Most practice does nothing. A person can do a task for ten thousand hours and plateau early. The amateur golfer who has played weekly for twenty years is no better than they were after two, because they groove the same swing with the same flaws forever. What separates the people who keep improving is not more reps but a particular kind of rep, and the difference is entirely mechanical. Anders Ericsson's research on expert performance names four ingredients; here is the analyst's version.
How do you run a deliberate-practice rep?
(1) A defined target at the edge of your ability. A rep is not "do a teardown." It is "do a teardown of a bank, because banks are where my ROIC instinct breaks" or "do a teardown under a 45-minute clock, because I keep needing 80." The target names the specific sub-skill you are stretching. Reps inside your comfort zone (another clean FMCG compounder when you already read those well) feel productive and change nothing. Deliberate practice is uncomfortable by design; if the rep felt easy, it was maintenance, not improvement.
Advertisement
(2) Full effort, produced before you peek. The single most-broken rule in self-study: you must produce the complete output, all twelve questions, the full eight-line verdict, the actual reverse-DCF arithmetic, before you open the model answer or the rubric solution. The struggle is the learning. The moment you glance at the answer first, the rep converts from active retrieval (which builds the skill) to recognition (which builds only the comfortable illusion of competence). Write before you peek is working-agreement number three of the whole program (Section 0.1); it is nowhere more load-bearing than here.
(3) Immediate, specific feedback against a standard. The instant you finish, grade the output, line by line, against the rubric and against a model answer where one exists. Feedback that is immediate and specific ("your reverse DCF assumed 20% growth for a decade the base rates say survives ~5% of firms") teaches; feedback that is vague and delayed ("that felt a bit off") does not. This is why the rubrics here are printable and why you self-grade every rep, not just the certification: the rubric is a practice instrument before it is a gate. Read it before the rep so you know the shape of "good," and score against it after so you know your distance from it.
(4) Error analysis that feeds the next rep. The grade is not the point; the diagnosis is. Every miss goes into a miss log with its category, and every few reps you read the log for the pattern. This is the machinery that turns a graded rep into an improved analyst, and it deserves its own subsection.
The rep loop, then, is: pick a target at your edge → assemble the source stack → produce the full output under the intended constraint (open-book or timed) → self-grade against the rubric and model answer → log the systematic error → design the next rep to attack it. One full turn of that loop is worth ten casual teardowns.
How do you track and fix your mistakes?
The miss taxonomy — track your error signature. Not all misses are the same, and the fix depends on the type. Categorize every miss into one of six families, and keep a running tally. Over twenty reps a personal signature emerges, and it is almost never what you would have guessed.
| Miss type | What it looks like | The usual root cause | The fix |
|---|---|---|---|
| Coverage | You skipped a question entirely (no capital-allocation read; never scanned the pledge) | Time ran out, or the question felt boring | Time budget; the tripwire scan as a fixed reflex, not an optional step |
| Calibration | You covered it but reached the wrong verdict (called a moat that the ROIC series denies; "cheap" on a value trap) | Anchoring on the story; ignoring disconfirming numbers | The numbers-first order (read the ten-year panel before the deck); the steelman |
| Speed | Right answers, but you needed 90 minutes for a 60-minute job | Plumbing not yet automatic; backtracking | More reps on the slow sub-skill in isolation; the one-pass rule |
| False precision | A to-the-decimal "fair value" from a teardown; over-confident point estimates | Mistaking a teardown for a valuation | Ranges and verdicts, never targets (see the failure modes); "demanding/generous," not "₹2,847" |
| Sourcing | Built a thesis on a number you shouldn't trust (adjusted EBITDA, a promoter-flattered metric) | Skipped the quality-of-earnings reflex | The M2.06 Ctrl-F pass as a pre-condition to believing any headline number |
| Tripwire | Missed a governance/liquidity landmine (pledge at 70%, RPT ecosystem, ALM gap) | Ran the "good business" questions and forgot "what can kill it" | The fixed tripwire scan (see the failure modes), run on every company regardless of how clean it looks |
The discipline is to write the miss down by category every single rep, and to re-read the tally before your next one. If your signature is "coverage misses on capital allocation," you are not a worse analyst than someone whose signature is "calibration misses on cyclicals"; you have a different thing to fix, and the fix is different. This log is also the raw material for the calibration journal entry at the end of the module and for the lifelong error log (see the weekly rep cadence). Your edge in year five is disproportionately built from what you honestly recorded about your misses in year one.
What are the two feedback horizons?
Two feedback horizons. Process feedback (the rubric, the model answer) is fast: you get it in minutes, and it is what builds the skill now. Outcome feedback (did the market later confirm your variant view?) is slow, months to years, and noisy, because a good process can lose and a bad one can win (working-agreement eight). During certification and the reps around it, you live almost entirely on process feedback, because it is fast and controllable. After certification (see the weekly rep cadence), the decision journal starts harvesting outcome feedback too. The two together (did I run the process well, and did reality later agree) are what calibrate a mature analyst. Never let a good outcome excuse a sloppy process or a bad outcome indict a sound one.