skip to content

A staffing tool reports only that 6 of 20 items went unassigned, so what certificate should it surface instead and what decision does that unlock?

level: principalimportance: nice to knowfreq 24%

answer

  1. a count is not a finding
  2. publish the constrained set
  3. shortfall equals items minus eligible engineers
  4. capacity outside the set changes nothing
  5. witness of infeasibility, not a number

basics

~20 s

Surface the constrained set: the group of items whose combined eligibility names too few engineers, with its exact shortfall. That set is a proof rather than an outcome, and it says precisely where cross-training or hiring raises the ceiling.

solid answer

~50 s

A bare count is an outcome, not a finding: it cannot be acted on and it invites someone to blame the scheduler. The actionable artefact is the **deficient set** - a set `S` of items whose combined eligibility reaches only `|N(S)|` engineers, with shortfall `|S| - |N(S)|`. If nine items are eligible only for three engineers, the shortfall is six, the ceiling is `20 - 6 = 14`, and the six unassigned items are proven unavoidable rather than a scheduling failure. Equivalently, a minimum vertex cover of the eligibility graph names the same bottleneck as a small set of items and engineers on which everything concentrates. The decision it unlocks is *where* to invest: broadening eligibility into that set's items, or adding an engineer eligible for them, is the only change that moves the ceiling.

go deeper

for a junior

Recall that unassigned work can be impossible rather than merely unlucky, and that the useful question is which items were blocked and by what.

for a middle

Compute the ceiling from a constrained set: items minus the shortfall between the set's size and the number of engineers eligible for it, and explain why re-running changes nothing.

for a senior

Design the report so it distinguishes a structural shortfall from a suboptimal run, and name the exact constrained set rather than handing a reviewer a bare number.

for a principal

Own the trade-off. Certifying optimality and publishing a bottleneck set costs work and exposes an organisational judgment about scarce skills, so decide who reads it, which witness is binding, and whether the stronger claim is worth its price.

## Why the count is the wrong output "6 of 20 items unassigned" answers none of the three questions a lead will immediately ask: is this the best achievable, which items are stuck, and what would fix it. Worse, it is ambiguous between two very different states of the world: - the assignment is **optimal** and the shortfall is structural - the eligibility graph simply cannot do better; - the assignment is **suboptimal** and a rearrangement would have staffed more. The distinction between those is the difference between a hiring decision and a bug report. A number cannot carry it; a certificate can. ## The certificate and the arithmetic it implies The structural obstruction has a precise shape. For a set `S` of items, let `N(S)` be the engineers eligible for at least one item in `S`, and call `|S| - |N(S)|` the **deficiency** of `S`. The maximum number of items that can ever be staffed is the item count minus the largest deficiency over all sets. A worked instance matching the report above: of 20 items, 9 are eligible only for the same 3 engineers. 1. That set has deficiency `9 - 3 = 6`. 2. So at most `20 - 6 = 14` items can be staffed, and at least 6 must go unassigned. 3. The observed 6 unassigned therefore matches the ceiling exactly - the run was optimal, and no rescheduling will help. Two arithmetic slips are worth naming because they surface in real reviews: `20 - 3 = 17` subtracts the engineer count rather than the shortfall, and `20 - 9 = 11` writes off the whole constrained set although three of its nine items do get staffed. | Output | What it proves | What a lead can do with it | |---|---|---| | "6 unassigned" | Nothing beyond this run | Ask whether the tool is broken | | "6 unassigned, and here is a larger assignment" | The run was suboptimal | File a defect against the assignment logic | | "These 9 items are eligible for only these 3 engineers" | A shortfall of 6 is unavoidable | Target cross-training or hiring at that set | | A minimum vertex cover of the same size as the matching | The assignment is maximum | Close the question of optimality entirely | ## Which investment actually moves the ceiling This is where the mathematics decides the design rather than merely describing it. Given the deficient set `S`, the interventions divide cleanly: - **Adding an engineer eligible for items in `S`** reduces that set's deficiency by one, so the ceiling rises by at most one. Each additional such engineer buys at most one more staffed item. - **Adding an engineer eligible only for items outside `S`** changes the deficiency of `S` not at all. The ceiling does not move. This is the expensive mistake the bare count encourages, because headcount looks like the obvious lever. - **Broadening the eligibility of items inside `S`** - cross-training, relaxing a requirement, splitting a specialist skill - enlarges `N(S)` for the same cost profile, and is usually cheaper than hiring. - **Reducing `S`** by deferring or removing constrained items lowers `|S|` directly, which is the lever available inside the current sprint. Note also what does **not** help: re-running with a different ordering, retrying, or raising a priority. The ceiling is a property of the eligibility graph, and it is indifferent to effort. ## Reporting choices a lead owns Surfacing the certificate is not free, and the trade-offs are genuine: 1. **Which witness to publish.** Several deficient sets may exist. A set of maximum deficiency bounds the whole plan and is the binding constraint; a smaller violator bounds only part of it and can mislead a reader into a partial fix. 2. **How much to expose.** A bottleneck set is, in a staffing context, a statement about specific people and specific scarce skills. A tool that publishes it widely is publishing an organisational judgment, which may need to be scoped to the people who own the hiring decision. 3. **Whether to certify optimality at all.** Proving "this is the maximum" costs more than producing an adequate assignment. The question is whether the audience needs the stronger claim - a plan a team commits to usually does, a background sweep that reruns hourly usually does not. 4. **Stability of the report.** A certificate is reproducible in a way that a merely adequate assignment is not; two runs that report different counts with no explanation erode trust in the tool faster than a lower but explained number. The underlying discipline generalises past staffing: when a system reports that it could not satisfy everything, the useful output is the **witness to infeasibility**, not the shortfall count. The witness is what turns an outcome into a decision.

  • Does adding one more engineer always reduce the shortfall by one?
    Only if that engineer is eligible for items inside the constrained set - then its neighbourhood grows by one and the deficiency drops by one. An engineer eligible only for items outside the set leaves the deficiency untouched and the ceiling unchanged, which is why headcount added without regard to eligibility can buy nothing at all.
  • If several constrained sets exist, which one should the report name?
    One of maximum deficiency, because that is the binding constraint on the whole plan; a smaller violator bounds only part of it and can suggest a fix that leaves the ceiling where it was. If two sets tie, either is a valid witness, and reporting both is fine when they point at different skills.
  • How would you show the run was optimal rather than merely adequate?
    Publish a set of vertices touching every eligibility edge whose size equals the number of staffed items. Anyone can verify the touching claim edge by edge, and on a bipartite graph that equality certifies the assignment is maximum, so no rearrangement can staff more.

saying these in an interview costs you the question

  • Reports an unassigned count with no witnessing set
  • Proposes hiring without checking eligibility for the blocked items
  • Treats a structural shortfall as a scheduler defect to retry
  • Assumes a different processing order would clear the backlog
  • Reads the shortfall as an estimate rather than a proven ceiling