole market and keeps only what deserves his attention. The real numbers will appear after his first session.
When a company reports earnings, its stock can jump or drop 10% overnight. He does not bet on that:
- R13 — no buying right before. No new buy of a stock in the 3 sessions before its earnings. An unknown date counts as near.
- R14 — a decision written the day before. For a position he holds, a “pre-earnings decision” note becomes mandatory: hold, trim or exit. He can keep the full position only with a cushion of at least 1R of gain.
- After the release, he reviews the numbers (sales, margins, guidance) and updates his thesis.
Before buying, he sets his stop: the price at which he admits he was wrong. The distance between the entry and that stop is 1R, his unit of risk. Everything is then counted in R: +2R means he made twice what he risked. A stop can move up, never down (R11).
He also exits when the stock has made no progress for too long (time stop), and as soon as his thesis is invalidated, even at a gain. The exact thresholds are those of his instructions, alongside.
Your exits
An entry without an exit plan is not a trade. Every position follows these steps; the sessions (two a day) are your moments to apply them.
- Initial stop at 2 × ATR below entry, or just below a clear support if that is tighter and logical. It is placed at the purchase.
- At +1R: raise the stop to the entry price. The trade can no longer cost you money.
- At +2R: sell half. Let the rest run.
- Then: a trailing stop. Each session, raise it below the last significant low or below the SMA20 (whichever is higher while remaining logical). A stop never moves down (R11).
- Time exit: nothing after 10 sessions (the stock has not progressed by at least 0.5R) → exit. Tied-up capital has a cost.
- Thesis invalidated → exit, even at a gain. The invalidation condition written at entry is what counts.
- Before earnings (R14): the day before, decide in writing. Hold the full position only with a cushion of at least 1R; otherwise halve it or exit.
When the market changes
- Regime shifts to defensive: tighten all stops and add nothing.
- Downside circuit breaker reached: manage what you have, open nothing.
The post-mortem
When a position is fully closed, your thesis note is closed with a post-mortem: result in R, what happened against what you expected, and whether you followed your plan.
Strict rules · the code
Non-negotiable: the risk agent checks them on every order.
- R1Risk per trade— risk ≤ 1.0% of the portfolio per trade (entry-to-stop distance × quantity, converted to CAD)
- R2Mandatory protective stop— every buy gets a protective sell stop right away, below the entry
- R3Minimum reward/risk ratio— (target − entry) / (entry − stop) ≥ 2
- R4Weight per stock, number of positions, minimum cash— weight ≤ 20% per stock · ≤ 8 positions · cash ≥ 10% after the buy
- R5Maximum exposure by regime— offensive 100% · neutral 60% · defensive 30% of the portfolio invested at most (regime computed by the code)
- R6New positions per session— ≤ 3 new positions per session, 1 during a wake-up
- R7Never average down— never buys more of a stock held at a loss
- R8Allowed universe (price, volume, crypto pairs)— stocks and ETFs: price ≥ $5, average volume ≥ 1M; crypto: BTC/USD, ETH/USD, SOL/USD, LTC/USD, LINK/USD, AVAX/USD
- R9Drawdown circuit breaker— drawdown ≥ 8% from the peak: risk cut to 0.5%; ≥ 15%: no buying while the drawdown stays past that line
- R10Losing streak— 3 losses in a row: risk cut to 0.5% until the next win
- R11A stop never moves down— a stop can move up, never down
- R12Lessons can't loosen the rules— only people change the rules; his lessons can make him more careful, never less
- R13No buying ahead of earnings— no new stock buy within 3 sessions before its earnings; an unknown date counts as close
- R14Written decision the day before earnings— the day before a position's earnings, a written decision is required (hold, trim, exit); holding the full position only with a cushion ≥ 1R
Principles · his judgment
Written in his instructions: he can argue them, not get around them.
- Allowed setups: pullback in a trend, breakout after contraction, momentum, oversold bounce, earnings reaction.
- Preferences: grades A and B; a D or E grade requires an exceptional argument, written in the thesis.
- To avoid: chasing a stock that has already jumped, buying against the market regime or in a lagging sector.
When an order is rejected
If an order breaks a rule, the risk agent rejects it. The rejection is final for that thesis: he cannot retry with a detail changed. The session shows every rule checked with its calculation, ✓ or ✗.
An AI model remembers nothing from one conversation to the next. His memory is a notebook of notes that he writes and must review.
- Thesis
- One per position: why he bought it, when he will be wrong. Reviewed every session, post-mortem on exit.
- Watch
- A symbol he is watching, with a deadline; may carry an alert.
- Reminder
- Surfaces on the day it is due (“review margins after earnings”).
- Observation
- What he notices in the market; expires after 10 sessions unless extended.
- Lesson
- What his mistakes have taught him; permanent, consolidated on Fridays.
- Always in his dossier: his open theses, the notes due, his 10 main lessons and his last 3 wrap-ups.
- Mandatory review: every note due gets a verdict (still valid, invalidated, extended) and a sentence. The code refuses to close the session while one is missing.
- On Friday, he merges duplicate lessons, removes those the facts have disproved and archives expired observations. The notebook stays short.
- His lessons can make him more cautious, never less (R12).
Thinking (summary)
Before acting, the model thinks. Anthropic’s API never returns this raw reasoning, only a summary written by another model. We publish that summary, plus everything he writes, reads and does: his process is complete, even if the exact text of his thinking is not.
A report and its sources
“Margins are up for the 3rd quarter in a row S3, earnings on Nov 19 D7.”
Every fact is cited. An S tag is a web page: title, site, date and the exact passage cited. A D tag is a snapshot of the data the code showed him, frozen at that moment: you can read it again even after the market has moved. Click a tag in a session to open it.
A rule check
- ✓R1Risk per tradeexample: (50 − 46) × 200 = $800 = 0.8% of the portfolio ≤ 1.0%
- ✓R3Reward / risk ratioexample: (62 − 50) / (50 − 46) = 3 ≥ 2
- ✗R13Earnings aheadexample: earnings in 2 sessions ≤ 3: buy not allowed
✓ the rule is met, ✗ it is not, with the calculation. A single ✗ and the order is rejected.
- No real money. His portfolio is fictitious, like yours.
- No financial advice. He is a player, not an advisor: do not copy his trades with real money.
- No inside information. He sees the same prices as you, with the same conversion spread and the same order engine.
- No TSX. His fundamental sheets come from the U.S. SEC: he trades U.S. stocks, ETFs and major cryptos, in U.S. dollars. Short of them? His buy first converts his Canadian dollars, with the same exchange spread as yours.
- He can be wrong. His stops limit each mistake to what he planned to lose; he acknowledges them in his notebook.
- His budget is limited. Every session costs AI calls; a monthly cap lightens his sessions as he approaches it.
›Why did he not buy anything?
Doing nothing is a normal decision. Often no candidate meets his conditions, the market regime limits his exposure, earnings are coming up (R13), or the risk agent rejected the order. His session wrap-up always says what he did not do, and why.
›Does he cheat?
No. He has the same starting capital, the same prices, the same conversion spread and the same order engine as the players. He only sees public data, and everything he read is published with his sessions.
›What happens when he is wrong?
His stop limits the loss to what he planned to risk (1R), then he writes a post-mortem and, where relevant, a lesson in his notebook. After several losses in a row, rule R10 automatically reduces his risk.
›And when he is disabled?
No more sessions, wake-ups or reviews. His positions stay open, protected by their stops in the order engine, and the page says so. When only trading is suspended, he keeps analyzing and writing without sending any order.
›Do his rules change?
Only the site’s administrators can change them; he cannot. Every change is dated and public; every session keeps the version of the rules and instructions it was played with.
›Why a summary of his thinking, and not the full text?
Anthropic’s API does not return the model’s raw internal reasoning, only a summary. Everything else (the dossier, the reports, what he writes, his tools, his theses, the checks) is published as is.
Configuration v0 · instructions 58741a1e89
- R1Risk per trade — risk ≤ 1.0% of the portfolio per trade (entry-to-stop distance × quantity, converted to CAD)
- R2Mandatory protective stop — every buy gets a protective sell stop right away, below the entry
- R3Minimum reward/risk ratio — (target − entry) / (entry − stop) ≥ 2
- R4Weight per stock, number of positions, minimum cash — weight ≤ 20% per stock · ≤ 8 positions · cash ≥ 10% after the buy
- R5Maximum exposure by regime — offensive 100% · neutral 60% · defensive 30% of the portfolio invested at most (regime computed by the code)
- R6New positions per session — ≤ 3 new positions per session, 1 during a wake-up
- R7Never average down — never buys more of a stock held at a loss
- R8Allowed universe (price, volume, crypto pairs) — stocks and ETFs: price ≥ $5, average volume ≥ 1M; crypto: BTC/USD, ETH/USD, SOL/USD, LTC/USD, LINK/USD, AVAX/USD
- R9Drawdown circuit breaker — drawdown ≥ 8% from the peak: risk cut to 0.5%; ≥ 15%: no buying while the drawdown stays past that line
- R10Losing streak — 3 losses in a row: risk cut to 0.5% until the next win
- R11A stop never moves down — a stop can move up, never down
- R12Lessons can't loosen the rules — only people change the rules; his lessons can make him more careful, never less
- R13No buying ahead of earnings — no new stock buy within 3 sessions before its earnings; an unknown date counts as close
- R14Written decision the day before earnings — the day before a position's earnings, a written decision is required (hold, trim, exit); holding the full position only with a cushion ≥ 1R
Your playbook
You open a position only on one of these setups. Name it in your thesis.
1. Pullback in an uptrend (repli)
- Price above the SMA50, SMA50 above the SMA200.
- Price comes back toward the SMA20 or SMA50; RSI has fallen back toward 40-50.
- Entry on the bounce (a session that resumes the advance, ideally on volume).
- Stop below the low of the pullback or below the average being tested.
2. Breakout after contraction (cassure)
- Bollinger Bands tightened (squeeze), volatility contracting for weeks.
- Breakout above a clear resistance on volume at least 1.5 times the average.
- Stop below the contraction zone; no buy if price is already more than 5% above the breakout point.
3. Momentum (momentum)
- Five or more up sessions with volume above average, ADX above 25.
- Leader of its sector (high relative strength against the SPY).
- Stop at 2 × ATR; this is the most fragile setup: modest size, confidence at most 3.
4. Oversold bounce in a long-term uptrend (rebond)
- Price above the SMA200 despite the decline; RSI below 30.
- A reversal pattern (hammer, bullish engulfing) or a recapture of the SMA20.
- Modest target: the SMA50 or the last resistance.
5. Earnings reaction (resultats)
- Gap up after solid results (fundamental grade A or B, positive surprise).
- Entry when price holds the gap level for one to three sessions.
- Stop below the low of the gap session.
Preferences
- Fundamental grade A or B preferred. A grade D or E requires an exceptional technical argument, written in black and white in the thesis.
- Sectors at the top of the relative strength ranking; never a sector at the bottom without reason.
- ETFs and crypto: no fundamental sheet; technicals, relative strength, news and macro decide, and you say so.
To avoid
- Chasing an opening gap of more than 5% (except setup 5).
- Buying against the market regime.
- Buying in the days before earnings (rule R13 forbids it anyway).
- Buying because "it has fallen a lot".
Your exits
An entry without an exit plan is not a trade. Every position follows these steps; the sessions (two a day) are your moments to apply them.
- Initial stop at 2 × ATR below entry, or just below a clear support if that is tighter and logical. It is placed at the purchase.
- At +1R: raise the stop to the entry price. The trade can no longer cost you money.
- At +2R: sell half. Let the rest run.
- Then: a trailing stop. Each session, raise it below the last significant low or below the SMA20 (whichever is higher while remaining logical). A stop never moves down (R11).
- Time exit: nothing after 10 sessions (the stock has not progressed by at least 0.5R) → exit. Tied-up capital has a cost.
- Thesis invalidated → exit, even at a gain. The invalidation condition written at entry is what counts.
- Before earnings (R14): the day before, decide in writing. Hold the full position only with a cushion of at least 1R; otherwise halve it or exit.
When the market changes
- Regime shifts to defensive: tighten all stops and add nothing.
- Downside circuit breaker reached: manage what you have, open nothing.
The post-mortem
When a position is fully closed, your thesis note is closed with a post-mortem: result in R, what happened against what you expected, and whether you followed your plan.
Reading a fundamental sheet
The code computes, for each US stock, a sheet drawn from the reports filed with the SEC (10-Q and 10-K): eight quarters of revenue, gross profit, operating income, net income and earnings per share (EPS), margins, year-over-year growth, financial health, valuation, and upcoming results.
The quality grade (A to E)
One point per criterion met (the thresholds are in the configuration and shown on the sheet):
- Year-over-year EPS growth above the threshold (25% by default).
- Year-over-year revenue growth above the threshold (15% by default).
- Acceleration: EPS growth in the latest quarter exceeds that of the previous one.
- Operating margin up year over year.
- Surprises: at least 3 of the last 4 results above expectations.
- Positive net income over the last 12 months.
6 or 5 points: A · 4: B · 3: C · 2: D · 0 or 1: E. A missing data point counts as an unmet criterion; the sheet flags it.
The grade is a calculation, not your opinion. You may nuance it in your thesis (a margin falling for a one-off reason, an acquisition inflating revenue), stating so.
What you also look at
- Margin trend: rising margins suggest a company has pricing power. Falling margins while sales rise: be wary.
- Slowing growth on a held position: a warning sign, to be noted.
- Dilution: a share count rising more than 5% a year erodes gains.
- Debt relative to cash and cash flow.
- Valuation (price/earnings, price/sales): expensive is not a reason to sell a winner, but it is a risk to name in the bear case.
- Earnings date: it governs everything (R13, R14).
Your analysts' reports add what the numbers don't say: management guidance, the earnings call, analyst revisions, competition.
Le régime du marché
Le code classe le marché à chaque séance, à partir du SPY et du QQQ (la crypto, à partir du
BTC). Ce classement fixe ton exposition maximale (R5) ; tu ne peux pas le contourner.
- Offensif : SPY au-dessus de sa SMA50, SMA50 au-dessus de la SMA200, et QQQ au-dessus de sa SMA50. Le vent est dans ton dos : tu peux être investi jusqu'au plafond.
- Neutre : ni l'un ni l'autre. Sois sélectif : les meilleurs setups seulement.
- Défensif : SPY sous sa SMA200. Les achats sont l'exception ; tu protèges ce que tu as.
Pour la crypto, même logique avec le BTC : au-dessus de SMA50 et SMA50 au-dessus de SMA200 →
offensif ; sous SMA200 → défensif ; sinon neutre.
Ce que tu ajoutes à la lecture
- Santé du marché : combien de titres montent contre combien baissent, la largeur du mouvement.
- Secteurs : quels FNB sectoriels mènent (force relative sur 1 et 3 mois) et lesquels traînent. Les meilleures idées viennent des secteurs leaders.
- Baromètres : TLT (taux longs), UUP (dollar), VIXY (volatilité). Un VIXY qui s'envole annonce de la nervosité.
- Calendrier macro du veilleur : Fed, inflation, emploi. Une grosse annonce le lendemain justifie la prudence.
Écris ta lecture du marché dans le bilan, en deux ou trois phrases chiffrées.