Draw your test suite as a diagram and it will have a shape. Whether that shape was chosen or merely accumulated is one of the quiet predictors of how a software team’s week goes. TestOps — running the automated checks as a managed service — begins with exactly this act of looking: naming the shape you have, comparing it with the shapes available, and deciding on purpose. Eye clinics will recognise the method. Vision gets corrected against a measured prescription, never against whatever glasses were lying in the drawer.
Four TestOps Shapes, Compared Side by Side
The taxonomy most teams inherit comes from Mike Cohn, who drew the testing pyramid around 2003-04: a wide base of small, quick checks, a narrower band of service-level checks above it, and a few full-journey interface checks at the peak. It became canonical partly because the alternative in most shops was its inversion — and that inverted thing, mostly slow interface checks floating on a thin base, earned its own name: the ice-cream cone. Every experienced practitioner treats the cone as a finding, not a strategy.
Two later challengers are worth comparing honestly rather than as fashion. The honeycomb, proposed for products built from many small services, concentrates on integration-level checks between services, with fewer pure unit checks and fewer full-journey ones — on the argument that in a microservice product, the interesting behaviour lives in the seams. The trophy, popular among front-end teams, puts integration work at the widest part too but keeps a meaningful static-analysis layer and a slimmer unit base, reflecting what actually breaks in rich browser applications.
Set them against three practical axes and the choice clarifies:
- Cost per run. Unit checks are pennies; full-journey checks through a browser are pounds, sometimes with the waiting time counted in minutes.
- Determinism. The closer a check sits to the interface, the more it depends on rendering, timing and networks — and the more often it reports trouble that isn’t there.
- Where the defects live. Business logic in one codebase favours a wide base; behaviour emerging between services favours the honeycomb’s middle bulge.
None of the shapes is correct in the abstract. A claims engine and a consumer banking app will rationally choose different proportions, and both will be right — the same logic that makes one prescription useless for the next patient in the chair.
Size Classes: The Other Comparison Worth Making
Shape describes proportions; size class describes runtime budget, and the second comparison matters just as much. Microsoft’s internal grading is the cleanest public example. Level 0 checks are tiny — the guidance targets an average around sixty milliseconds, with nothing permitted over two seconds. Higher levels climb through component and scenario checks until level 4, which runs against production itself. Google’s equivalent grades by consumed resources, but the principle matches: the class decides where a check may run, how long it may take, and how much of it you can afford.
Comparing the two schemes teaches the real lesson. Shape asks “what proportion of each kind do we own?” Size class asks “what does each kind cost, and how steadily does it answer?” Teams that adopt even a crude version of the second scheme stop arguing about test counts and start arguing about budgets — a far more productive argument, and one with numbers attached.
A word on changing shape once the current one is diagnosed, because this is where good intentions usually expire. Reshaping wholesale — deleting a layer of interface checks over a weekend — is surgery without a plan, and it fails in the predictable way: one incident, one rollback, one permanent retreat to the old prescription. The workable pace is one contested check at a time, each demoted or rewritten only when its cheaper replacement has proven it catches the same defect. Months, not weekends. Suites, like corneas, respond badly to sudden pressure and well to steady, patient correction.
Choosing TestOps Proportions by Cost, Not Fashion
The decision rule that survives contact with reality is two questions per layer. What does one run of this layer cost? And how often does this layer lie? Whatever the answers, weight your suite toward the layers that are cheap and truthful, and keep just enough of the expensive, wobbly layers to cover what only they can see. That is the entire engineering content of TestOps shape-craft, and it explains why the online arguments about diagrams generate more heat than light: the diagram is a drawing of a budget, not the budget itself.
Two health-service analogies keep the rule honest. Choosing among hospitals is an exercise in matching capability to need rather than picking the most famous name — precisely the argument made in picking an eye hospital by capability and fit rather than by fame — and the same matching logic should pick your suite’s proportions. And the checks that continue after the main event matter as much as the operation itself, which is what both cataract post-care programmes and a post-deployment smoke pass understand: discharge is not the end of observation.
The Practical Scorecard
For a team that wants to act this week, in order:
- Sketch the current proportions honestly — count checks per layer if the data is cheap to get.
- Estimate cost and flakiness per layer, even roughly; two columns on a whiteboard suffice.
- Name the gap: if the expensive, wobbly layer dominates, you are wearing the cone and paying for it daily.
- Move one decision at a time — usually migrating a contested check down a layer, where it runs faster and steadier.
- Re-examine whenever the architecture changes, because the right shape is a property of the product, not of the team’s taste.
Full taxonomies, sizing guidance and the research behind each shape are collected in the Ops Atlas overview of TestOps, which makes the natural companion read to this comparison. One closing caution from the eye department: correction works best when measured regularly, because prescriptions drift. So do suites. The shape you chose two years ago may be quietly squeezing your delivery the way an old lens squeezes a temple — painlessly at first, and expensively by the time anyone complains.
Remote consultation earns its place in the same picture: just as telehealth eye-care services triage the routine so clinics can concentrate on the serious, a well-shaped suite reserves its expensive layers for the questions only they can answer. Triage, budgets, and honest prescriptions — the vocabulary of vision care turns out to fit testing rather well.