1Introduction
When a cake is left on the counter of a shared kitchen with a note inviting people to help themselves, nobody divides it. People arrive one at a time, look at what is left, and cut themselves a piece. Most take less than they would like. The cake shrinks quickly at first and then slowly, and in our experience, and in the data reported here, a narrow piece is very often still on the plate days later.
An audit of twelve offices found that cakes and similar shared items made up the largest share of food left unfinished in workplace kitchens (Dunleavy and Okoro, 2023), and a field study of group snacking found that people avoid the last item when others might still want it (Haugland and Sefton-Rye, 2021). This paper models the process and fits the model to data. The cutting phase is a stick-breaking process: each person takes a random fraction of whatever remains. Stick-breaking is well studied as a construction of random discrete distributions (Kingman, 1975; Sethuraman, 1994; Pitman, 2006), and its use as a model of literal division seems natural. We add a cutting threshold , the mass below which nobody will cut a piece, and a second phase in which the residual sits on the plate until someone takes all of it or it is cleared away.
The fair-division literature that began with Steinhaus (1948) also studies cakes, but it asks how to allocate a divisible good among agents with known preferences so that the allocation is fair (Brams and Taylor, 1996). Our question is descriptive. Nobody in the kitchen is trying to be fair, and the allocation emerges from sequential, uncoordinated choices.
The main results are exact and short. Under independent fractions, the logarithm of the remaining mass is a Poisson process in disguise (Lemma 1). The number of servings is then (Theorem 2), and the residual mass has a Beta law that does not depend on how many servings preceded it (Theorem 4). For the last piece, a visitor rate that decays like gives a Lomax tail, and the mean lifetime is finite only when a single dimensionless number, , exceeds one (Theorem 6).
We fit the model to 62 cakes in three kitchens. The cutting phase fits well. The estimated tail index for the last piece is , with an upper confidence limit of 0.88. Under the model the last piece in these kitchens has no finite expected lifetime, and in practice it is removed by the weekly clearance.
2Model
Let the initial mass of the item be 1. Visitors who take some of it are indexed in order of arrival, and is the mass remaining after visitor , with .
Definition 1 (Threshold stick-breaking). Fix and a sequence of random variables in . Visitor cuts if and only if , in which case they take and leave . The cutting phase ends at
the number of servings cut, and is the last piece.
Assumption 1. The fractions are independent and identically distributed with law for some .
Under Assumption 1 the mean fraction taken is , so measures restraint: larger means smaller pieces. The law is the one used in the stick-breaking construction of the Dirichlet process (Sethuraman, 1994), and we chose it for that reason before seeing data. Section 7 tests it.
Once the last piece exists, it stays on the plate until somebody takes all of it. We measure time from the moment the last piece is formed, in office hours.
Assumption 2. Visitors to the last piece arrive as a Poisson process with intensity and cumulative intensity . Each visitor, independently of everything else, takes the last piece with probability . The lifetime of the last piece is the time of the first visitor who takes it.
The particular intensity we fit is
with initial rate and a time scale over which interest falls. The decay is an empirical choice. It fits the visit counts in our data better than an exponential decay (Section 7), and its consequences are the subject of Section 4.
3Serving counts and the residual
Lemma 1 (Log-mass is a Poisson process). Under Assumption 1, let for the untruncated sequence (). Then are the points of a homogeneous Poisson process of rate on .
Proof. If then , whose distribution function is on . Hence satisfies , so . Since with independent , the are the arrival times of a rate- Poisson process.∎
Theorem 2 (Serving count). Under Assumption 1 and for any ,
Proof. Write . The first visitor always cuts, because . For , visitor cuts if and only if , that is, . The cut sequence agrees with the untruncated one up to the first with , so . By Lemma 1 the count is the number of points of a rate- Poisson process in , which is .∎
Corollary 3. and . As , almost surely, and converges in distribution to a standard normal.
Proof. The moments are those of Eq. (3). The almost-sure limit is the strong law for the Poisson counting process, and the central limit statement follows from the normal approximation to the Poisson law (Durrett, 2019).∎
The serving count grows only logarithmically in . Halving the cutting threshold, for example by providing a sharper knife, adds servings in expectation, which is 1.8 at our estimate of .
Theorem 4 (The last piece). Under Assumption 1, , and is independent of . In particular
Proof. In the notation of the previous proof, is the first point of the Poisson process at or beyond . By the memoryless property, and is independent of the configuration of points in , hence of (Kingman, 1993). Then , and for .∎
Theorem 4 says that the size of the last piece carries no information about how many people came before it. A small last piece is not evidence of a busy kitchen.
For completeness we record the sizes of the servings themselves.
Proposition 5. Under Assumption 1, for ,
where denotes the untruncated pieces. The untruncated sequence has the law, its decreasing rearrangement is Poisson–Dirichlet with parameter , and .
Proof. The inequality holds because pointwise. The expectation follows from independence: with . The identification with and Poisson–Dirichlet is the definition of those laws (Pitman, 2006; Kingman, 1975). For the last claim, . With and this equals .∎
The quantity is the probability that two crumbs chosen at random from the original item were eaten as part of the same serving. It also appears as the homozygosity in Ewens’s sampling formula (Ewens, 1972), where plays the role of a mutation rate.
4The last piece
Theorem 6 (Lifetime of the last piece). Under Assumption 2, . With the intensity of Eq. (2),
so that has a Lomax law (Lomax, 1954). Consequently if , and if . The median lifetime is for every .
Proof. Visitors who take the last piece form an independent thinning of the visit process with retention probability , which is again a Poisson process, with intensity (Kingman, 1993). is its first point, so is the probability of no points in , namely . For Eq. (2), , which gives Eq. (6). Then , which equals for and diverges otherwise. The median solves .∎
Corollary 7. The expected lifetime of the last piece is finite if and only if
The exact form of Eq. (2) is not needed for the divergence. All that matters is that interest in the last piece decays no faster than .
Proposition 8. Suppose for all and some constants . If then .
Proof. By Theorem 6, for , and when . Since , .∎
In practice the lifetime is never infinite, because somebody clears the plate. In all three of our kitchens, facilities staff removed any remaining food at 17:00 on Fridays. We treat this as right censoring. It truncates the observed lifetimes but not the underlying law, and it is the reason the divergence in Theorem 6 does not show up as cakes accumulating.
5Estimation
Proposition 9. Let be observed fractions under Assumption 1 and . The maximum likelihood estimator is , and
is an exact 95% confidence interval for .
Proof. By Lemma 1, is a sum of independent variables, so . The log-likelihood is up to a constant, maximised at . Inverting the pivot gives the interval.∎
For the threshold we use the largest observed last piece, over items. By Theorem 4 the maximum of independent variables has distribution function on , so its mean is . We correct for this bias by multiplying the maximum by .
For the last piece, let be observed lifetimes or censoring times and the number of lifetimes that ended with someone taking the piece. With fixed, the censored Lomax log-likelihood plus terms free of is maximised at
We report intervals from the approximation to (Lawless, 2003). The approximation is exact under type II censoring and was close to nominal under our censoring pattern in simulation (Appendix A). We estimate and separately from visit counts, which also gives a plug-in estimate of to compare with Eq. (9).
6Data
We recorded 62 cakes in three kitchens of a 340-person engineering consultancy between January and November 2025. Two kitchens served one floor each and the third served a ground-floor meeting suite. A cake was included if it was placed on the kitchen counter with an open invitation (a note, an email or a message on the office chat) and was not already cut. Traybakes and items sold in pre-portioned form were excluded, as were six cakes that were moved to a meeting room before the last piece formed.
A ceiling camera above each counter took one image per minute. The camera’s field of view covered the counter only; images were processed on a local machine, and only plate-level measurements were kept. Remaining mass was estimated from the visible area of the item multiplied by its initial height, measured once by hand. On nine cakes that we also weighed after each serving, the image-based estimate had a correlation of 0.97 with the true remaining mass and a mean absolute error of 1.8% of the initial mass. A serving was recorded when the estimated mass fell by more than 1% between consecutive images and the change was confirmed by a person reviewing the image pair. This gave 622 servings.
Visits to the last piece were counted with a separate ceiling occupancy sensor, which records the presence of a person within half a metre of the counter but no image. A visit was a presence of at least five seconds while a last piece was on the counter. We recorded 655 such visits. Time was measured in office hours (09:00 to 17:00, Monday to Friday), so that a last piece formed at 16:00 on a Tuesday and taken at 10:00 on Wednesday has a lifetime of two hours.
The study was approved by the Ellery College research ethics committee. All staff were told about the cameras in advance and could ask for any image to be deleted before processing. No such requests were made.
7Results
7.1Cutting phase
The 622 observed fractions give with exact 95% interval 2.40 to 2.81 (Proposition 9), so the mean serving takes 28% of what remains. A Kolmogorov–Smirnov test of against the unit exponential gives , . Measurement error matters most for small remaining masses, where a fixed error in area is a large error in the fraction. Restricting the estimate to cuts made with more than 20% of the cake remaining gave .
The largest last piece was 3.09% of its cake, which after the bias correction gives . Table 1 lists all estimates.
| Parameter | Meaning | Estimate | 95% CI | Based on |
|---|---|---|---|---|
| Restraint; mean fraction | 2.60 | [2.40, 2.81] | 622 servings | |
| Cutting threshold (fraction of initial mass) | 0.031 | [0.031, 0.032] | 62 last pieces | |
| Initial visit rate to the last piece (per hour) | 4.0 | [3.5, 4.6] | 655 visits | |
| Decay time of the visit rate (hours) | 3.0 | [2.2, 4.1] | 655 visits | |
| Probability a visitor takes the last piece | 0.055 | [0.039, 0.075] | 36 of 655 visits | |
| Tail index, direct, Eq. (9) | 0.65 | [0.46, 0.88] | 36 taken, 26 censored | |
| Tail index, plug-in | 0.66 | [0.42, 1.02] | as above |
Figure 1 compares the distribution of servings per cake with Theorem 2. The predicted mean is and the observed mean is 10.03. The fit to the shape is good: pooling bins with expected counts below five gives on 6 degrees of freedom (). The observed variance, 7.84, is lower than the predicted 9.01, a dispersion index of 0.87. Under-dispersion would follow if a large serving made the next visitor slightly more restrained, which would violate the independence in Assumption 1. With 62 cakes we cannot distinguish this from chance.
The last pieces averaged , against from Eq. (4), and a Kolmogorov–Smirnov test against gave . The correlation between the size of the last piece and the number of servings was (), consistent with the independence in Theorem 4. Table 2 collects these checks.
| Quantity | Model | Observed | Test |
|---|---|---|---|
| Mean servings per cake | 10.01 | 10.03 | |
| Variance of servings | 9.01 | 7.84 | dispersion index 0.87 |
| Serving-count distribution | Figure 1 | , 6 df, | |
| Log-fractions | KS , | ||
| Mean last piece, in units of | 0.722 | 0.718 | KS |
| Correlation of last piece with servings | 0 | ||
| Median lifetime of the last piece (hours) | 5.7 | 5.4 (Kaplan–Meier) |
7.2Last-piece phase
Visit counts during the last-piece phase fell off as Eq. (2) predicts. A Poisson regression of hourly visit counts on gave visits per hour and hours, and fitted better than an exponential decay (difference in AIC 11.4). Of 655 visits, 36 ended with the visitor taking the last piece, so . The remaining 26 last pieces were still on the plate at the Friday clearance.
With fixed at its estimate, Eq. (9) gives (0.46 to 0.88). The plug-in estimate agrees, with a wider interval because it combines three estimates. Both point estimates are below one, and the direct interval excludes one. By Corollary 7, the kitchens are in the regime with infinite expected lifetime, and the critical acceptance probability is .
Figure 2 shows the Kaplan–Meier estimate (Kaplan and Meier, 1958). The curve flattens after about 24 office hours because few last pieces remain at risk by then; most of those are censored on Friday. The Kaplan–Meier median, 5.4 hours, is close to the Lomax median of 5.7 hours. Means cannot be compared because the model has none.
7.3An unplanned comparison
During the study one member of staff began attaching a handwritten note reading “please finish me” to the last piece of cakes on one floor. This happened for 8 cakes. Of 64 visits to those last pieces, 7 ended with the piece being taken, so . With the other parameters at their pooled estimates this gives , above the critical value, and an expected lifetime of office hours by Theorem 6. Eight cakes on one floor with no randomisation are not evidence of an effect, and we report this only because it bears directly on Corollary 7.
8Discussion
The cutting phase is described well by a model with two parameters, and the exact results in Section 3 make it easy to test. We were surprised that independent Beta fractions fit as well as they did. Visitors differ in appetite and in how much they think is acceptable to take, and a mixture over would be a more realistic model. A mixture of laws is not a law, but over the range of that seems plausible the difference in the distribution of is small enough that 622 servings may not detect it.
The last-piece phase is the more consequential part. In a shared kitchen, a resource that nobody owns and nobody clears is a small instance of the problem described by Hardin (1968), although with the sign reversed: the difficulty is not overuse but the reluctance of any one person to finish what is there. Ostrom (1990) argued that commons are often managed well by local rules rather than by central authority. The Friday clearance is a central rule, and the note in Section 7 is a local one. Corollary 7 says that a local rule only needs to raise above . That target depends on how quickly interest fades, which a rule cannot easily change.
Heavy tails estimated from 36 events deserve suspicion, and Clauset et al. (2009) give reasons to be careful about inferring tail behaviour from small samples. Our conclusion does not rest on extrapolating the tail. The fitted model has a finite median, which matches the data, and an infinite mean, which the data cannot show directly because of censoring. What the data do show is that the survival curve lies above the critical curve over the whole observed range.
There are other limitations. All three kitchens belong to one organisation. Lifetimes are measured in office hours, which ignores the occasional evening visitor. The assumption that the acceptance probability does not change over time is probably wrong, since a last piece that has been on the plate for two days is less appealing than a fresh one. That would lower for old pieces and make the tail heavier still.
Data availability
Serving-level measurements, visit counts and code to reproduce every number in this paper are available from the first author. Images were not retained.
ACoverage of the interval for the tail index
We simulated 10,000 data sets with the estimated parameters, 62 items each, and the observed pattern of formation times within the week, with censoring at 17:00 on Fridays. The nominal 95% interval for from Eq. (9) covered the true value in 94.1% of replicates. The interval for from Eq. (8) is exact and covered in 95.0%, as it should. In the same simulations, the probability that the direct interval for excluded one, when the true value was 0.65, was 0.71.
BDerivation of the bias of the threshold estimate
Let . By Theorem 4 and independence across items, for . Then . With and the correction factor is 1.006. An exact one-sided interval follows from the same distribution function: with 97.5% confidence, which gives the upper limit in Table 1.
Acknowledgments
We thank the facilities team of the participating firm for keeping to a fixed clearance time throughout the study, and Ruth Okafor-Grange for reviewing 1,140 image pairs. T.W. was supported by a Kingsbury University early-career fellowship.