← Open data

Release notes frequency, 327 public changelogs measured

Entries on 327 public Canny and Featurebase changelogs, 16,060 of them, read on 3 October 2026. The hosts are teams with a public Canny or Featurebase board, the same 902 candidate addresses the board votes study reads. Each host is one row in the file below, with its entries in the last year, its total entries read, the age of its newest one, its median words, its notify share and whether it was capped. No company, board or host is named.

The median host with a changelog posts 4 entries a year, and 99 of 327, 30 percent, posted none at all in the year to the read date. 198 of 327, 61 percent, have gone 90 days without a new entry.

id,vendor,entries_365,entries_total,newest_age_days,median_words,notify_share,linked_share,capped
1,featurebase,12,76,247,55.0,52.6,,0
2,featurebase,2,7,269,116,71.4,,0
3,canny,0,135,883,60,0.7,0.0,0

Download the file. It is free to copy, quote and use commercially. To cite it, link https://votefirst.app/en/data/release-notes-frequency and give the read date, 3 October 2026.

How often do teams with a public Canny or Featurebase board publish release notes?

Four entries a year, at the median host. 99 of the 327 hosts, 30 percent, posted none in the year to 3 October 2026, and a quarter posted 15 or more.

Of the 902 candidate hosts, 160 no longer exist under that name. Of the 742 that still answer, 327, 44 percent, publish a changelog with at least one entry, 392 answer with an empty one, and 23 refused the request. The method below sets out each count.

How long does a release notes entry run?

84 words, the median across all 16,060 entries. A quarter run 45 words or fewer, a quarter 161 or more. 4,555 entries, 28 percent, are under 50 words, and 432, 3 percent, run past 500.

How are entries labelled?

53.5 percent of entries carry the label new, 31.3 percent improved, and 17.1 percent fixed, counting Canny's own entry types and Featurebase's changelog categories the same way, lower case. 1,861 entries, 12 percent, carry no label at all.

How often are subscribers emailed when an entry publishes?

The two vendors record different things under that question, so the shares are not merged. Canny's notifySubscribers is set on 33.1 percent of its 10,843 entries. Featurebase's emailSent is set on 60.6 percent of its 5,217 entries.

A separate reader found the same gap on the request side of Featurebase boards: of 10,476 completed requests on 341 Featurebase hosts with 10 or more requests, where Featurebase records whether a status update email went out, one was sent for 4,710, 45.0 percent. A reply for each status change covers the other 55 percent, where Featurebase records no status email.

How often does an entry name the request it answers?

Only on Canny, where an entry can carry the IDs of the requests it closes. 3.1 percent of its 10,843 entries name at least one, and where one is named, the median entry names exactly one. The other 96.9 percent carry no linked request.

What does a release notes habit look like at the median and the top quartile?

At the median host: 4 entries in the last year, and a gap of 13 days between entries, counted on the 282 of 327 hosts that have posted on at least two dates. The two figures fit together: entries arrive in bursts around a release, then stop. 198 of the 327 hosts, 61 percent, have gone more than 90 days since their newest entry.

Entries go out on weekdays. Thursday carries 19.9 percent of the 16,060, Monday the fewest of the five at 17.0 percent, and Saturday and Sunday together 8.5 percent, by the UTC date of publishing.

At the top quartile: 15 or more entries in the last year, about one every three to four weeks at a steady rate. 91 of 327 hosts, 28 percent, have an entry within the last 30 days, and 228, 70 percent, within the last year.

What VoteFirst's changelog does

VoteFirst's release email goes only to the voters who were signed in with an account when they voted and whose vote is on a request in the release, naming the requests they voted for. Nobody who voted anonymously is emailed. More on how it builds and sends that email.

How was this built, and when does it change?

The candidate list is the same 902 hosts the board votes study reads, drawn from *.canny.io and *.featurebase.app addresses in Common Crawl's URL index with the two vendors' own marketing, app, docs and status hosts removed.

Each host's changelog was read through the same public endpoints its own /changelog page uses: Canny's POST /api/changelog/getEntries and Featurebase's GET /api/v1/changelog, with no sign in, one request a second. For Canny, the largest pages value the endpoint honours was found by trying values up from 50: pages:50 returns 500 entries; pages:51 and above return {"error":"invalid input"} with an HTTP 400 status. 5 hosts had more entries than that and are flagged capped in the download; their newest-entry age and totals still come from what was read, so they understate the host's full history.

Both endpoints were checked against the page a person sees before any figure was trusted. feedback.canny.io's newest entry and its custom label names matched the page's own embedded JSON. feedback.featurebase.app's totalResults matched the count the page renders.

Two silent filters were checked for and neither was found. Canny's getEntries already restricts its result to status: "published" and deleted: false; a 182-entry sample held no draft or scheduled row. Featurebase's listing already restricts to isPublished: true and state: "live"; a 118-entry sample held none in any other state, and locale duplication was checked for and not found, since every sampled entry's publishedLocales held only en.

Of the 902 hosts, 575 carry no changelog in this reading. 392 answered with an empty entries list. The other 183 were first logged as unreachable by the collector's broad error handling; a second, targeted request to each of those 183 found that every one of them answered. 160, all on Canny, answered with an error body. Ten of them, every sixteenth in host order, were opened again: each answered {"error":"invalid input"} to the changelog request and 404 at both its root and its /changelog page, a name no longer in use on Canny. The other 23 answered with a different HTTP error and no parseable body. None of the 183 failed to answer at all, so no host here is reported unreachable when it in fact responded.

Only the published date, a word count of the entry body, its labels or types, the notify flag, the linked request count on Canny, and the reaction count were kept. No title, author or email was stored. Entries are counted once each; a Featurebase entry with more than one published locale is kept once regardless.

The status update email figure in the section above is a separate measurement, read the same day on the 341 Featurebase hosts with 10 or more requests, over the 10,476 completed requests whose record of whether an email went out is readable.

The next read is due by 1 January 2027.

Related reading

Feature request votes, six hundred and thirty four public boards measured Feedback tool pricing data, thirty two vendors read on one day Feedback tool link authority, seventy domains on the web's link graph Feedback tool feature comparison: 32 vendors on 9 questions Canny board audit