Skip to content

Leadership and readiness

Time to first accepted dataset: read the curve, not the average

By SourceX Editorial · Updated

10 sections

Short answer

Time to first accepted dataset is the elapsed time from a supplier's signature to the first written acceptance of data it delivered. Report it per signing cohort as a cumulative curve, with open cases censored and declines counted as competing outcomes, never as one average. Among 228 CNCF sandbox projects, promoted ones took a median 759 days, yet the estimated share promoted never reaches half, so the typical entrant has no promotion date.

Checklist

Time to first accepted

This page covers time to first accepted dataset: read the curve, not the average.Illustrative

Key takeaways

  • Start the clock at the supplier's commitment signature and stop it at written acceptance of its first dataset; report the steps between as sub-intervals.
  • An average of accepted suppliers only describes the survivors: in the CNCF test, the 759-day median covered 21.9 percent of entrants.
  • Declines and withdrawals are competing outcomes, not censored cases; censoring them lifted the CNCF estimate from 33.7 to 37.3 percent.
  • Compare signing cohorts at the same age: CNCF's 2017-2020 entrants were promoted within two years at 34.5 percent, later cohorts at 2.8 and 1.9 percent.
  • A credible claim shows every signer in the denominator, the number still open and a median marked not reached when fewer than half are accepted.

What does time to first accepted dataset measure?

It measures how long a newly signed supplier takes to produce data that a reviewer accepts in writing: the clock starts at a dated signature and stops at the first written acceptance of that supplier's data. The unit is the supplier, not the dataset, which is why the word first matters. A supplier's first acceptance is the moment rights review, extraction, de-identification, delivery and the acceptance check have all worked once, end to end, for that company.

For an investor, that makes the interval a test of whether signed supply turns into accepted supply, and how fast. A count of signed suppliers can rise while the share that ever reaches acceptance falls; the interval, reported by cohort, shows which is happening. It is also harder to dress up than a signature count, because the stop event is a written decision by the party reviewing the data, not by the party selling it.

Every supplier must be able to land in one of three places: accepted in writing on a date, ended without acceptance on a date (declined, withdrawn or past its deadline), or still open on the report date. A metric with no room for the second and third measures the speed of the survivors, not the speed of the business.

Which start event and which acceptance should the clock use?

Start at the signature that commits the supplier and stop at written acceptance of its first dataset, then report the steps in between as sub-intervals rather than as rival headline numbers. An earlier start, such as first contact, mixes sales time into operations. A later one, such as delivery, hides the slowest part of preparing business records.

The choice moves the number more than most readers expect. The CNCF lifecycle document, which governs the public test later on this page, tells projects that moving a level 'can range from 5 months to 15 months' once an application is in. That clock starts at the application. The clock measured below starts at entry to the sandbox, and its completed cases took a median of 759 days. Both describe one review process; they simply start in different places.

SourceX's process page describes its own path in these words: apply and qualify, sign the Master Supplier Licensing Agreement once, which the page calls Day 1, sign a Dataset Deal Confirmation that fixes the exact dataset and its deadlines, prepare and deliver, then acceptance and Closing. Closing happens, the page says, when SourceX has accepted the data in writing and every condition is met, and a deal that does not close ends with a no-deal notice. That vocabulary maps onto the clock: either signature is a candidate start, written acceptance is the stop, and a no-deal notice is a competing outcome. SourceX publishes no figures for this interval, and none are given here.

The headline clock and its parts:

Which start event and which acceptance should the clock use?
ClockStarts atStops atWhat it showsWhat it hides
Headline: time to first accepted datasetSupplier's commitment signatureFirst written acceptance of its dataWhether signed supply becomes accepted supplySales time before the signature
Scoping sub-intervalCommitment signatureDataset-level confirmation signedHow long agreeing the exact dataset takesPreparation and review
Preparation sub-intervalDataset-level confirmationDelivery received by the reviewerExtraction, transfer and de-identificationReview queues
Review sub-intervalDelivery or the agreed review-start eventWritten acceptance or a declineReviewer capacity and data qualityEverything upstream
Not recommended: from first contactFirst contact or applicationFirst written acceptanceSales and operations combinedWhich of the two is slow

Why does an average time to acceptance mislead?

Because an average can only be taken over suppliers already accepted, it leaves out the two groups that decide the answer: suppliers still waiting and suppliers who will never get there. Three distortions follow, and each makes a business look faster or healthier than it is.

Dropping open cases shortens the clock. The slowest suppliers are, by definition, the ones still open, and a fast-growing company has a large share of recent, open cohorts, so growth itself flatters an average of completed cases. Open suppliers belong in the estimate as censored observations: counted as at risk until the report date, then removed without being called successes or failures.

Censoring declines overstates success. The competing-risks vignette of the R survival package, by Terry Therneau and colleagues, calls it 'a common mistake' to run a Kaplan-Meier estimate on one event type while treating the other event types as censored, and notes that the incorrect curve comes out higher. A declined dataset is not a supplier who might still be accepted; it is an ending. The right estimate is the cumulative incidence of acceptance, computed with the Aalen-Johansen method, which counts declines as a competing outcome.

Pooling cohorts lets the launch cohort speak for everyone. Early suppliers are often chosen by hand, and a pooled curve reflects them for years. The report that avoids all three is a curve per signing cohort: at each checkpoint, the share accepted, the share ended without acceptance, the share still open and the number still at risk.

What a public review clock shows: 228 CNCF sandbox entrants

The Cloud Native Computing Foundation runs a public, written review process that behaves like this clock, and its landscape file records the dates. Its lifecycle document defines four stages: sandbox, incubation, graduated and archived. Projects may enter as sandbox projects or directly at incubation, and a move up a level passes when two thirds of the Technical Oversight Committee votes for it after due diligence and two weeks of public comment. Archiving also takes a two-thirds TOC vote. Entry plays the part of the signature, the incubation vote the part of written acceptance, archiving the part of a decline.

In the landscape file at commit c239ac1 (27 October 2026), 259 projects carry a CNCF stage. Thirty entered directly at incubation and one has no acceptance date, leaving 228 that entered below incubation. Of those, 50 had been promoted to incubation, 25 were archived before any promotion and 153 were still in the sandbox on the data date.

The 50 promotions took a median of 759 days and a mean of 856. Both figures are correct and both mislead, because they describe 21.9 percent of entrants. Counting all 228, with open projects censored and archiving as a competing outcome, an estimated 11.5 percent were promoted within 730 days and 24.9 percent within 1,460 days. The estimate reaches 33.7 percent at day 1,936, the last promotion observed, and stays there through the longest follow-up of 3,087 days. It never reaches half, so the median time to promotion for an entrant is not reached: the typical sandbox project has no promotion date at all.

Treating archived projects as censored, the mistake the survival vignette warns about, lifts the estimate to 25.6 percent at 1,460 days and 37.3 percent at the end of follow-up. That curve describes a foundation in which no project is ever archived.

Cohorts tell the rest. Of 58 projects that entered from 2017 through 2020, 20 (34.5 percent) were promoted within two years. Of 72 entering in 2021 and 2022, 2 were (2.8 percent), and of 52 entering in 2023 and 2024, 1 was (1.9 percent). For the 46 entrants of 2025 and 2026 the two-year window has not closed, so no share is reported. The 2021-2022 cohort now counts more archived projects (18) than promoted ones (12). A pooled curve is carried by the first cohorts; the cohort view shows the later ones moving far more slowly.

Step curves for 228 CNCF sandbox entrants showing the estimated share promoted to incubation rising to 33.7 percent and the share archived to 20.8 percent, never reaching the 50 percent median line, beside bars of two-year promotion shares by entry cohort: 34.5, 2.8 and 1.9 percent.
CNCF landscape at commit c239ac1, 27 October 2026; cumulative incidence with archiving as a competing outcome and open projects censored. Computed by SourceX.

The cohort dashboard: one row per signing cohort

Lay the metric out as one row per signing cohort, with the share accepted in writing, the share ended without acceptance and the share still open at fixed checkpoints, and write window open until every member of a cohort has been observed that long. The example is invented: a hypothetical licensor that signs 6, 10, 14 and 20 suppliers in four quarters, each cohort signing on the quarter's first day, read at the end of the year. It describes no real company, SourceX included.

The accepted-only view reports a mean of 12.6 weeks and a median of 12 weeks to first acceptance, and 9 of the 14 accepted suppliers (64.3 percent) cleared it within 13 weeks. The cohort view disagrees. The share of each cohort accepted by week 13 fell from 66.7 percent to 20.0, 14.3 and 5.0 percent while quarterly signings more than tripled, and across all 50 signed suppliers only 9 (18.0 percent) had an accepted dataset by week 13.

Each cohort is compared with the one before at each checkpoint and given one of four labels: faster, slower or same, by comparing exact accepted shares, or not comparable when either window is still open. A claim that the clock is improving needs a faster label on the latest comparison with both windows closed. Here every closed comparison reads slower. Every figure and label in the table was computed by script from the invented inputs:

The cohort dashboard: one row per signing cohort
Signing cohort (Illustrative)Accepted in writing by week 13Declined and open at week 13Accepted by week 26Label against the previous cohort
Q1: 6 signed, observed 52 weeks4 (66.7%)1 declined, 1 open5 (83.3%); median week 9First cohort; no comparison
Q2: 10 signed, observed 39 weeks2 (20.0%)1 declined, 7 open5 (50.0%); median week 22Slower at week 13 and week 26
Q3: 14 signed, observed 26 weeks2 (14.3%)1 declined, 11 open3 (21.4%); median not reachedSlower at week 13 and week 26
Q4: 20 signed, observed 13 weeks1 (5.0%)0 declined, 19 openWindow openSlower at week 13; not comparable at week 26
All 50, accepted-only view9 of 14 accepted suppliers (64.3%)Declined and open suppliers droppedMean 12.6 weeks, median 12 weeksNo label: not a cohort measure

Seven checks before a time-to-acceptance claim enters a memo

Ask for the evidence behind each part of the claim, not for a better chart. These checks work for any licensor, and each can be answered from records a company already keeps:

At a glance

Seven checks before a time-to-acceptance claim enters a memo

  • Definitions: the start and stop events in writing, the date each last changed, and whether earlier cohorts were restated.
  • Acceptance evidence: for a sample of suppliers, the acceptance notice issued by the reviewing party, not an internal status field.
  • Denominator: every supplier who signed in the cohort, including those declined, withdrawn or still open.
  • Report date and censoring: how many suppliers in each cohort are still open and how long each has been observed.
  • Competing outcomes: declines and withdrawals counted as endings in a cumulative incidence estimate, never censored.
  • Same-age comparison: cohorts compared at fixed checkpoints, with the launch cohort shown on its own row.
  • Tail and median: percentiles beside the median, and the median marked not reached when fewer than half are accepted.

Could a fast clock still mislead?

Yes. A short interval can come from easier suppliers, looser acceptance or a changed definition rather than from better operations, and each has a test. A launch spike shows up when the first cohort beats every later one at the same age, as CNCF's 2017-2020 entrants did. A zero base shows up in counts: CNCF's 2017 cohort had three entrants, two of them promoted within two years, a share no one should quote alone.

A changed definition shows up in the clock table. Moving the start from first contact to signature, or the stop from written acceptance to delivery receipt, shortens the interval without anything getting faster; the CNCF document's 5-to-15-month range and the 759-day median above are the same process timed from different starts. Looser acceptance shows up as a falling decline share alongside a rising complaint or rework rate, so ask for the acceptance criteria version used in each cohort.

Reviewer capacity is the last alternative. The CNCF document says public comment and voting do not open within four weeks of its main KubeCon events in Europe and North America, so a reviewer's calendar adds time no applicant controls. A data licensor's review queue, or a buyer's, behaves the same way, which is why the review sub-interval should be reported separately.

What could make this metric, or the thesis behind it, wrong

Several things, and an investor should weigh each before treating a short or improving clock as evidence of a durable business:

  • Acceptance is not revenue. On SourceX's process page the supplier invoices only after Closing, and a licensor's own sale to a buyer is a separate event, so a fast clock shows that supply can be prepared and accepted, not that it can be sold, priced or collected.
  • Small cohorts swing. A young licensor's cohorts may hold a handful of suppliers, so one decline moves a share by double digits; counts must sit beside every percentage.
  • One queue can dominate. If one reviewer or one buyer's process drives most acceptances, the clock measures that queue rather than the company.
  • The analogy has limits. CNCF projects are open-source communities whose promotion depends on adoption, governance and volunteer reviewers, so the public test demonstrates the measurement, not the speed of data licensing.
  • The start can be moved. A company that signs suppliers only once their data is ready starts the clock late, which is why the scoping and preparation sub-intervals belong in the same report.

Questions to ask any licensor about the clock, SourceX included

Ask questions that can only be answered with dated records. SourceX is raising capital from accredited investors under Rule 506(c) and says it shares company and commercial metrics with investors directly; investor materials can be requested at sourcex.si/investors, which is also where these questions can be put to it. The same list applies to any licensor:

Checklist · 6 items

  1. Which signature starts your clock, which written acceptance stops it, and has either changed since your first cohort?
  2. For each signing cohort, how many suppliers are accepted, ended without acceptance and still open at each checkpoint?
  3. Do you estimate acceptance with declines as a competing outcome, or do you censor them?
  4. Where the median is not reached, what share of each cohort was accepted by the last checkpoint?
  5. How long do the scoping, preparation and review sub-intervals take, separately, for the latest closed cohort?
  6. Can you show the acceptance notices for a random sample of suppliers from one cohort?

0 of 6 ticked. Tick each item as you go.

Methods note: the CNCF count, the illustrative cohorts and the limits

Public test: landscape.yml from github.com/cncf/landscape at commit c239ac1 (committed 27 October 2026), read by a script on 28 October 2026 without running any repository code. A project is an item with a stage field; its dates are the file's accepted, incubating and archived values. An entrant has an accepted date and either no incubating date or a later one; projects whose incubating date equals their accepted date entered directly at incubation and were excluded (30), and one item without an accepted date was dropped. Each entrant's outcome is promotion on its incubating date if present, otherwise archiving on its archived date if present, otherwise open and censored on 27 October 2026. Elapsed time is whole calendar days between dates.

Estimates: the promoted-only median and mean use the 50 promoted entrants, with the median of an even count taken as the mean of the two middle values and rounded half up. Cumulative incidence is the Aalen-Johansen estimate with archiving as a competing outcome; the comparison curve is one minus a Kaplan-Meier estimate that censors archived projects. At tied days, events are processed before censoring. Percentages are exact fractions rounded once, half up, to one decimal. A two-year cohort window is closed when the cohort's last acceptance date plus 730 days falls on or before the data date. The CNCF stage definitions and voting rules come from the TOC's lifecycle and archiving documents at commit 95230b2 (26 October 2026), and the competing-risk point from the survival package vignette at commit b76de62.

Illustrative dashboard: invented inputs listed in full in the script, with no randomness; every share, median and label is computed by it. Not measured: any SourceX interval, cohort or rate, none of which is public; the time CNCF projects spent before applying; whether landscape dates mark votes or announcements; and how CNCF timing compares with any commercial review process.

Frequently asked questions

What is time to first accepted dataset?

It is the elapsed time from the signature that commits a data supplier to the first written acceptance of data that supplier delivered. It is measured per supplier and reported by signing cohort, so each supplier ends up accepted, ended without acceptance or still open on the report date.

Why is an average a poor summary of this clock?

An average can only include suppliers already accepted, so it leaves out the slow ones still waiting and the ones that will never be accepted. In a public test on 228 CNCF sandbox projects, the median for promoted projects was 759 days, yet the estimated share ever promoted stayed below half.

What does censoring mean for this metric?

A supplier still open on the report date is censored: it counts as at risk until that date and is then removed without being called a success or a failure. Dropping open suppliers instead makes the clock look shorter, especially for a fast-growing company with many recent cohorts.

Should a declined dataset be treated as censored?

No. A decline or withdrawal is an ending, so it should be a competing outcome in a cumulative incidence estimate. Censoring it estimates a world where no deal fails, and the survival package's documentation calls that a common mistake that produces a higher curve.

Does SourceX publish its time to first accepted dataset?

No. Its process page describes the steps from signature to written acceptance and Closing but gives no interval figures, and none appear in this article. SourceX says it shares company and commercial metrics with investors directly, through the materials request at sourcex.si/investors.

Sources

  • The CNCF landscape file records accepted, incubating and archived dates for 259 projects with a CNCF stage, the data behind the 228-entrant count. Source
  • The CNCF lifecycle document defines four stages, lets projects enter as sandbox or directly at incubation, requires a two-thirds TOC vote to move levels, pauses comment and voting near KubeCon and gives a 5-to-15-month range once an application is in. Source
  • Archiving a CNCF project requires a TOC vote with a two-thirds majority; archived projects are inactive or no longer recommended. Source
  • The survival package vignette calls Kaplan-Meier on one event type with other events censored a common mistake that yields a higher curve, and uses the Aalen-Johansen estimate for competing risks. Source
  • SourceX's process page lists its steps from qualification and signature to written acceptance and Closing, with a no-deal notice when a deal does not close and invoicing after Closing. Source
  • SourceX's investor page says it shares company and commercial metrics with investors directly and offers an investor materials request. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies