Every number in Vouchd is a count of something that happened — never an average of how strangers felt. This page explains each one, exactly as computed. If a number can't survive this page, we don't show it.
Methodology v1.1 · 17 July 2026 — this document is versioned. When a counting rule changes, the version steps and the change is logged in the version history; a number never quietly changes meaning under the same name.
Product pages show one 0–100 poor→best track. Each colored marker on it is a behavioral aspect — owner verdicts, would-buy-again, still-going — and appears only once it has 3+ answers (tagged “early” under 5, joining the mean at 5). The thumb carrying the number is the V-Score™: the plain average of the markers on screen.
No secret weighting — you can check the math against the track itself. The sections below define each marker, exactly as computed. Category aspects (quality, fit, durability) will join the same track as owners answer them.
The survival rate at a fixed horizon: of owners whose 1-year fate is on record, the share whose unit is still alive and in use. It answers the one question a purchase-day review can't — does this thing last?
Every cohort starts at 100% and falls — that's physics, not failure. Quality is how slowly it falls, and how one generation's curve compares to the last. One year is the marker's fixed horizon; the survival record below runs the same arithmetic at 1, 2, 3 and 5 years.
The same still-going arithmetic, run at every horizon a product has earned — because “lasts” means different things at 1 and at 5 years, and the EU legal guarantee alone runs 2. This family of numbers is the durability evidence: it feeds our evidence sheets and rankings, while the product page carries the record itself.
Recompute it yourself. To check a printed row — say “2 yr — 71% still going · 12 of 17” — for any product:
If your recount disagrees with our print, one of us is wrong in public — write us. That is the point of this page.
A death is a unit that failed on its own: wore out, broke, or was recycled because it was dead. When an owner retires a unit as broken or worn out, one follow-up question settles the verdict — “Its own fault, or something that happened to it?”
How firmly a unit's start date is known decides how far its evidence reaches.
The 1-year numbers are untouched by this rule.
A unit that reached five years through two repairs and one that reached five untouched are different facts, so we run two clocks — EN 45552 draws the same line.
Both curves print on our durability evidence sheets; the product page carries the survival record.
Daily-use and occasional-use units die at different speeds, so owners tell us how hard a product works — daily, weekly, occasional — and the survival record can split by it.
Censored units — sold, gifted, lost, accident-damaged, recycled while working — count while observed and become unknown the day they leave. Buried in that (in every survival method, not just ours) is an assumption: that units leaving the story would have gone on to fail at the same rate as units that stayed. Statisticians call the failure of that assumption informative censoring, and it can cut both ways:
We disclose this rather than pretend to model it away; the prospective cohort below measures the recruitment half of the problem.
People often find Vouchd because something already broke — and failure-first arrivals inflate death rates. So our evidence sheets compute a second, cleaner cohort: owners who demonstrably arrived to own, not to complain. A journey registered within 48 hours of getting the product is prospective by construction — its owner joined before anything could have gone wrong, even if it broke on day 3. A backdated “had it a while” journey is prospective only if its first failure-flavored moment came 14 or more days after registration — failure-first arrivals log the failure within days of joining. Journeys from before we recorded registration time carry the date recording began — since any failure they ever logged predates that stamp, a failure-first arrival among them can never read as prospective.
Recruitment bias is measured, not hidden: the prospective cohort is computed under the same gates as everything above and printed beside the all-owners figures, both bases shown — the gap between the two lines IS the bias. And one sentence rides verbatim wherever these numbers travel: a low score is an observation, not an allegation of illegality.
EN 45552 is the EU's general method for assessing the durability of energy-related products; our estimators map onto its concepts one for one.
| EN 45552 concept | On Vouchd |
|---|---|
| Durability — function until a limiting state, under stated conditions of use, maintenance and repair | The survival record — repairs keep a unit alive; only a death stops the clock. |
| Reliability — durability without repair | The first-failure curve — the clock stops at the first breakdown, issue or repair. |
| Limiting state / failure | A death: worn out, broke, or recycled dead — the unit's own fault (an “inherent” verdict). |
| External damage — not a durability failure | The accident verdict — censored, never a death. |
| Conditions of use | The intensity strata: daily / weekly / occasional. |
| Maintenance and repair | Repair moments on the record — they end the reliability clock and keep the durability clock running. |
Behind every share sits a second, lower number we never print on product pages: the worst the true rate is statistically likely to be, given how many owners answered — the Wilson 95% lower bound.
For the record, once more: a low score is an observation about tracked units, not an allegation of illegality.
A counted verdict, asked of tracked owners months into ownership — never of strangers on delivery day. Each owner answers once per product: yes, no, or maybe — one answer per owner, and the latest wins if they change their mind. We count the yes votes against everyone who answered.
A verified maker can mark a version release — a dated claim, on the record. The mark can't touch a number: it only draws a line. Units that started before it keep their record forever; units after earn their own score under identical rules — same floors, same words-under-5, same denominators. The system never says “improved” or “fixed”; the after-column speaks for itself when it has the answers.
A percentage may only appear when it can't lie:
When you scan, Vouchd queries the official safety registers — the EU's Safety Gate and the US CPSC — at that moment. This works for any product, including ones no Vouchd owner has tracked yet. “No active recalls” means none found in those registers at scan time; it is a lookup, not a guarantee.
✓ Verified The ownership is backed by hard evidence — a purchase receipt attached to the journey, or in-box QR registration. This is the tier manufacturers can rely on.
Owns it The moment was logged live on a tracked, owned journey — a real owner, without receipt-level proof.
Anything less gets no badge. A badge that appears on everything means nothing — that's the “Verified Purchase” disease, and we don't have it.
Same here lives only on failure reports. It counts owners reporting the same fault — the single strongest signal a manufacturer gets from Vouchd: 41 owners, same zipper, same month is an engineering ticket, not a rating.
Helped decide counts real buying decisions a moment guided. It's how we measure a post's usefulness — a compass, not a like.
One owner's own verdict on their unit, on a 0–100 scale. It's testimony — one voice, clearly labeled. At product level it appears only as the labeled “owner verdicts” marker on the aspect scale, carrying its answer count — never as a hidden average behind a single number.
Under the hood it starts as a familiar 1–5 rating, one per owner (latest wins), then gets rescaled onto the same 0–100 track as every other marker:
Stars minus one, times 25 — so the worst possible answer really is 0, not a floor of 20%. It's the only place a 1–5 number ever touches Vouchd, and it never reaches you as anything but this rescaled marker.
EU consumer law — the Unfair Commercial Practices Directive, Annex I, point 23b — makes it a banned practice to say reviews come from real consumers without taking reasonable and proportionate steps to check that they do. These are our steps, in public; on Vouchd they are the architecture, not an afterthought:
Makers pay Vouchd for tools — verified replies, version marks, reports on their own products. Not one of those payments is an input to any number: the scoring pipeline has no field where money could enter, and nobody at Vouchd can move a digit by hand. A paid claim can never bury a fault report, reorder a list, or retouch a score.
If the data embarrasses a paying brand, the data stands and the invoice is unchanged.
Every number here is about someone's product, so the maker gets a voice — on the record, never on the scale.
A low score is an observation, not an allegation of illegality — and the reply channel exists so it never has to be the last word.
✗ Star averages — everything on review sites lands between 4.2 and 4.7; that's noise, not signal.
✗ A percentage without its denominator.
✗ A number touched by money — no ads, no affiliate links, and a paid brand claim can never bury a fault report or move a digit.
If you ever catch a Vouchd number that can't be explained by this page, write us: hello@vouchd.it.
v1.1 — 2026-07-17: prospective cohort rule made precise (48-hour registration window; 14-day grace for backdated adds; unknown-registration rows excluded).
v1 — 2026-07-17: first versioned edition. Added the survival family in full (multi-horizon record, cause-of-death verdicts, the exact-dates rule, reliability vs durability, intensity strata, the censoring caveat, the prospective cohort), the EN 45552 mapping, the verification pipeline, the money firewall and the right of reply.
A counting rule never quietly changes meaning under the same name: a change steps the version and leaves a line here. This page mirrors the “How we count” screen in the app, section for section.