Signal and noise

The roadmap ends with a short list of things to avoid: books called How to Trade [Anything], finance Twitter/X, and trading YouTube. This lesson is why — and the reasons are all things the course has already established, which makes it a fair final test of whether they landed.

Why the avoid-list is one problem, not three

Every item fails the same way: you are shown a selected sample.

   what you see                     what generated it
   ─────────────────────────────────────────────────────────────
   a screenshot of a great          thousands of accounts trading;
   month                            the ones with bad months stop
                                    posting. SURVIVORSHIP (M2.7)

   a strategy with a beautiful      one of hundreds tried, and the
   backtest                         best kept. MULTIPLE TESTING —
                                    best-of-1000 on zero edge
                                    scores Sharpe 1.4 (M2.7)

   "up 340% since inception"        inception chosen after the
                                    fact. And no costs (M2.6),
                                    no capacity (M1.5), no
                                    drawdown (M3.6)

   a confident causal story         needs a villain and a daily
   about yesterday's move           deadline (M6.1)

There is also a business-model tell worth stating plainly. If someone has a strategy that works at scale, selling a course about it is a strictly worse use of their time than trading it. Capacity (M1.5) means every additional participant makes the edge worse, so publishing a genuinely profitable short-horizon strategy is self-harm. The exceptions are instructive: AQR publishes because factor premia are risk compensation rather than secrets (M5.4), and academics publish because that is the job. Neither is selling you anything.

This is not a claim that nobody on the internet knows anything — plenty do, and some of the best market-structure writing is on blogs. It is a claim about the base rate of a feed, and about what filtering costs you when you cannot see the denominator.

The checklist, turned outward

M2.7 built a checklist for interrogating your own backtest. The same questions work on anyone else's claim, and they are more useful pointed outward because you are less motivated to fudge them.

   ✓  What is the effect size, in basis points, BEFORE any p-value?
      A statistically significant edge of 5bps/day that costs 5bps
      to trade is worth nothing. Significance is necessary and
      nowhere near sufficient (M3.7).

   ✓  What are the costs at the size actually intended?
      Not a flat fee — M1.5's square-root law at that participation
      rate. M2.6's pairs trade went from Sharpe 1.39 to 0.17 on
      this line alone, with the signal unchanged.

   ✓  How many variants were tried before this one?
      Including the abandoned ones. If nobody counted, the number
      is larger than they think (M2.7).

   ✓  How much of it rests on a handful of days?
      Drop the ten biggest and re-run. On the real S&P, ten days
      out of 2,513 move the Sharpe between 0.37 and 1.17 (M3.7).

   ✓  What is the capacity?
      Alpha scales with Q, cost with Q^1.5. Every edge has a size
      at which it stops existing (M1.5, M5.4).

   ✓  Who is on the other side, and why are they wrong?
      If the answer is "they're dumb", be suspicious. Usually the
      other side is being paid for a risk (M4.6, M2.4).

   ✓  What regime does this assume?
      A relationship fitted across one regime need not survive a
      change of it — ask the 60/40 portfolio (M4.6).

   ✓  Drawdown, not variance?
      What is the worst path, not the average one (M3.6).

Applied honestly, that list disqualifies the overwhelming majority of claims you will encounter, including some made by serious people in good faith. That is the correct outcome, and being comfortable with it is most of what separates literacy from credulity.

Two failure modes to avoid in yourself

Having a checklist creates its own risks, and they are worth naming.

Selective scepticism. Applying the list to claims you dislike and waiving it for ones you find congenial. The test is whether you have ever used it to kill something you wanted to be true.

Corrosive scepticism. Concluding that since everything fails the checklist, nothing works and the whole enterprise is noise. That is also wrong — market makers really do earn the spread, factor premia have survived decades of out-of-sample scrutiny, and Medallion's record is long enough to be beyond the survivorship critique. Some things are real. The point of the list is to tell which, not to conclude that none are.

The honest position after this course is narrower than either: most claims of edge are artefacts, a few are real, the difference is usually costs and capacity rather than cleverness, and you can now tell which questions separate them.

Where this leaves you

Six modules, and the arc is worth stating once at the end.

M0 gave you the machinery — order books, spreads, the path from click to settlement. M1 explained why the machinery behaves as it does, and derived the spread from adverse selection and inventory. M2 built the toolkit for pricing and strategy, and showed that costs, not signals, usually decide. M3 removed the Gaussian assumption underneath all of it. M4 supplied the macro context that explains when things move. M5 named the firms and showed that their structures follow from where their money comes from.

M6 is the one that doesn't finish. Ten minutes a day, one weekly, and the books over months — and the machinery above is what turns that from a stream of anecdotes into an accumulating map.

That was the whole goal, stated at the very start of the roadmap: not to make you a practitioner, but to make you literate — which is the prerequisite for everything else, and the thing that lets you hold a conversation with someone who does this for a living and ask them a question they find interesting.

Source: nothing new here — every argument in this lesson is from M1.5, M2.6, M2.7, M3.6, M3.7, M4.6 or M5.4. That is deliberate. If any of the checklist items above needed explaining, that is the module to revisit.