Maintain the trade journal

Setup. The scripts sit beside this file. Run them from a working folder that holds your .env (E*TRADE consumer key/secret) and a data/ directory; install deps with pip install -r requirements.txt where one is included. Part of a set: etrade-pull, order-check, portfolio-review, realized-pnl, rebalance-plan, price-history, trade-journal, intrinsic-value, sec-financials, buffett-checklist.

Start from JOURNAL.template.md in the repo root; copy it to JOURNAL.md, which is gitignored. The template carries the full structure — this skill is how to operate it.

The purpose is narrow and worth stating plainly: the hold/exit decision gets made before conviction is tested, so it isn't made live in a moment of fear or greed. Everything below serves that.

Opening a position

Write the block before the buy, not after. It must answer:

• Thesis — the real-world mechanism. Not "it's cheap."
• Why I'm early — what the market hasn't priced, and why not. If this can't be answered, the position is consensus and should expect consensus returns. Writing it down is the only cheap way to find that out.
• Invalidation — specific, fundamental, and non-price. A named event or number meaning the thesis was wrong.
• Sell-into-strength signal — what recognition looks like.
• Catalyst / timeline — the dated events that would close the gap.

Rules

• Invalidation is never a price. "Down 20%" is volatility, not evidence. A price-based invalidation converts every drawdown into a sell signal, which is the exact behavior the journal exists to prevent — and see realized-pnl's counterfactual for what that habit costs in practice.

• Never edit a past entry to match what happened. Amend by appending a dated line to the review log. A journal rewritten in hindsight records only that you were always right, which is worth nothing when you later need to know whether you were unlucky or wrong.

• Log what you considered and did not do, with the reason. The rejected trades are half the record and the half nobody keeps.

• Most bad news is not invalidation. When something happens, state explicitly whether it meets a written criterion or is merely bad. Forcing that answer in writing is the whole mechanism; "this seems concerning" is how a rule gets quietly abandoned.

• Keep exited positions in the file, marked EXITED <date> with what actually happened. A thesis that didn't work is the most useful thing in here later.

• Adding to a position is a new decision. It must re-clear the thesis, not inherit it. Averaging down without re-underwriting is how a small mistake becomes the largest line in the book.

The gate register

One table indexing every dated trigger already written in prose somewhere in the file. It is the index, not the authority — the position block wins on reasoning. It exists because once the file is long, reconstructing every live trigger from prose is exactly where one gets missed.

Date	Name	What	Action it unlocks	State

Fire on the transition, not the state. A gate reading "under $227" is true every day the price sits there. Acting on the state means re-deciding a settled trade every morning; ignoring the noise means missing the crossing. So each gate is ARMED until it crosses, then stamped FIRED <date>, and stays fired until deliberately re-armed with a reason.

When asked whether something fires today: read the register, check the transition, and answer yes or no with the date. Don't re-derive the trade — it was already decided, by design.

After a review

A review that changes nothing on paper gets re-litigated from scratch next month. When portfolio-review reaches a conclusion, append it to the review log with the date, what changed, and what it means for each affected thesis.