Ninety days of paper trading is not a substitute for a backtest and it is not a warm up. It answers a question no backtest can answer, which is whether you personally will execute the rule you wrote down when the rule tells you to buy something that looks terrible. That is the failure mode that kills retail strategies, and it is the one you can only measure on yourself.
Write the rule before you look at a single filing
The protocol starts with a paragraph, not a trade. Write the entry rule so specifically that another person could execute it without asking you a question, then do not change it for ninety days. Vagueness here is what makes the whole exercise worthless, because a rule you can reinterpret is a rule you will reinterpret in your own favour every time.
Use the module's own controls as the vocabulary, because they force you to be specific. On the performance view the filters at capture were a date range of last 6mo, last 1yr, last 5yr or all time, a side selector offering All, Buys only or Sells only, and a size dropdown with buckets at 25 thousand, 100 thousand, 500 thousand and one million dollars. Your rule should name a side and a size threshold in exactly those terms, because that is how you will later check whether the population you studied matches the trades you took.
Then add the three things the screen will not decide for you. A maximum number of open positions, so the exercise stays countable. A fixed holding period with a calendar exit date written at entry. And a liquidity condition, stated as a spread you have actually observed on the name rather than as a market cap band, because the spread is what will eat you and the market cap is not.
An example of the required specificity, so there is no ambiguity about the standard. Open market purchases only, trade value at 100 thousand dollars or above, quoted spread no wider than fifteen basis points at the time I look, maximum six open positions, equal dollar size, exit ninety calendar days after entry. That is a rule. Buy good insider signals is not.
The eight fields, logged the same day, in one place
Paper trading fails more often from bad record keeping than from bad rules. Keep one row per position and fill these eight fields at the moment of the decision, not at the end of the week when you remember it differently.
- Date and time you saw the filing, and the date the filing itself was made. The gap between those two is your personal latency and you will want it later.
- Ticker and the trade value bucket that qualified it.
- The price at the moment of the decision, and the quoted spread you observed at that moment. Not the closing price. The one on the screen when you decided.
- The planned exit date, written before anything else happens.
- Your conviction on a one to five scale, recorded before the outcome exists.
- One sentence of reasoning, in your own words, which is the field you will be most embarrassed by and which is the most useful of the eight.
- Whether the position was one your rule required, or one you wanted. Record the second kind as a rule break even though you are only on paper.
- The exit price and date, and whether the exit was the calendar one or an override.
The weekly review that makes the log worth keeping
Once a week, thirty minutes, same day each week. Do not look at profit and loss first. Look at compliance with your own rule first, because that is what you are testing.
Count three numbers and write them at the top of the week's entry. Signals that met your rule. Positions you actually opened. Positions you opened that did not meet your rule. If the second number is materially below the first, you are not testing the strategy you wrote, you are testing a discretionary version of it and you will not be able to attribute anything at the end. If the third number is above zero, write next to each one what you were feeling, because that pattern will repeat with real money and larger size.
Then update your running cost estimate. For every position opened that week, take the spread you observed, halve it, double it for the round trip, and add it to a column. This is the single most useful thing in the whole log, because it converts an abstract worry about friction into a number that will be sitting there on day ninety when you decide whether to fund.
The benchmark to hold it against is the eighteen basis points of gross relative return that the panel reported over seven days, being plus 0.25 percent for the filings against plus 0.07 percent for SPY. That figure came from the seven day horizon, and the 30D, 90D and 365D tiles each showed a dash and zero scored filings on the day I read them, so I cannot tell you what the module measured over the horizon you are probably paper trading. Use the seven day gap as a sanity check on scale rather than as your expected return, and be suspicious of any paper result that comes in far above it.
The three tests on day ninety, applied in this order
At the end of the quarter, run these three in sequence and stop at the first failure. The order matters because passing the later tests means nothing if you failed the first.
Test one is execution fidelity. Divide positions opened by qualifying signals, and count rule breaks. If you executed less than three quarters of your own signals, or if you broke the rule more than twice, the answer is no and the reason has nothing to do with insider filings. Run another quarter with a simpler rule, or with fewer maximum positions, until you can follow it. This is the test most people fail and the one most people skip.
Test two is sample sufficiency. Count the closed positions. If you have fewer than twenty, you do not have a result, you have anecdotes, and the correct action is to extend rather than to fund. Twenty is not a statistically comfortable number either, it is simply the point below which the exercise cannot distinguish a process from a coincidence at all. Say out loud what your average outcome was and then say how many trades it came from, in the same sentence, every time.
Test three is cost survival. Take your paper return, subtract the cost column you have been accumulating, and subtract it again as a margin of safety because paper fills are always better than real ones. Paper trading gives you the mid price and reality gives you the far side of the spread, and in the thinner names it gives you the far side plus your own impact. If the result after that double haircut is not clearly positive, the answer is no, and no amount of filter tuning after the fact repairs it, because tuning filters on ninety days of your own data is how you fit a rule to noise.
What funding actually looks like if you pass
Passing all three does not mean deploying the intended allocation. It means starting at the smallest size where the trade is real, which for most people means an amount where a total loss on one position is genuinely absorbable, and running the identical rule with the identical log for another quarter.
The reason for the small first tranche is that two things change the moment money is involved and neither is in your paper data. Your fills get worse, which you can measure by comparing every real fill against the decision price you have been recording all along. And your behaviour gets worse, which you can measure by the same rule break count you have already been keeping. Both are now baselined, which is the actual product of the ninety days.
Keep the log running after funding rather than retiring it. A paper protocol that ends when the money arrives has taught you about a period when nothing was at stake, and the interesting divergence is between that period and the one that follows. If your rule break count triples in the first funded month, you learned something worth far more than the strategy, and you learned it at the smallest size you will ever trade.