Methodology

Headline shares, meters and helpful scores come from one published rule and one set of published constants. The rule is below. The constants are at the bottom. Nobody here can override the count, and every number carries a Verify link that opens the events behind it in the public vote log.

Two tracks, never blended

Every listing has two tracks of reviews. The community track is anyone signed in with REVA. The vetted track is a short list of named reviewers, admitted against public criteria, who disclose their tradition. The two tracks sit side by side. The vetted column appears once an edition has 3 counted vetted reviews. There is no composite score, and nothing on the site adds the two tracks together. How vetted reviewers are admitted is on the vetted reviewers page.

What a vote is

A vote is one of two things. A one-tap answer to “Would you use this again?”, yes or no. Or a written review, which carries that answer plus the category’s meters and the text. Both are written to the public vote log as events. Helpful and Not helpful clicks on reviews are logged too, because they change sort order.

A vote is about the thing listed, never a person. For a Bible or a book, the vote is about the edition. For an app, the current version. For a channel or a streaming service, the channel or service as a whole. For a podcast or a worship catalog, the feed or catalog. For a kids’ title, the vote is an adult’s answer about their own household’s use.

Published first, counted by rule

Every review you can see is live. It counts toward the numbers only when it passes the eligibility rule: the account is 14 days old, the body is 200 characters or more, and the author has no connection to the maker. If a review is published but not counted, the reason is printed beside it. The rule has no exceptions and no manual switch.

The headline share

The headline on a listing is the share of counted responses that said yes. Counted responses are the counted one-tap votes plus the yes-or-no answers carried by counted written reviews. The raw share is shrunk toward a prior with a weight C:

share = (yes + C × 0.70) / (n + C), where C = 7 and n is the number of counted responses.

The prior is 0.70 in scoring v2, the same for every listing in every category. The effect of C is that a listing with few responses is pulled toward 70 percent, and the pull fades as responses arrive. Worked example: 27 yes out of 31 counted responses gives (27 + 4.9) / (31 + 7) = 0.839, shown as 84%.

The share is shown as a whole percentage. It appears at 3 counted responses. Below that a listing says “Not enough counted responses yet” and shows its count.

The meters

Each category has three or four meters. A meter runs from 1 to 5 and neither end is better. Only counted written reviews feed a meter; a one-tap vote carries no meters. Each meter is shrunk toward the middle with the same weight:

meter = (sum + C × 3.0) / (n + C), where C = 7, sum is the total of the positions given, and n is the number of counted written reviews.

A meter is shown to one decimal with its n beside it. The meters by category:

Category	Meters
Bibles	Readability (dense to easy); literal to idiomatic feel; study helps (bare to extensive); devotional to academic feel
Apps	Simple to deep; free to paid; offline to online
Websites	Simple to deep; ad-light to ad-heavy; text to tools
Kids	Screen-free to screen-heavy; read-to to read-alone; Bible text to retelling
Watch	Teaching to entertainment; short to long; tradition-neutral to tradition-specific
Listen	Teaching to conversation; episode length (short to long); music to spoken
Read	Devotional to academic; short to long; standalone to series

Fidelity questions, such as whether a translation is accurate, are answered by the spec sheet and the Disputes panel on each page, never by a number.

Confidence words

Every share carries one of three words. They describe how many people answered, not how good the thing is.

• few: under 3 counted responses
• some: 3 to 9 counted responses
• many: 10 or more counted responses

Segments

“Readers like me” filters the reviews by the tradition each reviewer chose to show. A tradition segment appears only at 5 or more counted reviews. Tradition never weights anything: the headline share and the meters are computed over everyone, and a segment only changes which reviews you are reading.

“Verified use” is a filter and a badge, not a weight. Verified use comes from the reviewer’s REVA reading record where available. Otherwise a review says “Self-reported use”. Reviews count in any language; the language filter only changes what is shown, and we never translate a review.

Sorting

On a hub, the default order is most counted reviews, then alphabetical. Three other sorts are offered:

• Would use again sorts by the headline share. Only listings with 10 or more counted responses join this sort.
• Most debated sorts by how split the counted responses are: how close the raw share is to one half, weighted by the logarithm of the count. This is used for sorting only and is never shown as a number.
• A to Z is alphabetical.

A listing joins the sorted list once 3 written reviews are counted. Until then it sits under “Listed, not enough reviews yet”, alphabetically, with a full spec sheet.

Within a listing, reviews sort by Most helpful or Newest. Most helpful uses the lower bound of the Wilson score interval over Helpful and Not helpful clicks, with z = 1.281551565545 (one-sided 90 percent). That bound is used for ordering only and is never published as a number.

Ties break against REVA. In every sort, filter and segment, if REVA, our own app, ties with another listing, REVA is shown below it. It never sorts above a tie. The rest of what we took away from ourselves is on the conflicts page.

Editions

Scores are computed separately for each edition and each track. A vote and a review are pinned to the edition they were written about. Older tap votes whose edition was never recorded remain visible in a separate “Edition not recorded” group; we do not guess an edition. A revised edition is a new row, not an edit to the old one. When a text is revised, readers see a note on each older review saying which text it describes.

How the numbers are produced

One job recomputes every listing’s numbers nightly from the exported events, using the open scorer pinned to scoring version v2. The same job produces the transparency page. Its operational request, report and reviewer-admission totals come from service records; they are not public replay scores. There is no field in which a number can be typed. The export, the scorer and the script that checks the chain are described on the Verify page.

What is never scored

• Theology or doctrine. We believe the Bible is the word of God. We do not score anyone’s theology here.
• Tradition. Named when the reviewer or the maker chooses to show it. Never scored, never weighted.
• Accuracy or fidelity. Answered by the spec sheet and the Disputes panel, with each position attributed. Never a number.
• Christian-made or general. A label from the maker’s own pages, with a source and a date. A general product is never marked down for being general, and a Christian-made one is never marked up for being Christian. Same facts, same meters, same vote.
• Institutional positions. An approval is not a quality endorsement, and a non-approval is not a quality defect. Positions are listed, dated and sourced.
• A person. Channels and services are reviewed as channels and services. A review of a person is removed as a review of a person.
• Popularity. Subscriber counts, sales and download figures are not shown. They are not a vote.
• Our own app. REVA’s page carries facts only, no editorial from us, and no rating markup for search engines.

Constants, scoring v2

Effective 22 October 2026.

Constant	Value
C (shrinkage weight)	7
Share prior	0.70
Meter prior	3.0
z (Wilson bound, sorting only)	1.281551565545
Rank floor (joins the “would use again” sort)	10 counted responses
Segment floor (a tradition segment appears)	5 counted reviews
Share floor (the headline share appears)	3 counted responses
List floor (joins the sorted list)	3 counted written reviews
Minimum body	200 characters
Minimum account age	14 days
Activity rule (10 chapter reads or 3 study artifacts)	off
Recent window	90 days

Any change to a constant or a formula gets a new version label, a date, a line in the changelog, and a scoring.version event in the public vote log. The constants are also served as JSON so that anyone recomputing the site reads the same values we do.

Scoring v2 retains the v1 formula constants and corrects the aggregation scope: editions and tracks are isolated. The public export includes eligibility attestations and missing legacy metadata. These attestations do not independently prove account identity or past eligibility. The resource JSON identifies the exact log prefix used for its results.