[Assumed Cause]

Definition #

[Assumed Cause] is the operator failure where the causal read runs correctly and still produces the wrong answer, because the variables tested were never the causal variables.

The operator does not skip the discipline. He applies it. He looks at a past event, names what he believes caused it, tests whether those causes have moved, finds them stable, and repeats the decision on the same criteria. Every step of that sequence is the right sequence. The method is sound. The inputs are not. He assumed the cause instead of establishing it, then ran a rigorous stability test on the assumption.

[Assumed Cause] is a failure-mode child of [Causal Read]. [Causal Read] is the physics every operator decision runs on — current signal plus causal model produces an anticipated future state that governs the action. The causal model is built from the operator’s own accumulated action-outcome pairs. [Assumed Cause] names what happens when those pairs were mislabeled at the moment they were recorded: the operator credited the wrong variable for the outcome, filed the pair, and every read that model produces from that point forward carries the error forward as if it were established. It is not a general reasoning error and it is not a posture problem. It is the specific way the discipline breaks while it is being run.

The name carries no Transactional lead word. This is an operator read failure, not a Road 1 mechanism — the same scope logic that leaves [Static Decline] unprefixed.

Mechanism #

The five-beat sequence. An outcome lands. The operator looks back at it. He names a cause. He tests whether that cause still holds. It holds, so he repeats the decision. Five beats, and the failure is entirely inside beat three, which is the only beat nobody inspects. Beats four and five are the disciplined part, and they are what the operator will point to when asked whether he did the work. He did do the work. He did it on a variable he never established.

Naming is not establishing. This is the whole term. Naming a cause is a single act of attribution the operator performs in seconds, usually while the outcome is still fresh and the most recent change he made is still the most available thing in his head. Establishing a cause is a body of work: generate the candidates that could have produced this outcome, size each one against the magnitude of the effect, check whether each candidate was present in periods that did not produce the outcome, and go looking for the evidence that would kill your favorite. Operators run the first and call it the second. The gap between them is where [Assumed Cause] lives, and the gap is invisible because both end in the same sentence — “the new item drove the quarter” — spoken with the same confidence either way.

The recency trap in the causal model. [Causal Read] already names how the model gets built: from action-outcome pairs weighted by how recent and how vivid the observation was. That weighting is what selects the assumed cause. The operator changed one thing on purpose and a hundred things changed around him without asking. The thing he changed on purpose is the one he can see, the one he can name, the one he is emotionally invested in, and the one that arrives first when he goes looking for an explanation. The competitor who closed two blocks over does not present itself as a candidate, because it was not a decision the operator made. The eleven days of unusual weather do not present themselves. The one strong lead who was holding the whole cast together does not present herself, because she never announced that she was the mechanism. So the operator’s own deliberate act wins the attribution by default, every time, on availability rather than on evidence.

Three shapes you will recognize. He attributes the strong quarter to the new menu item when the actual causes were a competitor closing and a stretch of unusual weather — so he builds the next three specials off a mechanism that never fired, and when the competitor’s space reopens he reads the drop as an execution problem. He attributes the low turnover to the training program when the cause was one strong lead who has since left — so he funds the program, protects the program, and cannot explain why turnover tripled while the program ran untouched. He attributes the check-average lift to the upsell script when the cause was a price increase that landed the same month — so he coaches harder on the script, drills the cast on language that never moved anything, burns cast credibility on a mechanism that does not exist, and never runs the price read that would have told him what he actually has.

Every shape has the same signature. One deliberate operator act, one or more undeclared co-occurring changes, and an outcome that arrived in the same window. The operator’s act gets the credit because it is the only candidate anyone bothered to name. Then the act gets funded, defended, and taught, and the real cause goes unmanaged — which is the second cost and the larger one. When the actual cause is a strong lead, [Assumed Cause] does not just waste the training budget. It leaves the lead unretained, unpaid for what she was actually producing, and available to somebody else.

Why it survives the audit. Nothing in the work looks wrong. The rigor is real. The documentation is clean. The date is on it, the variables are listed, the stability test is recorded, the review passes. And the decision is bad. This is the part that earns the term its own entry: the failure is invisible precisely because the discipline was followed. The operator’s own diligence is what makes him confident in the wrong answer. Review processes are built to check whether the work was done, and the work was done. Nobody in the room is auditing the premise the whole audit ran on. An operator who skips the read knows, somewhere, that he guessed. He holds the decision loosely. He watches it. An operator running [Assumed Cause] holds it tightly, because he earned it. That is why it does more damage than sloppiness — sloppiness comes with its own hedge attached, and this does not.

The confidence inversion. The determinations most likely to be running on an assumed cause are the ones the operator would defend hardest, because they are the ones that got named early, worked once, and were never re-established. Confidence is the marker, not the defense. When an operator says he is certain what drove a result, the correct next question is how many candidates he considered before he became certain. An operator who considered exactly one did not establish a cause. He assumed it and then spent years accumulating a track record on top of it.

The recognizable moment. It shows up as a because. “Covers are up because of the new patio hours.” “Retention held because we run better pre-shifts.” “Food cost came down because we tightened portions.” Every one of those may be true. None of them has been established, and the operator will not know which is which until somebody makes him list the other things that changed in the same window. It shows up in a P&L review where a favorable variance gets a one-line explanation and the meeting moves on. It shows up in the pre-shift where a lead teaches a technique as the reason something worked. It shows up worst in the annual plan, where last year’s attributed wins get scaled up and funded — and scaling an assumed cause is how an operation ends up spending its largest number on its weakest mechanism.

Load-Bearing Distinction #

Not [Causal Read]. [Causal Read] is the parent discipline and the physics underneath every decision the operator makes. [Assumed Cause] is one specific way that discipline breaks — not the only way, and not the way most operators expect. [Causal Read] already names the failure of the operator whose model never updates against observation, and the operator who attributes every miss to the market. [Assumed Cause] names something narrower and harder to catch: the operator whose model updates faithfully, on cadence, against observations that were labeled wrong on intake. The discipline is running. The intake is corrupt. Collapse the two and the correction gets aimed at the wrong layer — the operator runs the read more often and more rigorously, which multiplies the error instead of finding it.

Not [Case Study Reduction]. The discipline was never applied there. [Case Study Reduction] is the operator taking a retrospective account of someone else’s operation and running it forward as an executable path — the record substituted for the mechanism, borrowed from a building the operator does not stand in. [Assumed Cause] runs on the operator’s own building, his own cast, his own numbers, and his own causal read, honestly conducted. That distinction changes the fix. The fix for [Case Study Reduction] is to stop borrowing and run your own read. That fix does nothing here, because the operator already ran his own read. The fix here is upstream of the read: establish the cause before you test it.

Not [Orphaned Act]. These two get collapsed constantly and the relationship has to be stated exactly. [Orphaned Act] is the terminal negative outcome of [The Operator’s Filter] — a decision that cleared fewer than all ten spokes, disconnected from the discipline that would have made it a [Conscious Investment]. Its signature is process skipped. [Assumed Cause] has the opposite signature: process run. They sit under different parents, so they are not siblings in the family sense, and neither one produces the other by definition. What connects them is a chain, and the chain runs in one direction. An [Orphaned Act] that happens to work is an outcome with no established cause attached to it. [Assumed Cause] is the mechanism that then names that outcome’s cause — always the operator’s own act, because that is the only candidate in the room — and promotes an orphaned decision into a standing formula. So [Assumed Cause] is not the mechanism that produces an [Orphaned Act]; the filter-skip produces that. [Assumed Cause] is the mechanism that converts a lucky [Orphaned Act] into a repeated one. It is also why the filter does not catch it: the ten spokes test the decision in front of the operator, and [Assumed Cause] corrupts the premise every spoke runs on. A decision can clear all ten spokes cleanly and still be built on an attribution nobody established.

Not [Egotistical Ignorance]. That is a posture problem — the operator treating a snapshot of knowledge as complete, permanent, and sufficient, refusing to ask whether it still holds because the track record feels like authority. It is a curiosity failure, and it is diagnosable in the person. [Assumed Cause] afflicts the humble, diligent operator just as readily, which is the point and the reason it needs its own name. The operator running [Assumed Cause] is asking whether it still holds. He is asking on cadence, in writing, with the variables listed. He is asking about the wrong variables. Curiosity does not fix it. More curiosity aimed at an assumed cause produces a deeper, better-documented commitment to the assumption. The two also stack: an [Egotistical Ignorance] posture will refuse the audit that would surface an [Assumed Cause], which makes the posture an aggravator and not the mechanism.

Not [Constant Expiry]. [Constant Expiry] says the determination expired — it was accurate at settlement and the universe it was accurate in no longer exists. [Assumed Cause] says the determination was wrong about its own causes from the start. Different failures, and the diagnostics point in different directions. Under [Constant Expiry] the question is what has moved since. Under [Assumed Cause] the question is what was ever true. An operator can commit both at once, and most do: an attribution that was never established, running on variables that have also since moved, tested annually against the wrong list. Worse, the [Constant Expiry] discipline actively hides this one. The variable-stability test is the legitimate extraction from a past event, and running it on an assumed cause produces a clean pass, on the record, with a date on it. The rigor certifies the error.

Not [Outcomes Formula]. [Outcomes Formula] runs backward from outcome to thinking: if the outcome came in right the thinking was right, if it came in wrong the thinking was wrong, and luck is not a category it recognizes. It is a verdict instrument. [Assumed Cause] is a location instrument. The Formula tells the operator that the thinking was wrong; it does not tell him where. Its own mechanism already names the likely address — an assumption the operator did not know he was making — and [Assumed Cause] is that address named as a term. The two run together and neither substitutes for the other. Where they must not be collapsed is the right-outcome case. The Formula grades the thinking by the outcome, so an outcome that comes in right reads as thinking that held. [Assumed Cause] is how thinking that did not hold produces a right outcome anyway, once, and gets graded as sound. The Formula is not wrong about that — it is a verdict on the thinking that produced the outcome, and it makes no claim about the attribution the operator wrote down afterward. [Assumed Cause] governs the attribution layer, which is a separate act performed after the outcome landed.

Not 0055 Probability Over Time Not Certainty Per Event. That entry answers the variance objection at population scale: the principle claims probability over time, not certainty per event, which is why a defaulted operator can catch a wave and a designed operator can absorb a shock without either case disturbing the law. Different object entirely. That argument is about how a predictive law behaves across a population of operations over years. [Assumed Cause] is about what one operator did with one outcome inside one building on one afternoon when he decided what caused it. The variance argument does not answer this failure and this term does not re-derive that one. Where they meet is a warning: an operator who has read 0055 has a ready explanation for any single result, and “variance” is as available a misattribution as “the new item.”

Not [The ReRead]. This one is not primarily a distinction — it is part of the response, and it needs its boundary named anyway. [The ReRead] is the step in the operating loop that sees what execution actually produced versus what the design intended, and it feeds [Recalibration]. It is the closest thing in the framework to a catch mechanism for this failure, because it is the one step whose whole job is to look again. But a clean re-read does not catch [Assumed Cause] by itself. The re-read compares result to design intent. When the result matches the intent, the re-read closes green — and [Assumed Cause] is exactly the case where the result matched and the reason it matched was never established. So the re-read has to be extended to be useful here: not only did we produce what we designed, but did we produce it for the reason we designed it. That second question is the [Assumed Cause] question, and without it the re-read confirms the attribution instead of testing it.

Not [The Reader’s Unread Bias]. The closest structural cousin in the framework, and the one an operator is most likely to collapse this into, because both describe a rigorous read that is broken anyway. [The Reader’s Unread Bias] is the filter underneath the read — the operator’s fatigue, carry, identity story, and inflated or deflated certainty coloring the signal on the way in. Its object is the reader. [Assumed Cause] takes clean signal and mislabels the world: the object is the causal variable set out in the building, not the state of the operator holding the clipboard. An operator can run the filter read perfectly, arrive rested, unattached, and honestly dispassionate, and still credit the menu item for a quarter the weather bought. Both terms produce the same sentence — the discipline ran and the discipline was still broken — and they require different corrections. One is corrected by reading the reader. The other is corrected by generating candidate causes out in the world and requiring disconfirmation.

Not [The Biased Read]. [The Biased Read] names the condition that no operator reads their own building without bias, and it inventories the contaminants, attribution bias among them. That attribution bias is directional and self-serving: credit the strong period to your own leadership, blame the weak one on circumstance. [Assumed Cause] is not directional and does not require ego. The operator who credits his strong quarter to a competitor’s closing when the actual cause was his own repositioning has run [Assumed Cause] against himself, humbly, and lost the same money. Naming the bias inventory does not close this. An operator can prosecute all five biases, name each one out loud, and still hand the outcome to the one candidate he generated.

Not [Judgment Distortion]. Anchoring, overconfidence bias, and loss aversion name how judgment bends away from analysis through invisible shortcuts. Those are pressures on the read. [Assumed Cause] is a defect in a specific artifact the read produced — the labeled action-outcome pair sitting inside the causal model, wrong from the day it was filed. Anchoring may help select the assumed cause; overconfidence may protect it from challenge. Neither is the thing itself. Correct all three distortions and an operator with one candidate cause on his list still gets the wrong answer, calmly and without distortion.

Not [Default Perspective]. [Default Perspective] is the closed examination loop — real examination running inside an installed interpretive frame that cannot surface itself, every finding riding back to confirm the installation. That is a frame problem operating at the Perspective layer across everything the operator looks at. [Assumed Cause] is narrower and lives one layer down, at a single attribution inside a single read. An operator who has broken out of the installed frame, who has genuinely gained altitude on his own perspective, will still assume causes — because generating competing candidates is a separate discipline from examining your frame, and nothing about altitude supplies it.

Not [Hacksterism]. [Hacksterism] is the shortcut posture that reaches for someone else’s settled record to avoid paying the ongoing cost of the read. [Assumed Cause] pays the cost. It shows up in the operator who refuses hacks, runs his own reads, keeps his own records, and builds his whole operating discipline in-house — and who has never once been asked how many causes he considered before he picked one. That is why it cannot be fixed by posture correction. There is no shortcut to name and no shortcut to refuse.

Without this term named, operators default to auditing method and never auditing inputs. The review asks whether the read was run, whether it was documented, whether the variables were tested, whether the criteria transferred — and every one of those can pass while the whole exercise stands on an attribution that took four seconds and was never contested. The operation then does the most expensive thing available to it: it funds, scales, teaches, and defends its weakest mechanism, on the authority of its own clean paperwork. [Assumed Cause] is the term that moves the audit from the method to the premise.

Diagnostic Tests #

Every test here attacks the inputs. A test that asks whether the operator ran the causal read fails on contact, because he did. A test that asks whether the work was documented fails, because it was. The only tests that reach this failure are the ones that interrogate how the cause got named.

Test One — The Candidate Count Test. For any outcome the operator is explaining, ask how many candidate causes were on the list before one was selected. Not how confident he is in the answer. How many candidates. An operator who considered exactly one did not establish a cause, he assumed it, and the number is the whole finding. Two is not much better if the second was a straw candidate raised to be dismissed. Three or more independently plausible candidates, each sized against the effect, is the floor for a cause that has actually been established. Run this test on the operation’s five most-cited wins and count. The count will usually be one.

Test Two — The Timestamp Test. Ask when the cause was named: before the outcome was known, or after. A cause named in advance — written down as the anticipated mechanism at the moment of commitment, the way [Causal Read] discipline already demands — is a prediction that the outcome either confirmed or did not. A cause named after the result landed is a story constructed backward from a known ending, and the human brain builds those cleanly and fast. If the operator cannot produce a pre-outcome record of the mechanism he now credits, the attribution is retrospective by construction and gets treated as unestablished until it is tested forward.

Test Three — The Absent-Period Test. This is the cheapest disconfirmation available and it should be run first on every attribution in the operation. Take the named cause and go find periods where it was present and the outcome did not appear. The upsell script was in place in March and April; the check average moved in May, the month the prices changed. The training program ran for two years before turnover dropped and the drop tracks the lead’s hire date, not the program’s launch. If the named cause was present without the outcome, it is not sufficient on its own, and the operator’s model says it is. Five minutes in the numbers kills most assumed causes, which is exactly why nobody spends them.

Test Four — The Alternative-Sufficiency Test. Ask the operator to name what would have had to be true for a different cause to produce the same outcome. Then ask whether it was true. This is the test that forces the counterfactual into the open, and it is the one operators find hardest, because it requires arguing against a result they liked. For the strong quarter: what would have had to be true for the competitor’s closing to produce this by itself, and did the competitor close. For the low turnover: what would have had to be true for one strong lead to produce it, and was she there. An operator who cannot construct a single alternative account of his own best outcome has not established anything.

Test Five — The Co-Occurrence Sweep. For the window the outcome landed in, list everything that changed and was not the operator’s deliberate act. Weather. Calendar shifts and holiday placement. A competitor opening, closing, remodeling, or changing hours. A price change, yours or a vendor’s. A staffing change. A road closure, a construction start, an office tenant going hybrid, a school schedule, a local event, a new stop on a transit line. The sweep is mechanical and it takes twenty minutes. Its output is the candidate list Test One says the operator did not have. Run it as standing practice on every material variance, favorable ones included — favorable variances almost never get a sweep, which is why favorable variances are where the assumed causes accumulate.

Test Six — The Magnitude Test. Size the named cause against the effect it is being credited with. An item running at three percent of mix cannot produce a fourteen percent quarter no matter how well it sells. A script that touches one in six tables cannot move the check average four dollars. Do the arithmetic out loud. When the named cause is too small to have produced the measured result, the operator does not have an attribution problem at the margin — he has a missing cause somewhere, unnamed and unmanaged, doing the real work.

Test Seven — The Removal Test. Ask what specifically breaks if the named cause is removed tomorrow, and on what timeline. Then, where the exposure allows it, actually remove it and watch. If the operator cannot state a mechanism and a timeline, he is not describing a cause, he is describing a co-occurrence he grew fond of. This is the most expensive test on the list and the most conclusive, which is why it belongs on the items with the largest funding attached to them and nowhere else.

Test Eight — The Disconfirmation Test. Ask what evidence would prove the named cause wrong, and then ask whether the operator went and looked for it. The answer to the second question is almost always no, because the read closed the moment the first plausible cause appeared. This is [Bias Prosecution] pointed at the attribution rather than at the verdict: the operator takes his own named cause and runs it against his declared priors, naming where the evidence contradicts it. The contradiction is the signal. An attribution nobody tried to kill is an assumption wearing a conclusion’s clothes.

Test Nine — The Blind Second Read. Hand a second person the outcome and the raw window — the numbers, the calendar, the staffing, the sweep — and withhold your answer. Ask them to name the cause cold. A lead, a kitchen manager, a peer operator, anyone with the context and no stake in the attribution. When their answer differs from yours, you have two candidates and a real read to run. When their answer matches, you have the first independent support the attribution has ever had. Most operations have never run this once, because the operator’s attribution is stated first in every meeting it appears in, which contaminates every read in the room.

Test Ten — The Confidence Inversion. Rank the operation’s standing attributions by how hard the operator would defend them, and audit from the top. The instinct is to audit the shaky ones. The physics runs the other way: the shaky attributions are held loosely and get corrected by contact with reality, while the certain ones were named early, worked once, and have been accumulating funding and teaching authority ever since without a single contest. Confidence is the marker of the oldest unestablished cause in the building.

Test Eleven — The Audit-Trail Test. Take a decision that passed review and ask the reviewer what he checked. If the answer is that the read was run, the variables were listed, and the stability test cleared, the review checked the method. Ask him what he checked about the inputs — how the cause was established, how many candidates were considered, what disconfirmation was attempted. When the review has no answer to that, the operation’s audit function cannot detect this failure at all, and every clean review it has produced is evidence of nothing.

Family Position #

Failure-mode child of [Causal Read]. Sits inside Perspective, and cross-Fundamental in application — the read that names the cause runs at every altitude of the operation, so the failure appears at every altitude.

The parentage is exact and load-bearing. [Causal Read] names the physics: current signal plus causal model produces an anticipated future state that governs the operator’s action, whether he sees himself running it or not. The causal model is assembled from the operator’s own action-outcome pairs. [Assumed Cause] names the corruption of a pair at intake — the outcome recorded accurately, the cause recorded wrong — and the compounding consequence, which is that a model built from mislabeled pairs produces confident anticipations that will not arrive. This is a different failure from the ones [Causal Read] already carries. [Causal Read] names the operator whose model stops updating and the operator who attributes every miss to the market. [Assumed Cause] names the operator whose model updates diligently on bad labels. He is the operator with the best process and a mechanism that does not exist.

Perspective application. This is the term’s home. The operator’s read of his own history is the asset he trusts most, and it is assembled almost entirely out of attributions he made in seconds and never contested. Forty years of reps do not produce forty years of established causes. They produce a very large set of action-outcome pairs, weighted toward the vivid and the recent, labeled by whichever candidate was closest to hand at the time. The correction at the Perspective layer is not humility and it is not more examination. It is candidate generation as a standing discipline, applied hardest to the conclusions the operator holds most firmly.

Product application. Every element of the GX that is held because it works is held on an attribution. The signature item, the arrival sequence, the pacing, the room’s sound level, the pour. Each one earned its place because something good happened while it was in place, and almost none of them have been established as the cause of anything. The operator who has never separated the elements that produce the GX from the elements that merely accompanied it is carrying a Product he cannot edit, because he does not know which parts are load-bearing.

People application. The most expensive shape of this failure runs through People, because People causes walk out the door. A program gets credited for what a person was producing, so the program gets funded and the person gets nothing. The framework already runs this read in the other direction: the miss attributed to a cast member’s character when the system produced it. [Assumed Cause] runs it as the win attributed to the system when a person produced it. Same attribution failure, opposite sign, and the second one is quieter because nobody files a complaint about a good result.

Performance application. On the stage the attributions are fast, unwritten, and immediately taught. A save gets credited to a recovery script when the room was slow enough to absorb it. A smooth Friday gets credited to the pre-shift when the actual cause was the lineup. The lead then teaches the credited mechanism at the next pre-shift as established practice, and the operation accumulates a body of technique that has never been tested against an alternative account. Performance is where assumed causes enter the culture, because on the stage the attribution is spoken out loud within minutes and repeated as coaching before anyone has looked at the window.

Profit application. Profit is where the assumed cause gets funded. A favorable variance arrives, gets a one-line cause in the review, and that line becomes next year’s plan. Food cost improved because we tightened portions, in the same period the protein contract renewed lower. Check average lifted because of the script, in the same month prices moved. The operator scales the credited mechanism, the real cause reverses without warning because nobody was watching it, and the plan misses for reasons that read as execution failure. Favorable variances are the least-audited numbers in the operation and the highest-yield place to run the sweep.

Fundamentals Coverage.

Perspective read. [Assumed Cause] originates in Perspective because Perspective is where the causal model is stored. Every conclusion the operator holds about his market, his Guests, his cast, and his own capability is a label attached to a remembered outcome, and the labels were applied by availability, not by evidence — the deliberate act he made is always the most available candidate, and the environmental cause that arrived unannounced is never a candidate at all. It expresses itself as fluency. The operator answers new questions fast, from a deep store of what worked, and mistakes the speed for judgment. Detection at this layer is a count, not a feeling: for any conclusion he holds, how many competing accounts did he generate before he settled, and can he name one he rejected and why. An operator who cannot name a rejected candidate has never run the discipline on that conclusion. The response is to treat the causal model as a set of claims rather than a set of facts, to date each claim, and to invert the audit order so the firmest claims get contested first. Reps accumulate into evidence and into a sharper capacity to read. They do not accumulate into established causes, and an operator who believes they do has forty years of unaudited attribution running his building.

Product read. In Product the term operates on the operating output — what the operation produces and holds. Origination is the retained element: the operator keeps what worked and cuts what did not, and both decisions run on attribution. The expression is a Product nobody can edit. Ask an operator which three elements of the GX are actually producing the Guest’s return and he will name three, and he will not be able to tell you how he knows, because the elements were never separated from each other or from the conditions they ran inside. Detection is the removal question run on paper first: for each retained element, what specifically breaks if it goes, on what timeline, through what mechanism. Elements that cannot answer are accompaniments, not causes. The response is that every retained element carries a named mechanism and a magnitude — this element produces this part of the GX, at this scale, for this reason — and elements that cannot carry one get tested rather than defended. This is what makes the Product editable, and an operation that cannot edit its Product cannot respond to anything.

People read. People is where the attribution error is most costly and most invisible, because People causes have legs. Origination is the operator’s need to explain a People outcome he did not design: turnover fell, the cast held together through a hard quarter, a station started running clean. Something produced it, and the operator credits whichever instrument he installed most recently — the training program, the comp change, the schedule shape, the new handbook. The expression is a funded instrument and an unretained person. The operator protects the program and loses the lead, then reads her exit as an individual decision rather than as the removal of the actual mechanism. Detection is a timeline overlay: put the People outcome and the tenure dates on the same line and look at what actually moved when. Also run the reverse — the framework already teaches separating what the person did from what the system produced on the miss side, and the same separation is owed on the win side, where nobody asks for it. The response is that every People win gets a named cause with a person or a system on it, explicitly, and if the cause is a person she gets told, developed, and paid for what she is producing. An operation that credits its systems for what its people produce will keep its systems and lose its people.

Performance read. Performance is execution on the stage, in minutes, and attribution there is verbal and immediate. Origination is the post-shift read: something went well or went badly and the lead names why before anyone has left the building. The expression is technique taught as established practice on the strength of one attributed night — the recovery that worked, the seating pattern that flowed, the expo call that held the kitchen together. The operation accumulates a body of coached mechanism with no evidence under it, and cast members who follow it correctly and get no result learn that the standards are decoration. Detection is to ask, at the point of coaching, what else was true that night: the cover count, the pace, the lineup, the weather, the reservation shape. If the coached mechanism cannot be separated from the conditions, it is a hypothesis and gets taught as one. The response is a two-tier vocabulary in the pre-shift — this is what we know produces the result, and this is what we are testing — and a standing rule that nothing enters the taught standard on a single attributed shift.

Profit read. Profit is the money architecture, and it is where an assumed cause converts into capital allocation. Origination is the variance review, which is built to explain a number and not to establish a cause. A line moved, someone offers the most recent deliberate change as the explanation, the explanation is recorded, and the meeting moves on. The expression is the plan: the credited mechanism gets scaled and funded next cycle, the real cause continues unmanaged, and when it reverses the miss reads as an execution problem in the thing that was funded. Detection is the co-occurrence sweep run on every material variance, favorable ones first, plus the magnitude check — a mechanism too small to produce the number is a tell that something unnamed is doing the work. The response is that no variance closes with a single-candidate cause, no attribution enters the plan without a magnitude and a disconfirmation attempt on the record, and every funded mechanism carries the indicator that would show it is not the one producing the result. You own admin. Owning it means knowing which of your numbers you have established a cause for and which ones you have merely explained.

Cross-References To Locked IP #

Parent:

• [Causal Read] — the read discipline [Assumed Cause] is the failure-mode child of; the physics runs correctly and the inputs are wrong

Related:

• [The Read] — the aggregate discipline every assumed cause enters through and contaminates downstream

• [The Operator’s Read] — the multi-altitude read that produces the outcomes the operator then attributes

• [The Reader’s Unread Bias] — the sibling structural failure at the reader layer; that one corrupts the signal coming in, this one mislabels the world the signal came from

• [The ReRead] — the loop step closest to a catch mechanism; catches design-versus-result gaps and misses this failure unless extended to ask whether the result arrived for the designed reason

• [Recalibration] — the step that acts on the re-read, and the step that scales an assumed cause when the re-read closes green

• [Bias Prosecution] — the discipline of running the operator’s own read against his declared priors; pointed at the attribution, it is the disconfirmation requirement this term demands

• [The Read Log] — the instrument that can carry the candidate list, the magnitude check, and the disconfirmation attempt as fields rather than leaving them unrecorded

• [Perception Audit] — runs on the operator before the look; establishes what he already believed about what causes what

• [Just the Facts] — the intake discipline that catalogues what is on the stage without the operator’s explanation attached to it

• [Constant Expiry] — the adjacent determination failure; that one names the determination that expired, this one names the determination that was wrong about its causes from the start

• [Outcomes Formula] — grades the thinking by the outcome; [Assumed Cause] names the address inside the thinking where the unexamined assumption sits

• [Operating Helix] — the recalibration architecture that compounds a corrected attribution and equally compounds an uncorrected one

• [The Operator’s Own Audit] — [Causal Read] discipline turned inward on cadence; the natural home for the confidence-inverted attribution audit

• [The Biased Read] — the standing condition that no operator reads his own building clean; names attribution bias as one contaminant among five

• [Judgment Distortion] — the anchoring and overconfidence pressures that select an assumed cause and then protect it

• [The Summers Principle] — the design-or-default ground; an operator cannot design against a cause he has not established

Opposing patterns:

• [Case Study Reduction] — the discipline never ran; a retrospective account of another operation taken as an executable path

• [Orphaned Act] — the decision that skipped the filter; [Assumed Cause] is what converts a lucky one into a repeated one

• [Egotistical Ignorance] — the posture that refuses the audit that would surface an assumed cause; aggravator, not mechanism

• [Default Perspective] — the closed examination loop one layer up, where the interpretive frame cannot surface itself

• [Hacksterism] — the shortcut posture that avoids the read entirely; [Assumed Cause] pays the cost of the read and still misses

• [Stale Thinking] — the unrefreshed operating conclusion that an unestablished attribution hardens into

• [Dogma Trap] — the attributed mechanism defended as principle once it has been taught long enough

• [Predicted LTV] — attribution run on Guest history and converted into hospitality allocation

• [Franchisor Arbitrage] — sells a system whose causes were established in operations the buyer does not run

• [Static Decline] — the structural consequence of funding credited mechanisms while the real causes go unmanaged

• [Million Dollar Mediocrity] — the camouflaged form, where a healthy revenue line keeps unestablished attributions from ever being contested

Why This Matters #

Every audit the industry runs is a method audit. Did you do the read. Is it documented. Are the variables listed. Did you test them. Was the decision reviewed. That is the whole shape of quality control in an independent operation, and it is blind by construction to the failure that matters most, because [Assumed Cause] passes every one of those checks. The paperwork is the alibi. An operation can build a genuine discipline, run it faithfully for a decade, review it on cadence, and compound an error the entire time — and the better the discipline, the more authority the error accumulates.

This term earned a name because the alternative diagnoses are all wrong in the same direction. When a funded mechanism stops producing, the industry reads execution failure, market shift, cast turnover, or loss of focus. Every one of those reads keeps the attribution intact and puts the failure somewhere else. The operator doubles down on the mechanism, coaches harder on the script, protects the program, re-launches the item. He is defending something that never worked. Nothing in his own record will tell him, because his record was written by the same attribution.

It matters more than sloppiness, and that ranking is deliberate. Sloppiness is self-limiting. The operator who skips the read knows he guessed, holds the decision loosely, watches it, and corrects fast when it moves. Diligence removes the hedge. The operator who ran the discipline has earned his confidence and will spend real capital defending it, and the operation will scale the wrong mechanism at exactly the pace the operator’s competence allows. That is the inversion this term names: at the attribution layer, rigor is an accelerant, not a brake. The only thing that makes rigor safe is contested inputs.

It also protects the framework’s own instruments from being turned into liabilities. [Causal Read] is mandatory. The variable-stability test is the only legitimate extraction from a past event. [The ReRead] and [Recalibration] are the loop that keeps the operation honest. Every one of those can be run cleanly on an assumed cause and every one of them will return a pass. Without [Assumed Cause] named, the framework hands the operator a rigorous apparatus with an uninspected intake, and rigorous apparatus with uninspected intake is how confident, well-documented operations walk into [Static Decline] with clean files.

Across the framework this is the term that separates naming a cause from establishing one, and that distinction is upstream of everything the operator does next. Design requires a cause. [The Summers Principle] says every outcome inside the operating domain traces to design or default, and an operator cannot design against a mechanism he has not established — he can only fund the one he named. Read the attribution wrong and every downstream discipline runs perfectly against the wrong object. That is why this sits under [Causal Read] rather than beside it: it is not a competing discipline, it is the quality control on the one input the discipline never checks.

Operating Consequence #

Separate naming from establishing in the vocabulary. The operator strikes the unqualified causal “because” from operating speech and replaces it with one of two forms. “We have established that X produces Y, here is the evidence.” Or “X is our leading candidate for Y and it is untested.” Nothing gets to sit between those two states. This is not softening. It is the only vocabulary in which the operation can tell the difference between what it knows and what it assumed, and every plan, coaching moment, and funding decision downstream inherits the distinction.

Generate candidates before testing any of them. No attribution is permitted with one candidate on the list. Before the stability test runs, the operator produces at least three independently plausible accounts of the outcome, including at least one that requires nothing he did. The co-occurrence sweep is how the list gets built and it runs mechanically: calendar, weather, competitive set, prices, vendors, staffing, local conditions. The sweep before the test, never after.

Require disconfirmation, not confirmation. For every named cause the operator states what evidence would prove it wrong, then goes and looks for that evidence before the decision lands. The absent-period check runs first because it is the cheapest — find a period where the cause was present and the outcome was not. This is [Bias Prosecution] aimed at the attribution: the contradiction is the signal, and an attribution nobody tried to kill does not count as established.

Name the mechanism before the outcome arrives. Every commitment of consequence gets a written anticipated cause at the moment of commitment, not an explanation at settlement. This is the one move that makes attribution falsifiable, because a pre-registered mechanism can be wrong. A cause named after the result is a story built backward and gets treated as unestablished by default, regardless of how good it sounds.

Audit inputs, not method. Review changes its question. Not “did you run the read” but “how did you establish the cause, how many candidates were on the list, what did you try to disconfirm, and does the cause have enough magnitude to produce this effect.” A review that only certifies method is a review that cannot detect this failure, and the operator stops treating its clean passes as evidence of anything.

Invert the audit order by confidence. The attribution queue is ranked by how hard the operator would defend each item, and the audit starts at the top. The shaky attributions correct themselves on contact with reality. The certain ones were named early, worked once, and have been accumulating funding and teaching authority without a contest ever since.

Sweep favorable variances hardest. Every material variance gets the co-occurrence sweep, and the favorable ones get it first, because nobody demands an investigation into good news. The favorable variance with a one-line explanation is where the operation’s assumed causes accumulate, and it is where they get promoted into the plan.

Attach the falsifying indicator to every funded mechanism. When the operation funds a credited cause, it also names the indicator that would show the mechanism is not producing the result, and it reads that indicator on a schedule. This is how the re-read gets extended past design-versus-result into result-versus-reason. Without the indicator, a funded assumed cause is invisible until the real cause reverses.

Refuse single-shift attribution in coaching. Nothing enters the taught standard on one attributed night. The pre-shift carries two tiers of language — what the operation has established and what it is testing — and leads are held to the distinction, because a cast that follows coached mechanism and sees no result learns that the standards are decoration.

Credit people explicitly when people are the cause. Every People win gets a named cause with a person or a system attached to it, out loud. When the cause is a person, she is told, developed, and paid for what she is producing. The operation that credits its programs for what its leads produce keeps the programs and loses the leads.

What Changes Tomorrow #

Pick the single best result the operation produced in the last year, and pick it by defensibility rather than by size. The one where you would not hesitate for a second if somebody asked what drove it. The quarter that came in strong. The turnover number that finally held. The check average that moved. Confidence is the selection criterion, because the attribution you would defend fastest is the one that was named earliest and has been contested least.

Write down the cause you would name, in one sentence, before you look at anything. Then close the notebook on it and do not add to it. That sentence is the artifact under test, and the point of writing it first is that you cannot revise it quietly once the sweep starts producing candidates.

Run the co-occurrence sweep on the window the result landed in. Everything that changed and was not your deliberate act: the calendar and where the holidays fell, the weather, every competitor who opened or closed or changed hours or remodeled, every price change on your side and every vendor change on theirs, every staffing change including one person’s hire date and one person’s departure, and every local condition — construction, an office tenant going hybrid, a school calendar, a transit change, an event. Twenty minutes, mechanical, no judgment applied. When the list is finished, count how many of those items could plausibly have produced the result on their own. That count is your candidate list, and it is the list you did not have when you named the cause.

Then run the absent-period check on your original sentence. Find a period where your named cause was fully in place and the result did not appear. If you find one, your cause is not sufficient on its own and your model says it is. Then run the magnitude check: do the arithmetic and confirm the named cause is even large enough to have produced the size of the effect you are crediting it with.

The indicator is the count of surviving candidates. If your original sentence is the only account that survives the sweep, the absent-period check, and the magnitude check, you have established a cause for the first time — and you now know something about your operation you previously only believed. Fund it harder, teach it, and attach the falsifying indicator so you find out if it ever stops being true. If two or more candidates survive, the attribution is unestablished and every decision built on it is a bet you did not know you were placing; the work is to separate them, which usually means a removal test or a controlled period rather than more thinking. If your original sentence does not survive at all, you have just found the mechanism the operation has been funding, teaching, and defending for a year — and, more valuable, you have found the real cause, which has been running unmanaged and unprotected the entire time. Go protect it.

Then do the next one, working down the list in order of how hard you would defend each item. The inventory will be longer than you expect, and the worst entries on it will be the ones with the cleanest paperwork. Running the discipline is not the same as running it on the right variables, and the only defense against the second failure is refusing to let a cause into the operation with one candidate behind it.