Munim Technologies

Research/Software

How often W3C drafts become Recommendations

Fewer than half of the web standards W3C started in 2000-2014 became Recommendations within ten years. W3C now publishes about half as many Recommendations a year as in 2015-2019, but our prediction that the 2020 Process slowed specs down failed.

Oct 30, 2026 · 6 min read

Web standards such as CSS, HTML APIs and WebRTC are written at the W3C, where a specification starts as a Working Draft and is meant to end as a Recommendation. We followed every specification in W3C's public API through that process. Of the specs on the Recommendation track first drafted in 2000-2014, 44.8% became a Recommendation within ten years and 30.8% ended as a Note or Retired. W3C now publishes about half as many new Recommendations a year as it did in 2015-2019. We also predicted that the 2020 W3C Process would visibly slow the first four years of a spec's life. That prediction failed.

0.448

of 496 rec-track specs first drafted 2000-2014 reached Recommendation within 10 years

7.0 vs 15.2

new Recommendations per year, 2021-2025 against 2015-2019

14.2 : 1

Candidate Recommendation Drafts to Snapshots since 2021 (1,519 to 107)

65

specs in CR for 3+ years that have never been a Recommendation

The question

How often does a W3C Working Draft become a Recommendation, how often is it closed instead, and did the 2020 Process change that? The 2020 Process introduced the Candidate Recommendation Draft, a version that can be republished freely without the review step that a CR Snapshot requires. Some in the community call the resulting pattern "CR forever". W3C dashboards list stale drafts, and a 2026 hackathon project visualised time to Recommendation, but we found no study that measured outcome shares with censoring and compared the eras.

What we did

We downloaded the version history of all 1,721 specifications listed by the public W3C API on 2026-10-30, 17,756 versions with no failed fetches. After dropping six duplicate specs, 1,165 started as a Working Draft, and 912 of these are on the Recommendation track. A Recommendation and a Note or Retired document are competing endings; a spec with neither is censored, not counted as ended. Six predictions were published in a pre-registration before any lifecycle data were counted, and a second pre-registered follow-up measured time spent in CR.

What we found

Five of the six predictions held. Fewer than half of the 2000-2014 starters became Recommendations within ten years, and about three in ten were closed. New Recommendations fell to 7.0 a year in 2021-2025 from 15.2 in 2015-2019. Since 2021, Candidate Recommendation Drafts outnumber Snapshots 1,519 to 107.

New W3C Recommendations per year

01020200020052010201520202025YearNew Recommendations
2026 (4 so far at the 30 October snapshot) is left out as a partial year. Source: w3c-spec-lifecycle/results/run1-summary.json (metrics.P5.rec_by_year)

The prediction that failed

We expected that, four years after the first draft, specs started in 2017-2021 would be more than five points less likely to be a Recommendation, and more than five points more likely to be stuck in CR, than specs started in 2010-2014. Neither shift happened. In the main cohort the Recommendation share went from 25.0% to 24.3% and the CR share from 13.4% to 14.7%. A broader cohort moved in the predicted direction for Recommendations but the wrong direction for CR. Counting Note-track drafts as endings did not change the verdict.

Status four years after the first draft

CohortFirst draftsnAt RecommendationIn CR
A (rec-track)2010-201422425.0%13.4%
A (rec-track)2017-202113624.3%14.7%
B (broader)2010-201427320.5%11.0%
B (broader)2017-202119616.8%10.2%
Pre-registered prediction P3 required both shifts to exceed 5 points in both cohorts. Source: w3c-spec-lifecycle/results/run1-summary.json (metrics.P3).

So where did the drop in Recommendations come from? Exploratory counts, made after the run, point to fewer specs entering the pipeline. The main cohort had 224 first drafts in 2010-2014 and 136 in 2017-2021. Fewer specs are closed, too: 27.7% of the earlier starters had ended as a Note or Retired after four years, against 6.6% of the later ones.

Time in Candidate Recommendation

The follow-up started the clock at a spec's first CR. Specs that first entered CR under the 2020 Process reached Recommendation within three years half the time (16 of 32), against 67.5% (83 of 123) for 2011-2015 entrants. That passed the pre-registered bar of a 0.10 drop, but the new group is only 32 specs and a two-proportion test, not pre-registered, gives p of about 0.07. We read it as suggestive, not settled.

Share reaching Recommendation after first entering CR

First CRnWithin 1 yearWithin 2 yearsWithin 3 yearsEnded as Note/Retired within 3 years
2011-201512338.2%54.5%67.5%8.9%
Sep 2020 - Sep 20233225.0%43.8%50.0%0.0%
Source: w3c-spec-lifecycle/results/followup1-summary.json (metrics.G3).

Meanwhile, CR has become a place where specs are republished, not finished. CR publications never exceeded 69 a year before 2020; in 2024 there were 376, almost all CR Drafts. Sixty-five specs have been in CR for three years or more without ever being a Recommendation, including CSS Flexbox, CSS Grid, Service Workers, Media Capture and Streams, Resource Timing and WebXR.

Candidate Recommendation publications per year

CR SnapshotsCR Drafts (introduced 2020)
010020030020162018202020222024YearPublications
2026 is a partial year and is left out (400 Drafts and 14 Snapshots by 30 October). Source: w3c-spec-lifecycle/results/followup1-summary.json (metrics.cr_publications_by_year)

What it means

The 2020 Process did not make the first four years of a spec's life look different. What changed is how many specs start, how few are closed, and how many sit in CR as living documents. Developers already treat Flexbox and Grid as standards, so for them the label may matter little. For anyone who relies on the Recommendation stamp, such as procurement rules or patent commitments, the stamp now arrives for fewer specs and later.

More research