1. The Controlling Idea
Variance is a measurement problem before it is a people problem.
2. Why This Matters in the Room
Multiple stations and satellite bars, counted by one person, in a room where the well carries the volume.
And the module's most useful content is a habit: the location of a variance is frequently its diagnosis, per Module 5.
A variance concentrated in one category tells you far more than an overall number — and it usually points at the category with the least measurement rather than at anyone's conduct.
Which matters because the reflexive conclusion is theft, and starting there is both statistically wrong and corrosive to a bar that probably has a spout problem.
3. The Mechanism
Confirm the variance exists first
Before investigating anything.
Inconsistent counting between periods produces phantom variance. Different method, different person, different time of day, different treatment of partial bottles — any of those produces a number that is an artifact.
One method, one person, one time. And if the method or the counter changed, that is the first thing to eliminate, because there may be nothing to investigate.
Reading the location
Per Module 5, each category has a different amount of measurement on it.
Beer is system-dispensed and its loss shows as foam waste, per Module 31. Wine is poured from a bottle into a marked glass in front of a guest. Call and top shelf are usually jiggered because the product is expensive and the guest is watching. Well spirits are free-poured, at speed, in the largest volume, and unobserved.
So a variance isolated to the well points at the free pour — and that is a mechanical finding rather than a moral one.
The mechanical causes, in order of how cheap they are to test
Spout flow rate. Different spouts have different rates and a count-based pour is calibrated to a specific spout. A box of replacements changes every pour in the building overnight. Two minutes to check.
Pour calibration drift. Free-pour accuracy degrades with speed, with fatigue, and with never having been calibrated. Five pours per bartender into a graduated cylinder, framed as calibration.
New staff pouring to their own hand.
Volume — a busier bar pours faster and heavier.
The recording causes
Comps, staff drinks, remakes, and pours for regulars land disproportionately in the well.
If they are not being rung, they appear as variance — and that is a recording failure rather than a theft.
An audit is a week of ringing everything, including staff drinks, and seeing how much of the variance it absorbs.
Spillage and breakage
Real, measurable, and usually unrecorded.
And foam waste is the beer version — per Module 31, collect a shift's foam into a container once and look at the volume.
Theft, and where it belongs
On the list, last.
Not because it does not happen — because the four categories above are cheaper to test, more likely to be true, and starting with suspicion damages a bar that probably has a mechanical explanation.
And if the first four come back clean, the remaining variance is where it lives — investigated through inventory variance by category rather than through suspicion of individuals.
Which is a process rather than an accusation, and any action touching an employee belongs with a licensed attorney.
Counting method
Consistency matters more than precision.
A method that is slightly imprecise and applied identically every period produces a usable trend. A precise method applied inconsistently produces noise.
Which means the method should be written down — including how partial bottles are estimated, what gets counted, and in what order.
4. The Variables You Control
Set directly: counting method and consistency, spout type, pour method, what gets rung, whether variance is analyzed by category, the order of investigation.
Observed and responded to: the variance's location; whether the counting method changed.
5. The Numbers
Confirm the variance exists before investigating it.
One method, one person, one time.
Variance by category, not overall.
Check the spouts first. Two minutes.
Free-pour test: five pours per bartender.
Theft last.
Any matter touching an employee's conduct belongs with a licensed attorney.
6. The Sensory Standard
Not applicable. This module's discipline is measurement, and its failure mode is a number that is an artifact.
What to observe instead: whether the counting method is written down and whether it was followed.
What almost-right presents as
A variance that is real and misattributed. The number is correct and the explanation is wrong — usually theft assumed where a spout change or an unrung comp would explain it.
A counting method that is consistent in intent and inconsistent in practice. Two people estimating partial bottles differently produces a variance that is entirely an artifact.
What each failure presents as
Counting inconsistent: phantom variance, investigated at length, explaining nothing.
Variance analyzed only in aggregate: a number with no location and no diagnosis.
Spout change unnoticed: a sudden uniform shift in one category.
Comps unrung: variance that a pour test cannot explain.
Theft assumed first: a damaged bar and an unsolved problem.
7. The Worked Example
Four percent variance on well spirits and under one percent on everything else, consistent for three months.
The situation. Nobody has been accused of anything and morale is fine.
A variance isolated to one category with everything else clean is a highly informative result, and the isolation does most of the diagnostic work.
Ask what is structurally different about well spirits.
Beer is system-dispensed and its losses show as foam waste. Wine is poured from a bottle into a marked glass in front of a guest. Call and top shelf are usually jiggered, because the product is expensive and the guest is watching.
Well spirits are free-poured, at speed, in the largest volume, and nobody is watching.
So the category with the variance is the category with the least measurement. That is not a coincidence and it is the finding.
Now, four percent isolated there is consistent with several things and they are not equally likely.
Pour drift. The leading candidate. A half-ounce of over-pour on a high-volume well produces exactly this magnitude. It is not deliberate — free-pour accuracy degrades with speed, with fatigue, and with never having been calibrated against anything.
Spout variation. Different pour spouts have different flow rates, and a count-based free pour is calibrated to a specific spout. A box of replacements from a different supplier changes every count-based pour in the building silently. Check this first — two minutes.
Unrecorded drinks. Comps, staff drinks, remakes, and pours for regulars all land disproportionately in the well. If they are not being rung, they read as variance — and this is a recording failure rather than a theft, and it is extremely common.
Spillage and remakes. Also concentrated here.
Theft. It belongs on the list. It belongs last, and starting here is both statistically wrong and corrosive to a bar where morale is currently fine. The first four are cheaper to test and more likely to be true, and if they are eliminated the remaining variance can be investigated properly rather than by suspicion.
How to test, in order, and none of it requires accusing anyone.
Check the spouts. Compare current against what the bar used to use. Measure flow.
Run a free-pour test. Each bartender pours their standard count into a graduated cylinder, five times. Frame it as calibration, which is what it is. The spread within each person and the difference between people is the data. If everyone is heavy, it is systemic and the fix is retraining or jiggering. If one person is far off, it is individual and it is still probably not deliberate.
Audit the comp and remake recording for a week. Ask bartenders to ring everything, including staff drinks, and see how much of the four percent it absorbs.
Then recount, using one method, one person, one time of day, to make sure the variance is real.
And that last one deserves emphasis: before investigating a variance, confirm the variance exists. Inconsistent counting between periods produces phantom numbers, and three months of consistent four percent could as easily be three months of the same counting error.
What I rule out. Price or cost changes, which would affect cost rather than the physical variance between poured and sold. And a menu shift toward well drinks, which would increase well volume rather than well variance percentage.
8. Failure Taxonomy
Full treatment below. Counting method inconsistent. Variance analyzed only in aggregate. Spout change unnoticed. Comps unrung. Theft assumed before the mechanical causes were eliminated.
The named failures, in full
Counted inconsistently between periods Signature. Variance numbers that swing wildly with no operational change. Cause. Different counting methods, different people, different times of day. The variance is measuring the counting, not the inventory. Decision. Correctable. Recovery. One method, one time, documented. Verification. Count twice on the same night with two people and compare.
Variance calculated and never investigated Signature. A variance report produced monthly and filed. Cause. Measurement without follow-through. The number is only useful if it points somewhere. Decision. Systems. Recovery. Investigate by category. Variance concentrated in one place has a specific cause — well spirits point to free-pouring, beer points to foam waste, wine points to over-pouring or breakage. Verification. Four percent on well spirits and under one on everything else is a pour-control finding and nothing else.
Received without verification Signature. Paying for product that did not arrive. Cause. Signing rather than counting. Decision. Correctable. Recovery. Count against the invoice at delivery, every time. Verification. Spot-check a delivery against the invoice.
Over-pour absorbed as a cost of business Signature. A known pour problem that everyone has decided to live with. Cause. The cost is diffuse and the fix feels confrontational. Decision. Systems. Recovery. Put a dollar figure on it. Retrain against a measured baseline. Verification. The number usually ends the debate.
9. Texas Room Application
Multiple stations and satellite bars, counted by one person, in a room where the well carries the volume.
The named failure: variance concentrated in well spirits. The location is the diagnosis.
Recovery. Check the spouts, run a free-pour test framed as calibration, audit the comp and remake recording. Theft last, because the mechanical explanations are cheaper to test and more likely to be true.
And before investigating a variance, confirm the variance exists.
Full Texas Room Application
The Texas context. Multiple stations and satellite bars, counted by one person, in a room where the well carries the volume.
The named failure: variance concentrated in well spirits. The location is the diagnosis — the well is the free-poured category and the only one with no measurement on it.
Recovery. Check the spouts, run a free-pour test framed as calibration, audit the comp and remake recording. Theft belongs last, because the mechanical explanations are cheaper to test and more likely to be true.
And before investigating a variance, confirm the variance exists. Inconsistent counting between periods produces phantom numbers.
10. Volume Pressure
Volume generates the variance — free-pour accuracy degrades under speed, and the twelve minutes that produce most of the night's volume are the twelve minutes with the least accuracy.
What can flex: jigger versus count, decided in advance.
What cannot: the arithmetic.
11. The Diagnostic
Full scenario below. Four percent variance isolated to the well. The reasoning reads the concentration as the diagnosis, orders the tests by cost, places theft last with an explicit justification, and ends by questioning whether the variance is real.
The scenario, in full
The scenario. Your variance report shows four percent on well spirits and under one percent on everything else — beer, wine, call, and top shelf are all within normal range. This has been consistent for three months. Nobody has been accused of anything and morale is fine.
The variance has a precise location. Read it.
The reasoning.
A variance isolated to one category with everything else clean is a highly informative result, and the isolation does most of the work.
Ask what is structurally different about well spirits compared to every other category.
Beer is dispensed by a system and its losses show up as foam waste, which is visible in a drain. Wine is poured from a bottle into a marked glass, in front of a guest. Call and top shelf are usually jiggered, because the product is expensive and the guest is often watching the pour. Well spirits are free-poured, at speed, in the largest volume, and nobody is watching.
So the category with the variance is the category with the least measurement. That is not a coincidence and it is the finding.
Now, four percent isolated there is consistent with several things and they are not equally likely.
Pour drift. The leading candidate. A half-ounce of over-pour on a high-volume well produces exactly this magnitude. It is not deliberate; free-pour accuracy degrades with speed, with fatigue, and with never having been calibrated against anything.
Spout variation. Different pour spouts have different flow rates, and a count-based free pour is calibrated to a specific spout. A box of replacements from a different supplier changes every pour in the building silently. Check this first — two minutes.
Unrecorded drinks. Comps, staff drinks, remakes, and pours for regulars all land disproportionately in the well. If they are not being rung as comps, they read as variance. This is not theft; it is a recording failure, and it is extremely common.
Spillage and remakes. Also concentrated here.
Theft. It belongs on the list. It belongs last, and starting here is both statistically wrong and corrosive to a bar where morale is currently fine. The first four explanations are cheaper to test and more likely to be true, and if they are eliminated the remaining variance can be investigated properly rather than by suspicion.
How to test, in order, and none of it requires accusing anyone.
Check the spouts. Compare current spouts against what the bar used to use. Measure flow.
Run a free-pour test. Each bartender pours their standard count into a graduated cylinder, five times. Frame it as calibration, which is what it is. The spread within each person and the difference between people is the data. If everyone is heavy, it is systemic and the fix is retraining or jiggering. If one person is far off, it is individual and it is still probably not deliberate.
Audit the comp and remake recording for a week. Ask bartenders to ring everything, including staff drinks, and see how much of the four percent it absorbs.
Then recount, using one method, one person, one time of day, to make sure the variance is real and not an artifact of inconsistent counting.
That last one deserves emphasis. Before investigating a variance, confirm the variance exists. Inconsistent counting between periods produces phantom numbers, and three months of consistent four percent could as easily be three months of the same counting error.
What to rule out. Price or cost changes — would affect cost, not the physical variance between poured and sold. A menu shift toward well drinks — would increase well volume, not well variance percentage.
12. The Practice Protocol
Exercise one: write down the counting method, including partial-bottle estimation.
Exercise two: count twice on the same night with two people and compare.
Exercise three: analyze variance by category for one period.
Exercise four: check the spouts. Two minutes.
Exercise five: the free-pour test, framed as calibration.
What to expect. Exercise two frequently reveals that the variance was partly an artifact, which changes the entire investigation.
What this cannot teach. An accurate free pour. Calibration by repetition against a measure.
13. Where This Connects
Module 5 supplies measurement and the variance-location reasoning. Module 31 supplies foam waste. Module 54 supplies pour cost. Module 57 supplies the analytics. Module 51 supplies the remake tracking that explains part of the well variance.
Into the mastery schools: Program Leadership.
14. What This Does Not Qualify You To Do
Independent education, not accreditation or licensure. Any investigation touching an employee's conduct, and any action arising from it, belongs with a licensed attorney. Nothing here is employment advice.