Minimum demo set: cover every required feature

A minimum demo set is the smallest collection of supplied demonstrations that covers every required feature. The task is not to choose the individually most impressive demonstration. Complete this exercise to save a coverage ledger, every equally small complete collection and a reason no smaller collection works.

Content checked: 18 October 2026.

What does a minimum demo set decide?

Imagine a small equipment seller preparing an internal product briefing. The required features and the contents of four demonstrations have already been supplied. Our identifiers are invented teaching inputs, not verified product capabilities, observed sales data or a record of Anar conducting a customer demonstration. No demonstration is performed here.

A feature shown twice cannot replace a different feature that is missing. That makes this a set-union decision: merge the feature identifiers and count each distinct identifier once. Four feature mentions might represent only two distinct requirements. Read the identifiers rather than trusting a total or a confident recommendation.

Every demonstration has the same selection cost of one. Each can be chosen once, all are independently available, and there are no timing, money, sequence or compatibility constraints. Those are explicit exercise assumptions, not facts about a real showroom. A real brief may need additional conditions before anyone promises a demonstration.

The artificial intelligence guide covers broader starting points. An assistant is optional. The supplied lists and a text editor suffice; no account, upload, installed solver or customer record is needed.

Preserve four supplied demonstrations before choosing

Write the contract first: “Practice V1. Required features R1, R2, R3 and R4. Candidate demonstrations J, K, L and M. Select each candidate at most once. Cover every required feature. Minimise the number of selected demonstrations.” Do not add scores or quietly remove an inconvenient requirement.

J shows R1 and R2. K shows R3 and R4. L shows R1 and R3. M shows R2 and R4. Each list is complete for this fictional exercise. An omitted feature therefore means not included under this contract; it does not mean the source forgot to answer. Unknown membership is a separate case later.

Keep that source card above a separate decision table. Editing the source to match a preferred answer would change the question. If a real owner supplies a corrected membership list, preserve the change as another version and recalculate instead of mixing old rows with new conclusions.

The product variant matrix distinguishes an absent option combination from zero inventory and unknown facts. Our question begins after the demonstration memberships are fully supplied. It does not establish inventory, variant existence or whether a salesperson can actually perform a demonstration.

List every subset and inspect the feature union

With four candidates, there are sixteen possible selections: each candidate is either included or excluded, giving two times two times two times two. The empty selection belongs in that list. Selecting J then K is not a new collection compared with K then J; order is outside the model.

Organise the ledger by size. There is one empty selection, four single candidates, six pairs, four triples and one four-candidate selection. The control total is one plus four plus six plus four plus one, which equals sixteen. This prevents a duplicated pair from hiding a missing alternative.

For each row, merge the selected membership lists and remove repeated feature identifiers. Compare that union with the complete required list. Any required identifier left outside the union belongs in the missing column. A complete selection has no missing required feature, regardless of how many repeated mentions its source lists contain.

The table keeps selected candidates, distinct features, missing requirements, selection count and completeness separate. Read “none” in the missing column as no requirement missing. Read “empty” in the selection column as no candidate chosen. Those statements are not interchangeable, and the empty selection does not cover our four required features.

Why must the answer retain both minimum pairs?

J with K covers R1 and R2 from J, then R3 and R4 from K. L with M covers R1 and R3 from L, then R2 and R4 from M. Both collections contain two demonstrations and satisfy the entire requirement list. Neither has a numerical advantage under the stated objective.

No singleton is complete: every supplied demonstration covers only two of the four requirements. The empty selection also fails. Consequently a complete pair proves that the minimum is two, not merely that two is a feasible count. Checking every pair reveals all ties rather than stopping at the first satisfactory answer.

The other four pairs remain incomplete. J with L misses R4; J with M misses R3; K with L misses R2; K with M misses R1. Keep these missing identifiers as the smaller-choice evidence alongside the failed singletons. Every triple is complete here, but three is larger than the known minimum of two.

Report both J+K and L+M. If the owner wants one recommendation, ask for an additional supplied rule, such as availability already confirmed outside this exercise. Do not invent a preference from candidate names. List order does not break an objective tie.

The product mix exercise maximises additive contribution under capacity limits. Here duplicated coverage does not accumulate useful value. The shift coverage check examines fixed people and time windows, not the smallest feature-covering collection. Shared table formats do not make these the same decision.

Distinguish impossible coverage from missing input

Suppose the owner adds required feature R5 while none of J, K, L or M includes it. Even the union of all four candidates lacks R5. No subset can cover everything, so there is no complete solution under that revised contract. Choosing the least incomplete row would answer a different question.

Now consider a different revision: the contents of M are unknown. Do not turn that unknown list into an empty list or copy the old one without authority. J+K remains a demonstrated complete pair if their supplied lists are unchanged, but you cannot assert the exact minimum or all tied minima while an unknown candidate might cover everything alone. Ask for M's membership.

If a separately supplied requirement list is genuinely empty, the empty selection is a complete minimum with count zero. That edge case does not apply to our four-feature source card. Conflicting candidate identifiers or incompatible membership versions also need clarification before an exact result is accepted.

An optional assistant request is: “Use only this source card. List all sixteen subsets, feature unions, missing requirements and selected counts. Return every minimum complete collection and the evidence that smaller selections fail. Do not create features, weights, availability or an extra tie-break.” Mark the optional comparison unperformed if you cannot access an assistant.

A source total alone is not evidence of the right conclusion. The mean and median review likewise keeps named measures separate. Here the separation is between repeated feature mentions, distinct feature union, complete coverage and minimum candidate count. None of those labels can silently replace another.

Save a record someone else can reconstruct

Open a plain-text editor already available on your device and create a new document. Head it “Minimum demo set V1: fictional memberships”. Copy the contract, the four original membership lists and the complete sixteen-row ledger. Add the conclusion: “Minimum count two; all minima J+K and L+M; every empty or singleton selection fails.”

Use the editor's save command, choose a folder you recognise and name the file minimum-demo-set-V1.txt. Close the document. Open that folder in your file manager and reopen the saved file. Check both minimum pairs and read the missing feature for J+L. It must still say R4. Seeing an unsaved editor window is not the same as reopening the intended file.

A proposed fifteen-minute allowance is two minutes for the contract, three for empty and singleton rows, four for pairs, three for triples and the conclusion, then three for saving and reopening. This is a planning allowance, not a timed participant result. A real presentation, account setup and software installation are outside the exercise.

For this article's calculation check, all sixteen selections were generated and compared with a separate feature-bit representation. The supporting text files were written, closed and read back through filesystem operations. No Windows graphical interface or reader completion time was tested. This describes the actual evidence without implying a software walkthrough was executed.

The dependency audit checks predecessor relationships and cycles. Our selection model contains no order, so a chosen pair is not a presentation schedule or permission to contact customers. Finish with the small verified collection and its assumptions, leaving operational decisions to the responsible owner.

• Preserve all four requirements and original candidate memberships.
• Keep sixteen distinct selections, not repeated orderings.
• Retain both minimum pairs and every failed singleton.
• Keep unknown membership separate from a known empty list.