The confirmation bias trap in high-conviction trades.

High-conviction positions are where independent challenge earns its fee, and they're also where it's hardest to deliver. A PM with a 6% position and a three-year thesis has usually already stress-tested the idea more thoroughly than any outside reviewer can in an afternoon. The failure mode we see most often isn't a lack of rigor — it's that the rigor was applied selectively, after the position was already sized.

The mechanism, not just the label

"Confirmation bias" gets used loosely enough that it's worth being precise about which mechanism is actually doing the damage in a book of concentrated positions. We see three distinct patterns, and they call for different challenge techniques.

1. Evidence-search asymmetry. Once a PM has a variant view, new information gets processed through a filter: corroborating data points are logged and remembered; disconfirming data points are explained away as noise, timing, or "the market not getting it yet." This isn't dishonesty — it's how associative memory works under motivated reasoning. The tell in a research file is a lopsided ratio of bull-case updates to bear-case updates over the holding period, even when the stock's realized volatility says the story is genuinely two-sided.

2. Thesis creep. The original variant view was, say, "margins normalize as input costs roll off." Eighteen months later input costs haven't rolled off, but the position is still core, now justified by a different thesis — pricing power, or a buyback, or a multiple re-rate. Each individual pivot might be defensible. The pattern across pivots is the signal: a thesis that survives by mutation rather than by being tested against its original falsification criteria has stopped being a thesis and become an attachment.

3. Position-size anchoring. The size of the position starts to function as evidence for the quality of the idea, independent of any new information. "I wouldn't have 8% in this if it weren't right" is a sentence we've heard, almost verbatim, from very good PMs. It's a completely natural cognitive shortcut and it is also backwards — position size should be an output of conviction and risk budget, not an input to conviction.

Why "just ask more questions" doesn't fix it

The standard defense against confirmation bias is devil's advocacy — someone asks the hard questions. In practice this underperforms for three reasons that matter for how a challenge function should actually be built:

• —Social cost asymmetry. In most investment teams, the person voicing the bear case on the PM's biggest winner pays a real reputational cost if they're wrong and gets little credit if they're right. Rational actors underinvest in that role. An external challenge function doesn't carry the same career math, which is precisely why allocators want it external.
• —Question framing collapses to the same prior. A generalist analyst asking "what could go wrong here" tends to generate risks adjacent to the ones already in the model — because they're pattern-matching off the same research file. Genuine disconfirmation usually comes from a different information architecture entirely: a different factor lens, a different peer set, a different macro regime assumption.
• —One-time review vs. standing process. A single pre-mortem at initiation catches thesis errors at time zero. It does nothing for thesis creep that happens gradually over six quarters. Bias correction needs a cadence, not an event.

What a working challenge process actually checks

We build our challenge process around falsifiable, dated theses rather than around personalities. Concretely, for every position above a conviction/size threshold, we hold three things in a form independent of the PM's narrative:

1. The original thesis and its stated falsification conditions, logged at initiation before subsequent price action can contaminate the story. If a PM can't articulate what would prove them wrong at the time of entry, that itself is the finding. 2. A rolling variant-view scorecard — the specific ways the position's thesis differs from consensus, refreshed quarterly and marked against realized outcomes, not against updated narrative. 3. An independent bear case, built without reference to the PM's file. This is the part that actually breaks the evidence-search asymmetry — a separate analyst starting from the sell-side skeptic case and the short thesis (where one exists), not from "what would make me doubt the PM."

The output isn't a recommendation to trim or exit. It's a structured record of where the thesis stands relative to where it started, delivered on a schedule the PM can't quietly avoid.

The base-rate problem allocators actually care about

From an allocator's seat, the interesting number isn't whether any single position review was rigorous. It's the base rate: across a manager's full history of high-conviction positions, what fraction were held past the point where the original thesis had clearly broken, versus exited or resized on discipline? That number is knowable if theses are logged contemporaneously and checked against outcomes — and it's close to un-knowable if the historical record consists only of quarterly letters written after the fact, which by construction can't help but rationalize.

This is the practical argument for standing challenge infrastructure over ad hoc pre-mortems: it produces an auditable trail an allocator can actually examine during due diligence, rather than a set of assurances that the manager takes conviction risk seriously.

Where this cuts against the manager, and where it protects them

It's worth being honest that a rigorous challenge process will occasionally force a good manager out of a position that would have worked. Confirmation bias correction is not costless, and any framework that claims otherwise is selling something. The case for it isn't that it improves every single decision — it's that it improves the distribution of decisions, and it gives both the manager and the allocator a documented, dated account of why a position was sized and held, which is worth a great deal when a position goes wrong for reasons that had nothing to do with the thesis at all.

The managers who get the most value from an external challenge seat aren't the ones who need convincing that bias exists. They're the ones sophisticated enough to know their own blind spots are the expensive ones, and who want an outside team whose incentives are structured to find them.