Methodology · Decode Score
How Decode Score works.
Decode Score is our reading of what a product’s documented evidence shows it does, for one job at one exact configuration. Here’s what we evaluate, which evidence counts and how to read a result. The arithmetic stays private.
- What it is
- A 0–10 reading of documented evidence for one function of one configuration, with a confidence level.
- What it isn’t
- A star rating, a review average, or a verdict on whether it suits your home.
- Where you’ll see it
- Product pages, comparisons and Decode Score ratings. The product page shows the evidence behind it.
From evidence to a published result
Documented evidence
Records we hold
Certification listings, lab reports and brand documents, each recorded with who produced it, what it covers and when.
Rule-based evaluation
Read against one set of scoring rules
The evidence is read against the scoring rules for the product’s function. Every product of that type faces the same questions.
Published result
Scoped to one function
A Decode Score with its confidence and the records behind it, released in a dated revision.
That’s the whole public sequence. Weightings and arithmetic aren’t shown.What stays private
One job, one configuration.
A product is evaluated for each job it does, such as reducing contaminants at the tap, softening, conditioning or disinfecting. It’s evaluated at the configuration on record: the model and its cartridge or media.
Limit. A different cartridge or media is a different configuration, and it may have no result.
Illustration · one product, two jobs
A carbon filter and a salt-free conditioner in one system
Result 1
Filtration
Ranked with whole-house systems
Result 2
Scale reduction
Ranked with salt-free conditioners
Where the evidence comes from.
Two products may advertise the same lead reduction percentage. What differentiates them is who did the testing and under what conditions the products were tested. Every record is labeled with who produced it. We use three kinds of source and never blend or relabel them, and a brand’s own test never counts as independent.
Independent certification
A certifier lists it
A certification body tested the product against a standard or its own protocol, and lists it publicly.
Can show
The named claims on the listing, at the certified configuration.
Can’t show
Claims outside the listing’s scope. A standard number alone doesn’t cover every contaminant.
Independent testing
A lab or tester measured it
A laboratory or tester not run by the brand measured the product.
Can show
What that test measured, under its stated conditions.
Can’t show
Certification. A test report isn’t a listing, and we label it as a test report.
Brand-supplied
The brand says so
Test reports or specifications from the brand itself.
Can show
What the brand says it tested or specifies.
Can’t show
Independence. Brand tests are never counted as independent testing.
7 evidence families on product pages
Six of them belong to one kind of source each. The seventh marks an input that isn’t published anywhere we can source it. Families describe where evidence came from. They aren’t grades, and this order isn’t a ranking.
Scope. A standard number isn’t the whole claim. We show which claims a listing covers.
Verified certification
Independent certification
Certifier-measured result
Independent certification
Accredited laboratory report
Independent testing
Independent measurement
Independent testing
Manufacturer-documented test
Brand-supplied
Manufacturer-stated specification
Brand-supplied
Not published, so a fixed rule applied
No source on record
Reading a published result.
A result is more than a number. It tells you what it covers, how firmly the evidence supports it, what it rests on and when it was published.
Decode ScoreExample
One configuration from our research set
Evidence on record
- Manufacturer-documented test8 records
Covered, not quantified: how much it reduces bacteria, cysts and viruses
Example only. Not a published revision.
Function
The one job this score covers. Other jobs get their own results.
Score
Comparable only with others in the same group.
Confidence
How firmly the evidence supports the result.
Evidence
Every record behind it, labeled by the kind of source.
Gaps
What’s missing is named, never assumed.
Revision
When this result was published.
Illustration only. It shows what a result tells you, not how it’s calculated. We computed this result for one configuration in our research set. The product’s name is left off.
Confidence
High
The evidence behind this score is strong, and close to complete.
Confidence
Moderate
The score stands. There are gaps in the evidence behind it, listed on its product page.
Confidence
Limited
The score stands, and the evidence behind it has real gaps. Read it with the reasons on its product page.
When there’s no number
No Decode Score
It means only that we cannot currently calculate a trustworthy Decode Score for this exact function, product and configuration. It is never a zero, and never a finding that the product fails.
Why there may be no numberConfidence describes the evidence, not the product’s quality. A missing record counts as missing, never as a pass.
What can keep a result out of a higher band:
- Depth of the evidence
- No scored evidence
- Unpublished inputs
- Certification standing
- A conflicting result
- An adverse finding
- An aged independent test
Why a result can have no number
Every result is in one of four states. When there’s no number, the product page gives one of the reasons listed here. None of them is a finding about how well the product works.
The four states
- Decode Score
- Scored for this product’s own function, from the evidence we hold.
- No Decode Score
- It means only that we cannot currently calculate a trustworthy Decode Score for this exact function, product and configuration. It is never a zero, and never a finding that the product fails.
- Score temporarily unavailable
- We have paused this score. The reason is on our side of the record, not a finding about the product.
- Not included
- This product does not perform the function being compared, so there is nothing here to score.
We do not publish an uncertainty range. We can work out internally how far this score could move on what is unknown. We do not publish that as a range, so no interval is shown here.
The reasons we give
- We cannot pin down which exact configuration this is
- We do not score this category yet
- We do not score this function yet
- We support this function but don’t score it yet
- We hold no performance evidence that meets our bar
- The outcome that defines this function is not demonstrated
- We hold too little to score the defining outcome
- We are still researching this one
- A specification the score needs is not published
- A critical check in our scoring rules did not pass
- This record is under review
- A certification this score relied on was withdrawn
- We have paused this score
A great score isn’t always a great fit.
Decode Score judges the product. Finder judges the match with your home, and its answer is not a score.
The product
Decode Score
- Asks
- How well does the evidence support this product doing its job?
- Uses
- Product evidence only.
- Gives you
- A score for one job, with a confidence level.
- Same for everyone?
- Yes.
Your home
Household fit
- Asks
- Does this product suit your water, your home and your budget?
- Uses
- Your answers plus product facts.
- Gives you
- An explanation of fit. It is not a score.
- Same for everyone?
- No. It’s about your home.
What we publish, and what stays private.
Published with every result
- Which scoring rules a product is judged under
- Every evidence record behind it, labeled by the kind of source
- The result, its confidence and its revision date
- Any reason a result has no number
Kept private
- Internal weightings and thresholds
- The arithmetic that turns findings into a number
Publishing the recipe would let products be built to the score instead of to the job.
What you can check instead: every record behind a result, on its product page.
Updates and revisions
Results are published in revisions, and the ratings page shows the date of the current one. Each revision reads the records as they stand when it is made, so a change we record, such as a new listing, a lapsed certification or a recall, is read in the next one. A product with a recall on record isn’t ranked.
- New listing
- Lapsed certification
- Recall
Limit. We don’t run our own lab. A revision date is when results were published, not when anything was tested.
See what’s changedHow we’re funded
This site is funded by affiliate commission: when you buy through one of our links, we earn a cut. Rather than a promise about our own conduct, here’s the checkable part: price, retailer and link data are not inputs to the score. On a product page, a price is shown as a recorded price with the day it was recorded, never as a live price. The affiliate disclosure explains how it works.
Affiliate disclosureSpotted something wrong?
Tell us through Feedback, with the page and what you think the record should say. We check reports against the records we hold. When we correct a record, the next revision reads the corrected one.