Skip to content

How much time should I give a bug hunt?

I'm wondering whether the Kelly criterion can help me think about research time, and where that comparison breaks down.

planted April 22, 2026 · last tended November 9, 2026

A captured spark. Unverified, unpolished, possibly wrong.


I can spend a long time on a security target because it feels as though a finding must be just around the corner. Sometimes staying with a problem is useful, but I’d like a better way to decide when to continue and when to give something else a turn.

That led me to the Kelly criterion. It describes how to size bets to maximise expected long-term logarithmic growth, given a model of the odds and payoffs. The amount to risk depends on the advantage you expect to have, as well as what you could lose. Background on the Kelly criterion

I’m wondering whether there’s a useful comparison with research time. A target takes hours, has some uncertain chance of producing a finding, and might return a bounty, a write-up, or knowledge I can use later.

I’m not suggesting that I can put those things into a formula and get a reliable schedule. At this point, it’s a question I want to explore.

What the comparison makes me ask

First, how much do I actually know about the target? A large potential bounty is appealing, but it doesn’t tell me how likely I am to find something. Familiarity with a technology might make a less exciting target a better use of a particular session.

Second, how confident am I in that estimate? In investing, fractional Kelly means using less than the full allocation suggested by the model. The idea interests me because my own estimates of success are extremely uncertain, although reducing an allocation doesn’t magically make the underlying estimate correct.

Third, what am I giving up by continuing? Another hour on one target is an hour I can’t spend on a different target, a tool, or some background reading. I tend to notice the possible reward more readily than that opportunity cost.

Where the analogy gets awkward

Research doesn’t have a simple cash balance. A week without a finding can still teach me a technique or leave me with a better tool, and learning from one target can change my prospects on the next.

Kelly methods aren’t limited to one simplistic model of independent bets, but I haven’t built a model that captures those relationships. Calling every research decision “Kelly” would make the analogy sound much more developed than it is.

There’s also the effect of frustration. Continuing with the same hypothesis because I’m annoyed isn’t the same as continuing because the evidence looks promising. That’s something I want to notice in how I investigate a system, whatever framework I use to plan my time.

What I’d like to record

My next step is to revisit the Kelly discussion in Fortune’s Formula and keep a session log: the target, time spent, what I expected before starting, and what came out of it. I’d record learning separately from any bounty.

That won’t immediately give me dependable probabilities, but it might help me see which kinds of work are useful and when my expectations are unrealistic. It also connects with my notes on trading and risk, where the quality of an estimate matters just as much as knowing the formula.

Paths that lead here

  • Feeling behind in cybersecurity · My note on allocating research time asks a related question: how much attention should I give one problem before moving on?
  • Learning about quant trading · The Kelly criterion is one connection I'm interested in, although understanding the assumptions will matter more than memorising the formula.
  • The birthday paradox · It's a useful habit to carry into other probability questions I'm working through, where my first impression of the odds may leave out more than I realise.
  • Thinking like an attacker · That's one reason I've been trying out connections between security research and probability, while keeping in mind that a useful analogy still needs to be tested.
  • Understanding trading edge and risk · This connects with the Kelly criterion, which made me interested in how assumptions about odds affect the amount someone chooses to risk.

Where this note points

More from these beds