Votes on 634 public Canny and Featurebase boards, 67,473 requests in all, read on 3 October 2026. Each board is one row in the file below, with its post count, its vote total and the shape of its votes, and no company, board or host is named.
The median request holds 2 votes, and one of them is the vote of the person who posted it. Four requests in ten have no vote beyond that one. Half the requests on the median board carry a decision, 43 percent of boards have never declined one in public, and half the boards have gone 90 days without a new request. The median request still waiting for a decision was posted 592 days before the read.
id,vendor,complete,posts,votes,median_votes,share_le_2,top_post_share,top_tenth_share,decided_share,denied_share,days_since_newest,median_waiting_days
1,featurebase,1,76,121,1.0,84.2,5.8,24.0,77.6,22.4,415,742
2,featurebase,1,61,107,1.0,77.0,4.7,21.5,75.4,0.0,181,315
3,featurebase,1,10,19,1.0,80.0,36.8,36.8,20.0,0.0,32,410.5
Download the file. It is free to copy, quote and use commercially. To cite it, link https://votefirst.app/en/data/feature-request-board-votes and give the read date, 3 October 2026.
How many votes does a typical feature request get?
Two. That is the median across all 67,473 requests, and it is also the median of each board's own median, so it holds whether the requests are pooled or each board counts once.
Almost no request sits at zero: 2.7 percent on Canny and 1.3 percent on Featurebase, against 39.0 and 42.4 percent at exactly one. That is the shape of a count that includes the poster's own vote. Across both, 2.0 percent of requests carry no vote, 40.5 percent carry exactly one, and 59.8 percent carry two or fewer. So the typical request has been backed by one person besides the one who asked. On ten Discourse forums with topic voting, where a request starts at zero, the median request has one vote, the same backing.
The mean is 7.9, pulled up by a few requests with many votes. 12.3 percent of requests reach 10 votes, and 1.0 percent reach 100.
How concentrated are votes on one board?
On the median board, the most voted request holds 8.8 percent of every vote the board has collected. On a quarter of boards it holds more than 14 percent.
The top tenth of requests by votes holds 29.6 percent of a board's votes at the median, and more than 40 percent on a quarter of boards. A board's votes gather on a short list of requests, and most of the rest sit within a vote or two of each other.
How often do companies decide in public, and how often do they say no?
On the median board, 50.0 percent of requests carry a status other than the one a new request starts in: under review, planned, in progress, completed or declined. Canny lets each company name its own statuses and file each under one of four types, and the type is what is counted here. That is how often a company marks a decision where visitors can see it.
The spread is wide. A quarter of boards have marked fewer than 27.5 percent of their requests, and a quarter three in four or more.
Declining is rarer. On the median board, 1.4 percent of requests are marked closed or canceled, each vendor's word for no. 275 of the 634 boards, 43 percent, have never declined a single request in public, and 359 have declined at least one. The habit differs by vendor: 195 of 329 Featurebase boards have never declined a request, against 80 of 305 Canny boards. A reply for a decline, and for each other status, shows what the answer can say.
How long does a feature request wait for an answer?
Usually more than a year. 34,681 requests still sat in the status a new request starts in, and the median one was posted 592 days before the read. 67.1 percent of them were posted more than a year before, and 42.8 percent more than two years before.
Board by board, the picture holds. On the 528 boards with five or more requests waiting, the typical waiting request is 592 days old at the median board, and more than a year old on 72.5 percent of boards. A quarter of boards sit under 344 days, and a quarter over 1,054.
Most of the answering happens early. 71.3 percent of the requests posted in the 90 days before the read still wait, against 49.0 percent of those posted more than a year before. After that the share barely moves. Grouped by the year they were posted, close to half of every year's requests from 2019 to 2025 still wait:
| Posted in | Requests | Still waiting | Picked up | Completed | Declined |
|---|---|---|---|---|---|
| 2018 | 1,377 | 31.6% | 1.0% | 34.0% | 33.4% |
| 2019 | 1,379 | 48.3% | 2.5% | 34.9% | 14.3% |
| 2020 | 2,257 | 51.1% | 3.1% | 33.5% | 12.4% |
| 2021 | 2,980 | 54.9% | 6.6% | 28.7% | 9.8% |
| 2022 | 5,092 | 52.0% | 9.9% | 28.6% | 9.5% |
| 2023 | 7,919 | 51.4% | 8.2% | 30.3% | 10.0% |
| 2024 | 12,229 | 45.7% | 10.9% | 33.5% | 9.8% |
| 2025 | 18,178 | 51.9% | 9.6% | 32.6% | 5.9% |
| 2026 | 13,864 | 64.8% | 9.9% | 21.5% | 3.9% |
Picked up means under review, planned or in progress. 2018 is the exception, with a third of its requests declined. The 281 requests from 2017 and before are too few to show, and the 1,917 with no created date are left out of every age figure.
Do more votes get a request answered?
Less than a ranked board suggests. A request with three to nine votes carries a decision as often as one with two or fewer, 46.8 percent each. From ten votes the share rises to 55.6 percent, and from fifty to 64.1.
| Votes | Requests | Carry a decision | Completed | Declined |
|---|---|---|---|---|
| 0 to 2 | 40,355 | 46.8% | 29.0% | 10.0% |
| 3 to 9 | 18,792 | 46.8% | 28.8% | 7.2% |
| 10 to 49 | 6,883 | 55.6% | 34.0% | 6.3% |
| 50 or more | 1,443 | 64.1% | 38.9% | 3.6% |
Declines run the other way, from 10.0 percent at two votes or fewer to 3.6 at fifty or more. Votes and status were read on the same day, and a completed request can keep drawing votes, so this shows how the two go together rather than which one caused the other.
Atlassian's public tracker shows the same shape at a larger scale. Of 179,402 suggestions on jira.atlassian.com, 46.0 percent of those with one to four votes were closed unbuilt, against 3.5 percent of those with a hundred or more. On JetBrains' tracker, 149,466 feature requests, the share built more than 30 days after filing climbs from 24.1 percent at one to four votes to 59.5 percent past a hundred.
How many public boards have gone quiet?
Of the 621 boards with at least one dated request, 313 have had no new request in the last 90 days. That is 50 percent.
Some are mature boards with the easy requests already shipped. A visitor who does not check the dates cannot tell those apart from abandoned ones.
How big is a public board?
The median board holds 54 requests and 163 votes. A quarter hold 24 requests or fewer, a quarter more than 132, and 63 boards, one in ten, more than 275. The 68 boards left out for size are larger than all of these.
Counted by host, with every board a company runs on one address added together, 534 of the 648 hosts with a public board held 250 public requests or fewer, 82 percent. Fider's free hosted plan draws its fair use line at 250 feedback items.
What does that mean for deciding what to build?
When the median request has one vote besides its poster's and most requests sit within two votes of each other, a vote count alone separates very little. It ranks the short list at the top and says almost nothing about the rest.
VoteFirst scores demand from votes, comments and views together: votes count three, comments two and views a tenth, so a request with two votes and twenty comments scores 46, above one with fifteen votes and no comments at 45. It can weight each vote by the voter's tier, a whole number multiplier per tier, so a vote from an account on a higher tier counts several times over. And because 43 percent of the boards here have never declined anything in public, VoteFirst's public roadmap keeps a Denied column visible by default. When the median waiting request is 592 days old, telling voters is half the job, so VoteFirst emails the signed in voters whose vote is in a release, naming what shipped.
The board audit reads one Canny or Featurebase board by its address and shows its requests, its votes, its top request and how long ago the newest one arrived, to set against the figures here.
How was this built, and when does it change?
The candidate list is every *.canny.io and *.featurebase.app address in Common Crawl's URL index, reduced to one row per subdomain, with the two vendors' own marketing, app, docs and status hosts removed. That gave 902 hosts. A list drawn from a crawl of the public web leans toward boards that something links to, so this measures the public, linked part of the category and is not a census of every board either vendor hosts.
Of the 902 hosts, 254 returned nothing readable: a block, a login wall, a private board or no public board. The other 648 carried 919 public boards. 217 of those had fewer than 10 requests and are left out of every figure, though their rows stay in the raw file. Of the 702 boards with 10 or more, 68 held more requests than the reader takes, 500 per Canny board and 1,000 per Featurebase host, so only their most voted requests were read. Their medians, shares and newest dates would be biased by that, so they are left out of every figure as well, and appear in the download with complete set to 0. Every figure on this page describes the remaining 634 boards, which makes them boards of up to that size.
Each board was read through the same public endpoints VoteFirst's board importer reads, one request a second, with no sign in. Canny's endpoint returns completed and closed requests only when the request names every status, so the reader names them all. Canny's closed type also holds a few requests a company sent to support or marked already live, and they count with the declines. Only the vote count, the status and the created date of each request were kept. A request's age is the days from its created date to the read date. No title, author, email or text was stored. The public data carries no field for whether a company replied to a request, so replies are not measured.
The same day, justRead's public roadmap, which runs on VoteFirst and is made by VoteFirst's makers, held 131 cards and 884 votes, a median of 2 and a top card holding 5.3 percent of the votes. VoteFirst does not add a poster's vote, so its 2 is two people. 90 percent of its cards carry a decision, against the 50.0 percent median above, because most are marked completed or denied.
The next read is due by 1 January 2027.
Who is this dataset for?
Anybody running a feature request board who wants to know whether its vote counts, its quiet stretches or its habit of never declining anything is normal for the category.
The category's prices are published as a separate file, and the vendors are ranked on the web's link graph, for the commercial picture alongside the votes.
14 days, a card at signup, then $9 or $39 a month.