Somewhere in your home there is a small graveyard. A guitar in the corner that has not been tuned since spring. A stack of books still in shrink wrap. A domain name registered for a podcast that recorded one episode. Every time you notice it, you feel a specific mix of nostalgia and shame, and the internet has a name for the feeling: the hobby graveyard.
The Brazilian creator Leo Xavier opens a recent video with exactly that image, then does something more useful than the genre usually manages. Instead of another lecture about discipline, he sketches a resource-allocation model, calls it Sistema M, and argues that most focus advice is solving the wrong problem for people whose curiosity refuses to sit still. I watched it, traced the claims back to where they actually come from, and rebuilt the model as something closer to a systems diagram than a pep talk. This is the engineering version.
The graveyard is not a diagnosis
Society has a script for this. Pick one thing. Focus. Anyone with a shelf of half-finished projects has heard some version of “you’re scattered” often enough to start believing it is a character flaw rather than a pattern with a name.
It has a name. Barbara Sher started calling people like this “scanners” in the 1990s, after years of watching capable, high-output clients fail every piece of standard career advice because the advice assumed one true calling. Her 2006 book Refuse to Choose! is a field manual for the type: people who go deep into a subject, extract what they need, and move on, not out of weakness but because that is how their attention is actually shaped. Emilie Wapnick rebranded the same profile as “multipotentialite” in a 2015 TEDx Bend talk that has since been watched millions of times, and the reframe stuck: rapid learning, idea synthesis across fields, and adaptability are the actual superpowers this profile tends to have, not a lesser version of specialist focus.
None of that is a controlled trial. It is a named, widely recognized pattern, not a diagnosis with a base rate. What both writers leave open is the harder question: fine, you are wired this way, now what do you actually do with a finite week. That is where naming the pattern stops being useful and allocating around it starts.
Two failure modes, and reinforcement learning already named them
Run this pattern without a system and it tends to fail one of two ways.
Type one starts everything and finishes nothing. Python course to 30 percent, then graphic design, then content creation, then the gym, each one abandoned the moment the initial dopamine hit fades. Type two never starts at all: 47 browser tabs, a dozen saved courses, three hours of comparison research that ends in nothing shipped. If you have ever watched someone open Netflix and closed it 20 minutes later having selected nothing, you have watched type two in the wild.
Both are the same failure viewed from opposite sides, and reinforcement learning has a precise vocabulary for it: the explore-exploit tradeoff. Every unit of time is a turn you cannot spend twice, split between exploiting an option you already understand and exploring one you do not. Type one over-explores, switching before an option has been given enough turns to reveal its true payoff. Type two over-explores in a different sense, spending the entire turn budget gathering information and never actually pulling a lever. John Gittins solved a clean version of this problem in 1979 with an index that ranks options by the value of what pulling them would teach you, not just what they are expected to pay (Gittins, 1979). Brian Christian and Tom Griffiths popularized the practical shape of the answer in Algorithms to Live By (2016): explore hard early, for something close to the first third of your available window, then commit and exploit. Committing is not the opposite of exploring. It is what exploring is for.
The multi-armed bandit problem isn’t solved by picking one lever forever, it’s solved with an explore budget: spend it deliberately, then commit once it surfaces a winner. The failure in both hobby-graveyard patterns has nothing to do with curiosity itself. It comes from running exploration and exploitation off the same unbounded budget, with no rule for switching between them.
The curve is older than the creator economy
Here is the part of the video worth taking seriously on its own terms, because the graph it borrows is better sourced than the borrowing suggests.
Alex Hormozi has popularized a five-stage curve of felt competence for anyone starting something hard: uninformed optimism, informed pessimism, a trough where most people quit, informed optimism, then success. It is not his invention. Don Kelley and Daryl Conner mapped the same five stages in the 1979 Annual Handbook for Group Facilitators, built from organizational-change consulting through the mid-1970s, under the name the Emotional Cycle of Change: uninformed optimism, informed pessimism, hopeful realism, informed optimism, completion. The phrase “valley of despair” is not in Kelley and Conner’s original text. It is the nickname the trough earned later, popular enough that it now overshadows the model’s real name.
That lineage matters for two reasons. First, it means the curve was never a hustle-culture invention designed to sell you a course; it was a practitioner’s observation about how humans adjust to any voluntary change, mapped four and a half decades ago from a much less glamorous source: employees adjusting to a merger, or a new system rollout. Second, and more usefully, the trough doesn’t mean you picked the wrong pursuit. It’s the single most predictable stage in the whole cycle, the point where “this is harder than I thought” first collides with real difficulty, and by most practitioner accounts, where the largest share of attempts quietly stop.
Kelley and Conner’s model comes from consulting observation across decades of organizational change work, not from a randomized trial with an effect size attached. It has stayed in use in change-management training for 45 years because practitioners keep recognizing the shape in real projects, which is a meaningful kind of evidence, just not the kind you would cite in a meta-analysis. Use it as a map you can recognize yourself on, not as a formula.
What is actually training when you push the extra 10 percent
The video’s mechanism for crossing the trough is what it calls the “10 percent rule”: when you want to stop, do 10 percent more. Read one more page than planned. Train three more minutes. Debug one more attempt before closing the laptop. It credits this to training your anterior cingulate cortex, the brain region that lights up when you do something you would rather not.
The neuroanatomy is closer to correct than most pop-science shortcuts, and worth being precise about. The anterior cingulate cortex is a leading candidate for where the brain computes whether continued effort is worth it. Kurzban, Duckworth, Kable, and Myers (2013) argue it tracks the opportunity cost of your current task, the value of what you are not doing instead. Shenhav, Botvinick, and Cohen (2013) model it as computing the expected value of control: payoff from staying focused, minus the cost of staying focused. Both frameworks treat the discomfort of pushing through boredom as data, not noise: a live cost-benefit signal your brain is computing in real time.
But “train your ACC like a muscle” overstates what is actually happening, and the honest mechanism is arguably a better story. Robert Eisenberger’s learned industriousness research (Eisenberger, 1992) found that when effort is rewarded, the sensation of effort itself picks up secondary reward value, and that conditioning generalizes across unrelated tasks. Push through the boring page today, feel the small hit of finishing it, and the next boring task starts a little less aversive, not because a brain region got stronger, but because effort itself has been quietly re-conditioned as less bad. It is the same shape as progressive overload in strength training: a small, repeated, slightly-past-comfortable dose, not a single heroic push.
You are not training a brain region like a bicep. You are running a conditioning protocol on how aversive effort feels, one small rewarded dose at a time, and the effect is documented to transfer to tasks that have nothing to do with the one you practiced it on (Eisenberger, 1992). That is a stronger claim than “discipline,” because it predicts something specific: the domain you practice pushing through in barely matters. The transfer is the point.
The M-System as a control system, not a vibe
Here is where Leo Xavier’s actual model comes in, and where it earns the engineering treatment.
Most advice assumes one pillar. The specialist doctor who only does cardiac surgery, the backend engineer who never touches the frontend, gets rewarded for depth and told the rest is distraction. Wide-curiosity people who try to run that playbook end up shallow everywhere, the failure mode from the section above. The M-System proposes exactly two structural pillars plus one unstructured bridge between them, and the rule for which pillar gets which resources.
The stability pillar is load-bearing. It funds everything else, in time and in money, and the video’s readiness test is a genuinely good one: you are ready to allocate energy elsewhere once you can solve roughly 80 percent of your stability pillar’s routine problems without asking anyone. That is not a vibe check. It is a controllability threshold, the same kind of criterion you would use to decide whether a subsystem is stable enough to stop actively monitoring.
The growth pillar is the one deep pursuit under active construction, and only one runs at a time. It is timeboxed to a season (the video’s creator ran his YouTube channel as a growth pillar for 18 months) with an explicit graduation criterion set before it starts: it either becomes the new stability pillar, or it closes on purpose. Closing on purpose beats closing by neglect. One is an experiment that ran its course; the other is a project that died quietly, unfinished, for no better reason than that nobody made the call.
The curiosity lab is the unscheduled bridge between them, and it is where knowledge management stops being a nice-to-have and starts being the whole mechanism. Capture the spark when it hits. Archive it somewhere searchable. Revisit it later, ideally in a different context, so two unrelated notes can collide into something neither one was on its own. That loop, capture, archive, connect, is knowledge management by another name, and it is not incidental to the system, it is the reason abandoned interests stop being a graveyard and start being inventory.
If that loop sounds familiar, it should: it is exactly what the terms and concepts you are clicking through on this page are doing. This site’s own knowledge graph, the one that just linked “learned industriousness” back to a term page with its own citations, is the capture-archive-connect loop instrumented as software instead of a stack of Notion tags nobody revisits.
| Old framing | System framing |
|---|---|
| ”Just focus on one thing” | Run one exploit pillar plus a bounded, scheduled explore budget |
| ”You lack discipline” | You lack a conditioned reward signal for effort; discipline is downstream of learned industriousness |
| ”The valley of despair means you chose wrong” | The valley is a documented, 45-year-old stage most attempts pass through |
| ”Follow your passion, then find work” | Stability funds the portfolio; it does not have to be the passion itself |
| ”A hobby graveyard means you failed” | A hobby graveyard is an unindexed curiosity lab; the fix is capture, not shame |
Worth a footnote on “follow your passion, then find work”: Einstein wrote most of special relativity while employed at the Swiss Patent Office, a stability pillar that was undemanding enough to leave him the mental bandwidth for physics on the side. The job did not need to be the calling. It needed to fund the calling.
A worked portfolio, for a reader whose tabs look like mine
If your open-tabs situation spans robotics, 3D printing, data science, simulation engineering, software development, product management, psychology, neuroscience, behavioral science, and knowledge management (a specific list, chosen because it is a real one), the system does not ask you to pick a favorite. It asks you to sort that list into three roles, not eleven equal priorities.
| Role | Candidate domains from the list above | Governing rule |
|---|---|---|
| Stability pillar | Software development, product management | Fund everything else; pass the 80-percent-solo test before allocating elsewhere |
| Growth pillar, one at a time | Simulation engineering, robotics, or data science | 3 to 18 months, graduation criterion written down on day one |
| Curiosity lab, unscheduled | 3D printing, psychology, neuroscience, behavioral science, knowledge-management tooling | Captured on contact, archived weekly, promoted only after it resurfaces on its own twice |
Productivity is not a fourth pillar in that table, on purpose. It is the scheduling logic that runs the other three, the difference between a system and a list of good intentions.
What to run this quarter
- Name your stability pillar out loud, then take the 80-percent test honestly. If you are still asking someone else how to solve most of its recurring problems, that is real information, not a character flaw.
- Pick exactly one growth pillar for this season. Write its graduation criterion (becomes the new stability pillar, or closes) before you start, not three months in when sunk cost is doing the deciding for you.
- Build a two-minute capture habit for the curiosity lab. One inbox, not a new app for every spark. Archive it weekly. The value is in the revisit, not the capture.
- The next time you want to quit something in the growth pillar, apply the 10 percent rule once, deliberately, and notice afterward whether the urge to quit carried real information (wrong pillar) or was just the trough talking (Eisenberger, 1992; Kurzban et al., 2013).
- Review the whole system quarterly. Does the growth pillar graduate. Does anything in the curiosity lab earn promotion by resurfacing on its own, unprompted, more than once.
The guitar in the corner is not a verdict on your character. It is an unpromoted node in your curiosity lab that has not resurfaced twice yet. Tune it or don’t, but the graveyard framing was always the wrong data model.
Lucas Cazelli is CPO and Co-founder at North AI, where he builds neuroscience-inspired attention analytics for video. He writes about decision-making, cognitive science, and the places where an engineering background turns out to be useful outside engineering.
References and further reading
The source video and the multipotentiality framing
- Xavier, L. (2026). “Se você tem muitos interesses, você precisa de uma vida em M”. YouTube.
- Sher, B. (2006). Refuse to Choose!: A Revolutionary Program for Doing Everything You Love. Rodale Books.
- Wapnick, E. (2015). “Why Some of Us Don’t Have One True Calling”. TEDx Bend.
The change curve
- Kelley, D., and Conner, D. (1979). The Emotional Cycle of Change. In The 1979 Annual Handbook for Group Facilitators. University Associates.
- Hormozi, A. “The 5 Stages of Any Opportunity”. Skool / Acceleration Partners content, various dates.
Effort, the anterior cingulate cortex, and learned industriousness
- Kurzban, R., Duckworth, A., Kable, J. W., and Myers, J. (2013). An Opportunity Cost Model of Subjective Effort and Task Performance. Behavioral and Brain Sciences, 36(6), 661-679.
- Shenhav, A., Botvinick, M. M., and Cohen, J. D. (2013). The Expected Value of Control: An Integrative Theory of Anterior Cingulate Cortex Function. Neuron, 79(2), 217-240.
- Eisenberger, R. (1992). Learned Industriousness. Psychological Review, 99(2), 248-267.
Explore-exploit and portfolio allocation
- Gittins, J. C. (1979). Bandit Processes and Dynamic Allocation Indices. Journal of the Royal Statistical Society: Series B, 41(2), 148-164.
- Christian, B., and Griffiths, T. (2016). Algorithms to Live By: The Computer Science of Human Decisions. Henry Holt and Co.
- Markowitz, H. (1952). Portfolio Selection. The Journal of Finance, 7(1), 77-91.
Frequently asked questions
What is the M-System?
The M-System is a resource-allocation model for people with many rotating interests. It has three parts: a stability pillar that funds everything else and should run near-autonomously, a growth pillar that is the single active deep pursuit for the current season with an explicit graduation criterion, and a curiosity lab, an unscheduled bridge where new interests get captured and archived without any commitment attached.
Is the “valley of despair” a real, scientifically measured phenomenon?
It is a durable practitioner heuristic, not a controlled experimental finding. Don Kelley and Daryl Conner first mapped the five-stage curve in the 1979 Annual Handbook for Group Facilitators, based on organizational-change consulting, and it has stayed in use in change-management training for 45 years because practitioners keep recognizing the shape. Treat it as a useful map, not a formula with an effect size.
Does the “10 percent rule” actually train your brain, or is that a myth?
The specific claim that you are “training your anterior cingulate cortex like a muscle” overstates the evidence. What is well documented is learned industriousness (Eisenberger, 1992): rewarding effort makes the sensation of effort itself less aversive, and that conditioning transfers to unrelated tasks. The 10 percent rule likely works, just through a more precise mechanism than the popular framing suggests.
How is this different from just telling multipotentialites to focus?
Standard focus advice assumes one pillar and treats everything else as distraction, which fails people whose curiosity is a stable trait rather than a discipline problem. The M-System instead treats time as a fixed budget split across a known-good stability pillar, one deliberate growth bet, and a bounded exploration budget, the same structure reinforcement learning uses to solve the multi-armed bandit problem and modern portfolio theory uses to allocate capital across uncorrelated assets.