Most practice logs fail before the first fill. They fail at the calendar. You randomise a start, hide the date, walk candles honestly — and still grade a setup on one kind of tape because that is the kind of tape you keep drawing.
Period selection is the biggest hidden bias in replay. Not peeking. Not a sloppy stop. The quiet decision of which history you allow into the room.
The problem is not “too little data”
It is matched data. Ten randomised NSE sessions that all behave like a clean trend off the 09:15 open will congratulate the same breakout stub. Ten sessions that all grind from 11:30 to 14:00 will punish it. Neither block tells you how the stub behaves on the other tape. Both feel like “a sample.”
Favourite-chart practice is the obvious version of this. Randomising historical starts fixes the obvious version. It does not fix the quieter version: a randomiser drawing from a short recent window, a watchlist of names that only trended, or a habit of aborting days that “don’t look like my setup.”
Misconception: random dates equal an unbiased sample
Randomisation is a defence against cherry-picking a hero Tuesday. It is not a defence against sampling one regime family. If your date pool is “the last bull stretch I remember,” random picks from that pool are still that stretch. If you skip any session that opens flat, you have a filter, not a sample.
Hidden dates stop memory from naming the budget or the crash. They do not stop you from quitting a choppy open because it feels unproductive. The bias moved from the calendar stamp to the skip finger.
Better framing: mix before you measure
Treat period selection as a design step, not a mood. Before you replay, write the mix you will accept. After you replay, check whether the mix actually arrived. Grade the stub only against a mix you declared — or admit you graded it against a tape you like.
This is not a claim that a balanced mix proves an edge. It is the narrower claim that an unbalanced mix cannot falsify anything. You cannot say “the setup failed” if you only fed it the tape that never produces it. You cannot say “the setup works for me” if you only fed it the tape that produces obvious breaks.
The six-bucket sample card (copy this)
Fill the card before session one. Tick buckets as you complete them. Do not start a “results week” until each required bucket has at least the count you wrote.
SAMPLE CARD — setup: ________ product: cash / swing (pick one)
Window: 09:15–15:30 IST (or a written sub-window)
Universe: liquid large-caps / index chart (not a buy list)
Required mix (write a number in each box; sum is your block):
[ ] A Open-drive — first 45–60 min, directional open, range expands
[ ] B Midday grind — 11:30–14:00, overlapping bars, skips should dominate
[ ] C Late squeeze — last 60–90 min into 15:30, participation shift
[ ] D Gap / leftover — open away from prior close; your gap rule must fire or skip
[ ] E Event-adjacent — results / policy week *if you know it*; else tag UNCLEAR
[ ] F Quiet large-cap — liquid name, small range, boredom is the lesson
Abort rule: if I discard a day, I still tick the bucket it belonged to as SKIP-DAY.
Replacement: next random start must target the under-filled bucket, not “a better chart.”
Buckets A–C are session structure on ordinary NSE cash hours. D–F catch the data hazards people smuggle in: gaps, event memory, and the quiet tape that live trading will not spare you.
Corporate-action and index-reconstitution days belong in E or in a separate study pile. They change the series. Mixing them silently into “normal” breakout practice is how a survivorship-flavoured watchlist and a period bias stack.
Workflow for one 30-day block
- Freeze one stub. If the sentences are mush, they are not ready for a sample — write them testably first.
- Fill the sample card. Example mix for a finite practice month: three of A, three of B, two of C, two of D, one tagged E or F each week, not a holy ratio. Write your counts. Do not copy someone else’s.
- Draw random historical starts. After the open structure is visible — not after P&L — assign the bucket. If you cannot assign it by ~09:45, tag UNCLEAR and keep the session; unclarity is a bucket, not a reason to reroll.
- Replay candle-by-candle in decision zones. Fast-forward only where your written skip rule already fired. Speed is a different dial; do not use it to escape an under-filled grind bucket.
- Weekly audit: count ticks. If A is full and B is empty, next week’s randomiser is constrained: accept only B/C/F starts. Constraint is hygiene, not a new strategy.
- Hold out a sealed date list for a later test. Period mix in the train block does not licence peeking at the holdout.
Ten practice decisions a day × 30 days ≈ 300 practice decisions is a useful volume inside this mix. It is not a certificate that the mix was honest, and it does not make the stub profitable.
What this will not fix
Look-ahead still ruins a perfectly mixed card. If you peek, the bucket label is theatre. See look-ahead bias.
A mixed sample still sits on generous practice fills. Brokerage, STT and slippage apply live; replay will not charge them unless you haircut on purpose.
Index charts and single-name charts are different liquidity lessons. Do not fill six buckets on an index future and declare the cash stub validated.
If you only ever practise names you already survived with, period mix cannot save the watchlist. That is a separate bias.
Declare the tape mix before you declare a verdict on the stub. Join the waitlist.
Related reading
- Why Randomising Historical Markets Beats Favourite Charts
- Why Hidden Dates Make Trading Practice More Honest
- Walk-Forward Thinking for Discretionary Traders
- Look-Ahead Bias in Backtesting: How Traders Fool Themselves
Educational note
Educational practice content only. Not investment advice. No guaranteed returns. Practice ≠ live profitability.