The FFAI Practice Book

• Only 3 rows carry Sep-2026 results (Leela Palaces, SG Finserve, Heritage Foods); 31 show Jun-2026; 9 still show Mar-2026. Two rows have no quarter change at all.
• Negative P/E: RPSG Ventures and AeonX Digital. 11 rows have no NSE code (BSE or SME listings) and 14 have a market cap under ₹1,000 crore.
• Most common industry: Auto Parts & Equipment (4), then Realty and IT Consulting (3 each). Only 11 of 43 grew both revenue and profit quarter on quarter.
• Bottom of the alpha table: Websol Energy −32.8, Horizon Industrial Parks −20.6, Kirloskar Pneumatic −18.9.

Lesson: a screen that says "superstar bought" says nothing about freshness or quality. The flags matter as much as the ranking.

• CAGR = (2003.88 ÷ 100)^(1/13) − 1 = 25.9%. The simple average of the 13 periods is 32.1% and the median 27.6%; the gap between average and CAGR is volatility drag, and it is why "average return" overstates what compounding delivers.
• The drawdown runs from NAV 446.14 (Mar 2018) to 281.76 (Mar 2020). Measured on daily data it would be deeper; year-end NAVs hide the inside-year swings.
• Reasons it would not match: equal weighting of ~200 names needs constant rebalancing (costs and taxes); the list of "superstar" holdings is known only after disclosures (timing lag); names that were dropped or delisted may be missing (survivorship); nobody holds 200 stocks in equal weights.

Historical backtest from the Day-1 pack, used only as a reading exercise. It does not describe any future outcome.

• Your counts will differ by day; what matters is the direction. 250-day lookbacks give fewer results than 125-day ones, because a 1-year high is rarer than a 6-month high.
• The bearish scan with "volume < average × 2" is looser than "volume < average" (a bigger ceiling lets more stocks through). If you got the opposite, check which side of the inequality the multiplier is on.
• A good description names the segment, the lookback and the one thing the scan is for, e.g. "Nifty 500 six-month breakouts on double volume, RSI under 70 — watchlist only".

• Around 245–250 rows per year for NSE. Fewer means missing days; more means duplicates.
• If Colab says No module named yfinance, add a first cell with !pip install yfinance and run it before the code.
• If the Nifty file is empty, the symbol is wrong: the index is ^NSEI, not NIFTY.
• A split-adjusted file has no single-day 40%+ jumps with matching volume spikes. If you see one, say so in your notes and treat the data before that date with care.

• Every pair has the same entry rule (next day's open) and the same cost; otherwise the table compares apples with mangoes.
• Shorter pairs (9/21) trade far more often and lose more to costs and whipsaws; longer pairs (50/200) trade a handful of times and sit through big drawdowns. Neither is "right".
• The honest result of step 4 is usually that the winner changes. Write that sentence down; it is the most valuable line in this lab.
• If the AI reports a return without a trade list, ask for the list. No list, no result.

• Every month shows "n = 8" or similar. A 7-of-8 month sounds strong until you remember 8 coin flips land 7-of-8 one time in 28.
• A high average with a low win rate means one or two huge months are carrying the average; the median tells the truer story.
• Differences between tools usually come from the start year, whether the current month is included, and summing vs compounding monthly changes (the Pine script sums).

• Rules in three lines: buy when price first trades above the listing-day high (at the open if it gaps above); stop at the lowest low from listing to the breakout day; exit at entry × 1.13 or at the stop, whichever hits first, and if both hit on the same day the stop is assumed first.
• Expectancy = 0.619 × 13.21% + 0.381 × (−11.37%) ≈ +3.9% per trade before costs. A 62% win rate with a payoff near 1 is a thin edge; costs, slippage on gaps and partial allotments eat into it.
• The "Total return" cell under the journey table shows #REF!; it should read (final ÷ 1,00,000) − 1. Broken formulas in a sheet are a reminder to recompute anything you rely on.

Historical simulation from the Day-4 pack, used only as a reading exercise. It is one realised path with no costs; it says nothing about future IPOs.

• At least one "CANNOT TELL" per company. Pledging and auditor notes are often not on the summary page; "cannot tell" is the correct answer until you open the filing.
• A name that scores high on the rubric but fails cash generation (OCF far below PAT for 3 years) goes to the "set aside" list, not the top 2. Scores never override a red flag.
• Your one-liners name a number, not a feeling: "ROCE 31% but receivable days rose from 48 to 91" beats "looks strong".