When the calculator app went subscription, users went back to the hardware

The disconfirmation first: breadth could not be measured

The claim this study most wanted to test is the one it cannot report. A first pass at breadth and missing-calculator complaints failed its hand audit at an 85% false-positive rate: the pattern was matching paywall complaints such as "no free option" and "no option to buy" rather than missing trade functions. It was rewritten to require an explicitly named absent capability, and the honest result is a refusal rather than a finding. Breadth versus depth is not a measurable complaint in this corpus, so Tenon Tools publishes no breadth count from it, and nobody should quote one in either direction.

What the positive scan could measure is narrow. It tested exactly four patterns: pay once, no ads, works offline, and simple. The broadest match by far is simplicity, at 22,176 positive reviews, 24.1% of the calculator category's 92,009 positive reviews. Power, trade coverage and depth were never measured as praise patterns at all, so nothing here ranks simplicity against them.

Two caveats travel with that 24.1% and neither is optional. The pattern is deliberately broad, matching any "simple" or "easy", and the study reads it as generic satisfaction rather than as evidence that anyone owns simplicity. And it is concentrated: 14,554 of the 22,176 come from Invoice Simple, an invoicing app rather than a calculator. App names are stripped before matching, because "Invoice Simple" alone scored 18,015 false hits on its own name, but only the name was stripped. That one app's 14,554 genuine matches are still inside the total.

The switching table: hardware first, one app second, then nothing

Construction Master Pro is the reference case for three reasons: it is the highest-volume construction calculator in the study's published Tier-1 coverage table, at 2,123 reviews and 604 critical ones; its live listing carries an in-app-purchase and subscription badge; and it holds a 4.84 live rating while doing so. A separate app-level analysis of that app counts 280 critical reviews mentioning subscription or pricing, then probes the same app's reviews against a fixed list of nine named alternatives. The list is closed, so an alternative nobody thought to put on it returns nothing by construction.

Two of those zeros are a listing artifact

The study's Tier-1 coverage table separately records Handyman Calculator and DEWALT Mobile Pro as not on either store, with 0 reviews each. A name that has no live listing cannot be switched to, so those rows are a delisting artifact rather than a rejected choice. The other five zeros are simply names that nobody in this set volunteered.

Churn left the app market, for the reviewers who name a destination

The strategic content of that table is in its first row, and its limits are in its totals. Two rows carry every mention: 29 and 9, 38 in all. Construction Master Pro has 2,123 reviews in this corpus, 604 of them critical and 280 of those mentioning subscription or pricing, and the study does not state which of those sets the probe ran over. Under any of them the reviewers who name a destination are a small minority, and the corpus cannot say what the rest did. What it can say is that among the reviewers who name one, hardware outranks every app on the list.

One reviewer of Construction Master Pro writes that they will "just buy another physical calculator so that I have one in my truck" (Google Play review, Construction Master Pro Calc, 2026). Another puts the objection in one line: "it's a calculator bro not a netflix account" (Google Play review, Construction Master Pro Calc, 2026).

BuildCalc, the one app on the probe list that registers at all, is a one-time paid app with 172 reviews, 21 of them critical, and a 4.70 live rating. It is a small listing, and on this list it is the only app reviewers name as somewhere to go.

The design implication is narrow, and it is the only one this table supports: for the reviewers who name a destination, the thing being compared against is a pay-once physical device, not the next subscription app in the search results.

The anger has a curve, and the curve is not proof the objection died

Subscription-anger reviews for Construction Master Pro, by year. Read the caveats before the shape. 2026 is a partial year because the corpus ends 2026-08-14, so its figure is not comparable to a full one. And a falling complaint count is not evidence that the objection went away: the population that objects is finite, a user who churns to a physical calculator stops writing reviews about the app, and a declining curve after a pricing change is also what a completed exodus looks like from the inside.

• 2022: 38 reviews
• 2023: 86 reviews (the peak)
• 2024: 78 reviews
• 2025: 44 reviews
• 2026: 34 reviews (partial year, corpus ends 2026-08-14)

Subscription anger is a Tier-3 phenomenon. Ads are Tier-1. Never average them.

This split is market-wide across all three categories the study covers: construction calculators, timestamp cameras and site-diary apps. The four tier rows sum to the whole in-scope corpus, so none of them is a calculator-only figure. The denominator in each row is that tier's own critical reviews.

• Tier 1, fixed list, all three categories, 35 apps, 79,429 reviews, 13,716 critical: ads 2,080 (15.16%), subscription anger 375 (2.73%), crashes 483 (3.52%)
• Tier 2, fixed list, all three categories, 39 apps, 16,249 reviews, 2,861 critical: ads 153 (5.35%), subscription anger 113 (3.95%), crashes 133 (4.65%)
• Tier 3, fixed list, all three categories, 11 apps, 60,594 reviews, 6,895 critical: ads 127 (1.84%), subscription anger 1,153 (16.72%), crashes 370 (5.37%)
• Sweep-discovered, all three categories, 150 apps, 52,267 reviews, 4,940 critical: ads 347 (7.02%), subscription anger 241 (4.88%), crashes 212 (4.29%)

Why that split changes the conclusion

Ads are a Tier-1 phenomenon: 2,080 of 13,716 Tier-1 critical reviews, 15.16%, against 127 of 6,895 in Tier 3, 1.84%. The free consumer-facing apps at the top of the store carry the ad load, and the professional tools essentially do not. Subscription anger is the exact mirror image: 1,153 of those Tier-3 critical reviews, 16.72%, against 375 of the Tier-1 ones, 2.73%. Tier 3 is where the software-as-a-service products sit, and the study describes its volume as dominated by one invoicing app.

A blended market average for either complaint therefore describes no real app. Averaging 15.16% and 1.84% into a single ads figure invents a market segment that does not exist, and the same applies in reverse to subscription anger. Any competitive claim built on this corpus has to name both its tier and its category, because those are two different cuts of the same reviews and neither substitutes for the other.

Crashes behave differently again, and that is itself the finding: 3.52%, 4.65%, 5.37% and 4.29% across the four tiers. Stability is flat on the tier axis, which makes it a baseline quality obligation rather than a differentiator anyone can market. Flat across tiers is not the same as flat across categories, and this page makes no claim about the second.

What positive reviews praise, and what that cannot prove

Against 92,009 positive reviews in the calculator category, praise for the things a pay-once product would advertise is thin. "Pay once" or "no subscription" praise appears in 71 reviews (0.08%). "No ads" praise appears in 234 (0.25%). "Works offline" praise appears in 15 (0.02%).

This is a weak instrument and it should be reported as one. Satisfied users describe how an app felt to use, not how it was priced; the pricing model surfaces in anger, not in gratitude. So 0.08% is not evidence that pay-once pricing does not matter, and it is not evidence that it does. It is evidence that positive reviews are the wrong place to look for a pricing signal. Critical reviews are where one can be looked for at all, and they are an imperfect instrument too.

The hypothesis counts across the calculator category

These are category-wide counts that cross all four tiers, with the calculator category's 9,373 critical reviews as the denominator. They answer a different question from the tier table above and must not be substituted for it: the ads row here is 241 across this one category, while the 15.16% above is Tier 1 across all three.

• Subscription anger: 1,540 mentions, 16.43%, across 17 apps
• Ads: 241 mentions, 2.57%, across 14 apps
• Crashes and stability: 223 mentions, 2.38%, across 18 apps
• Quote, PDF and estimate output: 201 mentions, 2.14%, across 6 apps
• Trial and free-tier friction: 167 mentions, 1.78%, across 15 apps
• Unit-switch and conversion friction: 81 mentions, 0.86%, across 8 apps
• Wrong results and accuracy: 68 mentions, 0.73%, across 9 apps
• Offline expectation: 33 mentions, 0.35%, across 9 apps
• Dated or emulator-style interface: 18 mentions, 0.19%, across 5 apps
• Localization gap (German): 1 mention, 0.01%, across 1 app

What that ranking hides, and what is missing from it

One pattern is deliberately absent. Breadth and missing calculators failed its audit and was rebuilt, and the study treats it as unmeasured rather than as a count, so it has no row here and no figure anywhere on this page.

Concentration matters more than rank order in this category. Of the 1,540 subscription-anger mentions, Invoice Simple accounts for 766 and Joist for 238, both of them Tier-3 invoicing products rather than calculators. Of the 201 quote, PDF and estimate-output mentions, Invoice Simple is 117 and Joist is 71. Tenon Tools builds PDF quote export into its own calculator, and that is the honest state of the evidence for it: the demand signal is measured almost entirely in invoicing apps, not in calculators.

What the corpus volunteered on its own

A separate rule set counts unprompted themes rather than the ones the study went looking for. The two rule sets are not comparable, and their counts must never be added, averaged or printed as one series. On this one, pricing and subscription anger leads at 1,666 mentions, 17.77% of category critical reviews, across 35 apps, with Invoice Simple alone at 829. Ads follow at 386 (4.12%), then unit and measurement friction at 344 (3.67%), crashes at 304 (3.24%), missing features at 217 (2.32%) and PDF or export complaints at 150 (1.6%).

Unit friction at 344 is the entry worth stopping on. It is the third-largest volunteered theme in the category, ahead of crashes, and it is a mechanical problem: entering feet and inches, moving between metric and imperial, getting fractions back instead of decimals. Of the 304 crash complaints on this rule set, 45 (14.8%) are from 2025 or 2026.

The two rule sets disagree by design on accuracy, and that is worth showing rather than hiding. The hypothesis pattern counts wrong-results complaints at 68 (0.73% of category critical reviews). The volunteered-theme rule set counts the same subject at 130 (1.39%). Neither corrects the other. They are two classifiers with different rules over the same reviews, and 68 and 130 must never be printed as one series.

Germany, and what 204 critical reviews can and cannot say

The German storefront is small in this category: 1,628 reviews and 204 critical, against 99,755 reviews and 9,169 critical in the United States. German-language reviews are 50.1% of the German storefront's total, 816 of 1,628. Every German percentage below carries its absolute count, and a reader should refuse any German-market statistic that arrives without one.

On the volunteered-theme rule set, wrong-results complaints are 34 of the 204 German critical reviews, 16.7%, against 96 of 9,169 in the United States, 1.0%. That gap is the subject of a separate Tenon Tools article, linked below, which sets out what a 204-review base can and cannot support.

On the hypothesis rule set, subscription anger on the German storefront counts 0, with all 1,540 of the category's mentions on the United States storefront. Against 204 German critical reviews, that zero is a thin-instrument result, not a finding about German buyers.

Language requests are rare in this category: 21 reviews out of 101,383, or 0.02%, ask for a language or complain about translation quality. Reviews are the wrong instrument for that question in an obvious way. A tradesman who never finds an app in his language does not install it and does not review it, so 0.02% is not evidence that language coverage does not matter.

What Tenon Tools did with this

We ran this study and we build a construction calculator, so the conflict of interest is on the table. Three findings shaped the product. The pricing model is pay once, because the alternative those reviewers name most often is a device that is bought once. The scope is closed and small, because the only praise pattern that measured large is simplicity and the breadth complaint could not be measured at all. And live metric-to-imperial switching is a headline feature rather than a settings screen, because unit friction at 344 mentions is the third-largest thing this category volunteers.

Two claims we refused to make. We did not build the pitch on "users hate subscriptions", because that claim holds in one tier, 16.72% of Tier-3 critical reviews, and collapses in another, 2.73% of Tier-1. And we did not claim that pay-once pricing is a proven demand signal, since it appears in 71 positive reviews, 0.08%, and the instrument cannot carry that weight in either direction.

How to check this

Denominators matter and are stated per figure. Hypothesis and emergent-cluster shares use critical (1 to 3 star) reviews in that scope. Wedge-occupancy shares use positive (4 to 5 star) reviews. Localization-gap shares use all reviews in the category. Tier shares use critical reviews within that tier.

Three instruments in this study count Construction Master Pro's subscription and pricing anger, and they do not agree: the app-level analysis gives 280, the hypothesis classifier gives 297 and the volunteered-theme classifier gives 302. This page quotes 280 throughout, names the instrument at the point of use, and prints the other two here so that nobody finds them later and assumes something was hidden. They are three rule sets applied to one app, not three attempts at one number. Do not average them.

Two earlier figures from this study are void because their patterns failed a hand audit and were rebuilt: a calculator-category ads count of 350, where the correct post-fix figure is 241, and a timestamp-camera ads count of 2,269, where the correct figure is 1,906. Neither should be quoted from any older copy of this research.