Rules for Overriding the Model

First published 26 Aug 2026 · Last verified 29 Aug 2026

Every investor who builds a written constitution eventually sits in front of it holding a decision that the rulebook did not anticipate. This is not a hypothetical. It happens to disciplined investors more often than it happens to careless ones, because disciplined investors are the ones who actually follow their systems closely enough to notice when reality has stepped outside the lines the system drew. The question this chapter answers is not whether you will ever face this moment — you will — but what you are allowed to do about it, and under what conditions, without quietly becoming the same impulsive trader your constitution was written to protect you from.

I am going to tell you about two afternoons from my own investing life on NEPSE, the Nepal Stock Exchange. On one afternoon I broke my own rules on purpose, in writing, with a plan, and it turned out to be the right call. On another afternoon I wanted to break my own rules just as badly, felt just as certain I was right, and refused — and that refusal saved me from a loss I would still be explaining to myself today. The difference between those two afternoons is the entire subject of this chapter. It is not about whether rules can ever be broken. It is about how to tell, in the heat of the moment, whether you are exercising a rare and disciplined power called an override, or simply watching your own discipline collapse while calling it something nobler.

Lesson 96.1 — The Difference Between an Override and a Collapse

Let's start with two words you have met earlier in this Part: the Canon Score and the constitution. Your Canon Score is the numeric output of the scoring model you built earlier in this book — a system that takes a company's fundamentals, valuation, governance signals, and sector context and turns them into a single comparable number, so that you are not deciding stock by stock from your gut every single day. Your constitution is the broader written document that governs how you behave as an investor: your position sizing rules, your buy and sell triggers, your rules about leverage, your rules about what you do during a NEPSE crash. Together they form what the rest of this Part has called your personal operating system — the thing that makes decisions for you when your own judgment, on a bad day, cannot be trusted.

A model override is a deliberate, pre-specified exception to that system, made for a reason you can write down in one clear sentence before you act, which the system itself did not have the information or the structure to account for. Notice every part of that definition, because every part is doing work. It is deliberate — you chose it, calmly, rather than being swept into it. It is pre-specified — meaning your constitution already contains a clause describing the category of situation in which an override is permitted, even though it could not predict the exact situation. And it is reasoned — you can state, in advance, in writing, exactly why this case falls outside what the model was built to judge.

Discipline collapse is something else entirely. It is an emotional decision — usually driven by fear, greed, excitement, social pressure, or the simple exhaustion of watching a price move without acting — that gets dressed up, after the fact or in the heat of the moment, in the language of an exception. The investor tells himself "this is different," "the model doesn't understand this stock," "everyone knows something I don't," or "the rules were made for normal days and today isn't normal." These sentences sound like reasoning. They are not. They are rationalisations, and the tell is always the same: they arrive after the emotional urge, not before it, and they cannot be written down as a clean, falsifiable justification that a calm friend reading it tomorrow would find convincing.

Here is a simple analogy. Think of a commercial airline pilot. Autopilot flies the plane for the overwhelming majority of a flight, because autopilot does not get tired, does not panic in mild turbulence, and does not "feel" that the runway is closer than it is. But every trained pilot also knows there are specific, rehearsed situations in which they must disengage autopilot and fly manually — a wind shear alert, a system malfunction, an instruction from air traffic control that autopilot cannot execute. Those situations are named in the manual before the flight ever begins. What a pilot must never do is disengage autopilot because they feel a sudden urge to show the plane who is in charge, or because a passenger shouted that they saw something out the window. The disengagement itself is not the problem. An undisciplined, unrehearsed disengagement is the problem.

Your Canon Score and your constitution exist precisely because your in-the-moment judgment, on the days that matter most, is the least trustworthy version of you. NEPSE investors know this instinctively from remittance season. A household that receives a lump sum from a family member working in the Gulf or Malaysia often makes its worst financial decisions in the excited week right after the money lands — not because the money is bad, but because sudden liquidity plus emotional relief is a famously poor environment for cold arithmetic. Your written system is supposed to be the version of you that shows up on that excited week and says, calmly, "we follow the plan." An override clause exists so that the system is not so rigid that it becomes stupid in the face of genuinely new information. But the override clause must be narrower than the system itself, or it swallows the system whole.

Lesson 96.2 — Why Systems Need an Escape Hatch, But Only a Narrow One

If a constitution allowed no exceptions at all, it would eventually break — not bend, break — because reality occasionally produces situations no model builder foresaw. Imagine a Canon Score model that scores a bank primarily on loan growth, net interest margin, and capital adequacy ratio (a regulatory measure of how much capital a bank holds against its risk-weighted assets, set and monitored by Nepal Rastra Bank, NRB, the central bank). Now imagine NRB announces an unusual one-time directive requiring all commercial banks to write back a specific provisioning charge because of a policy change, instantly improving the reported profit of every bank in the sector for one quarter only. Your model, built on historical ratios, might score this as a genuine earnings improvement and tell you to buy aggressively. You, the human, might know — because you read the regulatory circular carefully — that this is a one-time accounting effect that will reverse next quarter and tells you nothing about the bank's real earning power. Refusing to act on your own better information here, purely because "the model says so," is not discipline. It is discipline mistaken for stupidity.

This is why every well-built constitution needs what I call an escape hatch: a formally defined, rarely used, heavily guarded mechanism for departing from the model's output. But notice the phrase "rarely used, heavily guarded." An escape hatch on an airplane is not a second door you use whenever the first one has a queue. It exists for emergencies, it requires a specific lever to be pulled, and pulling it without cause is itself a serious violation, sometimes a criminal one. Your override clause needs the same character. It should be something you almost never touch, something that requires visible, deliberate effort to activate, and something whose overuse is itself evidence that your system has failed.

Here is a useful distinction borrowed from decision theory that fits neatly onto this problem: the difference between a one-way door and a two-way door. A two-way door decision is reversible at low cost — if you are wrong, you walk back through it and the damage is small. A one-way door decision is expensive or impossible to reverse. Most day-to-day portfolio choices — trimming a position by ten percent, adding a small tranche to a stock you already hold — are two-way doors, and your constitution can afford to be a little flexible about them without formal overrides, because a mistake is cheap to correct. A genuine override, by contrast, is almost always a one-way door decision dressed as urgent: sell a core holding entirely, buy a large new position outside your normal universe, abandon a stop-loss rule mid-crash. Because these are expensive to reverse, they deserve the heaviest procedural friction, not the least.

I want to introduce a concept here that I use in my own practice: the override budget. Just as your constitution should specify a maximum percentage of your portfolio you are willing to lose in a year (a subject the next chapter covers in full), it should also specify a maximum number of overrides you permit yourself in a given period — I use one every calendar quarter as a soft ceiling, meaning if I am reaching for a second override within three months, I am required to stop and treat that as a five-alarm warning about my own state of mind, not a coincidence of an unusually eventful market. A budget forces you to ration a scarce resource, and scarcity is exactly the quality that keeps an escape hatch from becoming a second front door.

Lesson 96.3 — The Pre-Override Checklist

Now the operational heart of this chapter: the specific, narrow checklist that must be satisfied, in full, before any override is permitted. I present this as a checklist deliberately, in the same spirit as a pilot's pre-flight checklist or a surgeon's pre-operative checklist — professions where the cost of skipping a step under time pressure is exactly why the checklist exists in written form rather than living only in memory.

The first requirement is that you write down the reason before you act, not after. This single rule does more work than any other item on this list. The physical act of writing forces a pause, and the pause is where discipline collapse gets caught. If you cannot produce a clear written sentence — before the trade, not as a justification afterward — stating exactly which named exception in your constitution this situation falls under and why, you do not have an override. You have an impulse. I keep a single page in the back of my paper trading journal headed "Override Log," and the rule I have set for myself is brutal in its simplicity: no entry in that log, no override trade. If I cannot write the sentence calmly, I am not calm enough to be trusted with the decision, and the checklist has already done its job by stopping me.

The second requirement is a mandatory cooling-off period. For any override above a small threshold size, I require myself to wait — in my own constitution the number is twenty-four hours for anything affecting more than two percent of the portfolio — between writing the justification and executing the trade. Almost everything that feels like a screaming emergency at 11 a.m. looks calmer, and often simply wrong, by the next morning. A rights issue deadline or a genuine time-boxed corporate event might compress this window, but even then the rule requires some non-zero delay, because zero delay is indistinguishable from acting on pure adrenaline.

The third requirement is a second opinion from someone who is not inside your excitement. This does not need to be a professional advisor, though it can be. It can be a spouse, a fellow investor in your circle, or, in my own case, a friend from my university days who now works in a completely unrelated field and has no stake in NEPSE at all — which is precisely why his skepticism is valuable. The point of the second opinion is not that the other person has better information than you. The point is that explaining your reasoning out loud to someone with no emotional investment in the trade exposes weak reasoning almost immediately. If you cannot explain the override clearly enough for a reasonably intelligent outsider to understand why it is not just excitement, that is diagnostic information about the quality of the override itself.

The fourth requirement is a hard position-size cap on the override trade, set in advance and independent of how confident you feel in the moment. My own rule caps any override-driven trade at half of what my normal position-sizing formula would otherwise allow for a position of that conviction level. The logic is straightforward: an override, by definition, is a situation your model was not built to evaluate, which means your ordinary confidence calibration does not apply. You are, in a very real sense, flying without your normal instruments for this one decision, and a pilot flying without instruments reduces speed, not increases it.

The fifth requirement is that the override must be traceable to a named clause in your constitution, not invented on the spot. Your constitution, from earlier chapters in this Part, should already contain a short, closed list of situation categories — corporate actions the model cannot score, confirmed regulatory windfalls, and so on — for which overrides are even eligible to be considered. If your justification requires you to invent a brand-new category of exception that has never appeared in your written document before, that itself is close to disqualifying, and should trigger a much higher bar of scrutiny, ideally requiring you to first amend your constitution through its formal amendment process (covered earlier in this Part) rather than acting first and amending later.

Lesson 96.4 — Legitimate Override Scenarios

Let me now walk through the category of situations where an override clause earns its keep, and then tell you about the one I actually used.

The clearest legitimate category is the confirmed one-off regulatory or corporate windfall — an event that is announced, verified through an official source, and structurally guaranteed to happen, but which your Canon Score model has no mechanism to price because it was built around recurring fundamentals, not one-time events. Examples on NEPSE include a company receiving confirmed compensation from the government for land acquisition tied to a hydropower or infrastructure project, a confirmed merger swap ratio between two listed banks or financial institutions that creates a temporary, calculable arbitrage gap between the merging entities' share prices, or a court ruling — already final, not under further appeal — that resolves a long-running dispute in a company's favour and releases previously provisioned funds back to the balance sheet.

The second legitimate category is a company-specific structural event the model was never built to score at all, as opposed to an event the model scores badly. Your Canon Score, like most fundamentals-based models, is built around the assumption of a going concern operating its ordinary business — the same business, roughly, next year as this year. A spin-off (where a company separates one division into an independently listed entity and distributes shares of the new entity to existing shareholders), a large share buyback funded from a demonstrated cash surplus, a strategic asset sale that permanently changes the balance sheet, or a regulator-mandated capital increase that dilutes shares but also strengthens the institution — these are structural discontinuities, not deteriorations or improvements in ordinary operating performance, and no scoring model built on trailing ratios can be expected to interpret them correctly on the day they are announced.

The third legitimate category, narrower and requiring the highest bar of verification, is a confirmed regulatory or macro policy shift that is specific enough, and certain enough, to be actionable before the model's normal update cycle would catch it. NRB monetary policy announcements, changes to loan-to-value ratios for margin lending against shares, or SEBON (the Securities Board of Nepal, the capital markets regulator) directives on sectoral exposure limits for institutional investors sometimes create a short window where a well-informed, careful reader of the official circular has real information the broader market has not yet absorbed. I want to be very clear that this category is the most dangerous of the three, because it is also the category most easily counterfeited by a rumour dressed up as regulatory insight — which is exactly why the checklist's requirement to trace your reasoning to a verified, published, official source, not a broker's WhatsApp forward, matters most here.

Now let me tell you about the one I actually used. Three years ago I held a position in a hydropower company whose Canon Score had been sitting in a comfortable, unremarkable middle band for over a year — decent generation numbers, ordinary debt load, nothing that triggered a buy or sell signal either way. Then the company disclosed, through NEPSE's official corporate announcement system, that it had reached a final, court-approved settlement with a government agency over land compensation dating back to its original construction phase — a dispute that had been provisioned as a liability on the balance sheet for four years. The settlement was smaller than the provisioned amount, meaning the company would release a material sum back into retained earnings in a single quarter, verified by the settlement document itself, not by rumour.

My model had no clause for this. It scores hydropower companies on generation capacity utilised, power purchase agreement tariffs, and debt servicing coverage — none of which this event touched. I wrote, before doing anything else, a single sentence in my Override Log: "Confirmed final court settlement releases NPR [amount] in previously provisioned liability back to equity, verified via company's NEPSE disclosure filing dated [date]; this is a one-off balance sheet event my Canon Score was not built to capture; category: confirmed regulatory/legal windfall." I let it sit for the required cooling-off period, sent the disclosure document itself to my university friend and asked him, with no context about my portfolio, whether he thought the document said what I thought it said. He agreed it was unambiguous. I sized the position at half of what my normal conviction-based sizing would have allowed for a position with this clear a catalyst, added to my existing holding rather than opening a fresh one, and logged the entire sequence.

The position appreciated meaningfully over the following two quarters as the market gradually recognised the balance sheet improvement and — separately, and not something I had predicted or was relying on — the company's next dividend declaration reflected the stronger equity base. I want to be honest that a good outcome does not, by itself, prove the override was correctly made; a badly-reasoned override that happens to work is still badly reasoned, and I would have judged this one the same way if it had gone nowhere. What made it legitimate was not the profit. It was that every item on the checklist was satisfiable honestly, in writing, before I acted, using a verified public document rather than a feeling.

Lesson 96.5 — Illegitimate Override Attempts, and the One I Refused

Now the harder half of this lesson, because refusing an override that your gut insists is obvious is considerably more uncomfortable than making one you can defend on paper.

The clearest illegitimate category is chasing a tip — acting on a piece of information whose only source is a person's confidence, not a verifiable document. NEPSE, like most emerging and frontier markets with a large base of retail participants, has an active rumour economy: tea-shop talk near the brokerage houses in New Baneshwor, Viber groups promising an "insider word" on a company, remittance-funded new entrants eager for a shortcut who pass along whatever they were told by a cousin's friend who "works at the company." A tip is not disqualified from being true. It is disqualified from being actionable through the override mechanism, because the override checklist specifically requires a verified public source, and a tip, by definition, has not passed through any verification you can point to later and defend.

The second clearest illegitimate category is panic-selling during a circuit-breaker day. NEPSE, like most exchanges, uses circuit breakers — automatic, rule-based trading halts triggered when the index or an individual stock moves beyond a specified percentage in a single session, designed to force a cooling-off pause on the entire market rather than allow panic to compound itself in real time. The circuit breaker exists precisely because regulators understand that extreme single-day moves produce exactly the emotional conditions in which investors make their worst decisions. Treating a circuit-breaker halt as new information that justifies overriding your constitution's stop-loss or holding rules gets the causality backwards: the halt is a symptom of collective panic, not a fact about the company's fundamentals, and reacting to it as though it were company-specific news is discipline collapse wearing the mask of prudence ("I'm just managing risk actively").

The third illegitimate category, subtler than the first two, is FOMO dressed as a corporate-action override — noticing that a stock is rallying hard, discovering after the fact that there was some corporate action attached to the rally, and retroactively constructing a justification that resembles the legitimate categories in Lesson 96.4 to license a purchase you actually wanted to make for the much simpler reason that the chart is going up and you feel left out. The tell here is sequence: in a legitimate override, the verified information comes first and the trade follows from it. In this illegitimate version, the desire to buy comes first, and the "verified information" is assembled afterward to fit.

Here is the one I refused. About eighteen months ago, a friend from my neighbourhood — genuinely well-meaning, genuinely excited — called to tell me that a manufacturing company I did not hold had "guaranteed news coming," something about a large export order, and that his brother-in-law, who apparently had a contact inside the company's finance department, said the stock was about to move sharply. The stock had already climbed nearly fifteen percent over the prior week on unusually heavy volume, which made the story feel more credible, not less, in the way that a rally always seems to confirm a rumour that arrived to explain it.

I wanted this to be true in a way that, looking back, I find slightly embarrassing to admit. I opened my constitution and tried, honestly, to write the required override sentence. I got as far as "confirmed export order," and stopped, because I had no confirmation at all — only a secondhand account of a conversation with someone I had never met, about a filing that did not exist on NEPSE's disclosure system, SEBON's website, or the company's own investor page. I could not name a clause in my constitution this fit under, because "a friend's brother-in-law's contact says so" was not, and will never be, a category I had written into the legitimate list. I sent the story to my usual second opinion, who asked one question I did not have an answer to: "has the company itself said anything?" It had not.

I did not buy. Nine days later, the stock gave back the entire fifteen percent gain in three sessions after the company issued a clarification stating it had no material undisclosed information to report — the standard disclosure a company makes on NEPSE when a regulator asks it to explain unusual price movement. The export order, as far as I have ever been able to determine, never existed in any form that reached an official filing. I want to be equally honest here that avoiding a loss is not proof the refusal was correct in principle, any more than the hydropower profit proved the override was correct in principle — sometimes tips are true and sometimes verified overrides lose money regardless. What made the refusal correct was that the checklist could not be honestly completed, and I stopped at the point where it failed rather than forcing the remaining steps to fit.

Lesson 96.6 — Logging, Reviewing, and Knowing When the Constitution Itself Must Change

An override is not finished when the trade is placed. It is finished when it has been logged, and it is only truly finished when it has been reviewed later against what actually happened, honestly, whether the outcome was good or bad. This closing discipline is what separates an investor who uses the override mechanism as a genuine safety valve from one who uses it as a permission slip they quietly stop examining once the trade has been placed.

Keep your Override Log as a permanent record, not something you tidy away once a position is closed. Each entry should carry the date, the written justification produced before the trade, the named constitutional clause it falls under, who gave the second opinion and what they said, the position size relative to your normal sizing formula, and — added later, after the position is closed — the actual outcome and a short honest note on whether, rereading the original justification with distance, it still holds up as sound reasoning independent of the profit or loss. This last step matters enormously and is the one investors skip most often, because it is uncomfortable to write "this reasoning was actually weak" about a trade that happened to make money, and even more uncomfortable to write "this reasoning was actually sound" about a trade that lost money anyway. Both entries are valuable precisely because they resist the very human urge to judge a decision purely by its result.

Review your Override Log at a fixed interval — I do mine every six months, alongside my broader constitutional review — and ask three questions of the whole set. First, how many overrides did I attempt versus how many did I actually execute, and is the gap healthy — meaning did the checklist catch weak reasoning before it became a trade, the way it caught the export-order tip? Second, of the overrides I executed, does the pattern of categories match the legitimate list in Lesson 96.4, or am I quietly stretching the definitions to cover situations that do not really belong there? Third, and most important, is the frequency of overrides increasing over time? An increasing frequency almost never means the world has started producing more genuine exceptions. It almost always means one of two things: either your Canon Score model has a structural blind spot that keeps recurring — in which case the correct response is not to keep overriding it case by case, but to formally amend the model itself through the amendment process covered earlier in this Part, so the exception becomes a permanent, principled part of the system rather than a repeated one-off — or your emotional discipline is eroding, and the override clause has quietly become your escape hatch of first resort rather than last resort, which calls for a much more serious conversation with yourself, and possibly with your second-opinion partner, about what is actually going on.

There is a final, quieter reason to keep this log with care: it is the only honest evidence you will ever have, months or years later, of whether the override clause is earning its place in your constitution at all. Some investors, after a few years of disciplined logging, discover that their overrides as a group have performed no better, and sometimes worse, than simply trusting the model every time — in which case the honest, if humbling, conclusion is to narrow the clause further, or in rare cases remove it altogether and accept the model's judgment even in edge cases, on the theory that a slightly wrong model followed with total consistency beats a slightly-less-wrong model undermined by human intervention every few months. Other investors find the opposite — that their handful of well-checklisted overrides meaningfully outperformed what pure model-following would have produced, which justifies keeping the clause but also raises the bar for keeping it narrow, since its value depends entirely on its rarity. Either finding is useful. What is never useful is skipping the review, because a constitution nobody audits is, in practice, no constitution at all — it is a document you consult when convenient and ignore when it is not, which is discipline collapse with better handwriting.