Methodology
How we track speculative dates
Most countdowns on WorldClockTools point at a date a vendor or authority has officially announced. A growing share, however, point at events whose date is still speculation — the next Apple keynote, an unannounced Anthropic model, a US government report with a statutory deadline that may or may not slip. This page documents how we treat those uncertain dates, end-to-end.
Updated 5 November 2026: earlier versions of this page described an automatic signal-ingestion pipeline, a quarterly-reviewed whitelist of tracked accounts with admission thresholds, and a liquidity filter for prediction markets. The site does not have those. The page now describes what the site actually does.
Methodology rules at a glance
- Source signals are tier-weighted: primary sources count 3×, trade press 2×, community signals 1×. Methodology — credibility tiers
- Signals are fused with a weighted median, not a weighted mean, so a single outlier cannot drag the estimate. Methodology — weighted-median fusion
- Each signal's weight falls linearly to zero 90 days after the signal's date, measured from when the page is generated; if no logged signal is newer than 90 days, no aggregate estimate is shown. Methodology — 90-day decay
- Confidence labels (confirmed, expected, rumored, estimated) are set by hand; "confirmed" is meant only for dates the organiser itself has announced. WorldClockTools editorial source rules
- Signals are logged by hand during editorial reviews, not collected automatically; prediction-market prices are shown for context but are not part of the calculation. Methodology — how signals enter the log
Why some pages show an estimated window
When a date is officially confirmed, the countdown shows a precise target. When it isn't, showing a single sharp date would be misleading — so the UI degrades gracefully: instead of a ticking clock you see an estimated window, an explicit confidence label, and a list of the signals that produced the estimate. We treat the absence of certainty as information the reader should see, not as something to hide.
This affects roughly the rumour-tracker family of pages: tech keynotes ahead of vendor invites, AI model launches, generative hardware rumours, and political events with a soft deadline. Confirmed events (FIFA World Cup kick-off, an SEC rule's effective date, Diwali) never use the speculative UI.
The credibility-tier system
Every signal we log is assigned to one of three tiers, each with a fixed numeric weight. Rumour-tracker pages show the weights next to the estimate so readers can see how it was produced.
- Primary (×3): the vendor itself, a government primary source, or a court filing. Examples: an Apple newsroom invite, a NARA disclosure schedule, a Federal Reserve FOMC calendar entry.
- Trade (×2): established trade press with named bylines and visible editorial standards. Examples: Bloomberg, Reuters, The Information, Financial Times. A single trade report counts as one signal; it never makes a date “confirmed”.
- Community (×1): named accounts and forums that regularly report on the topic. Each rumour-tracker page lists the accounts we follow for it, so readers can judge them. Anonymous claims with no track record are not logged.
Weighted-median fusion + 90-day linear decay
Once signals are tier-weighted we fuse them with a weighted median, not a weighted mean. Median is robust to outliers — one wildly wrong account doesn't drag the estimate. Each signal's effective weight also decays linearly over the 90 days following its publication, so a four-month-old report no longer counts at all.
The formula:
effectiveWeight(s) = tierWeight(s) × max(0, 1 - ageDays(s) / 90) estimate = weightedMedian(date(s), effectiveWeight(s))
In practice this means a fresh primary signal (×3, age 0) has an effective weight of 3.0; the same signal at 45 days old has an effective weight of 1.5; at 90+ days it has weight 0 and drops out entirely. The same 90-day window applies to every tier. Because age is measured from when the page is generated, an event whose signal log has not been updated for 90 days shows no aggregate estimate until it is reviewed again.
How signals enter the log
Signals are added by hand when we review an event: the date of the report, its source and URL, its tier, and the date it implies. There is no automatic collection, no fixed review schedule, and no formal scoring of tracked accounts' past accuracy yet. Each rumour-tracker page shows when its log was last updated.
Where a well-defined prediction market exists for an event, the countdown page shows its current price, fetched from Polymarket or Kalshi when the page is generated. These prices are displayed alongside the estimate for context; they are not part of the weighted-median calculation, and we do not apply a liquidity filter, so treat thinly traded markets with care.
Each logged signal is stored with its source URL, date and tier, and is listed on the event's page. For the catalogue-level source rules, see Sources.
Editorial responsibility
The line between "rumoured" and "confirmed" is an editorial label, set by hand. A countdown should get confidence: "confirmed" only when the source is the organiser's own announcement on the record (vendor newsroom, official press release, court filing). Events without a confirmed date are tagged speculative or a similar lower-confidence label and, where an estimated window exists, rendered with the estimated-window UI and a "not officially confirmed" label. Some older entries still carry only a general link to the organiser's website as their source; we are reviewing these.
When an organiser confirms, moves or cancels a date, we update the event's date and label at the next review. Because reviews are manual, there can be a delay; the date on the organiser's own announcement always takes precedence over ours.
We do not accept paid placement or sponsored signals. The tracker is editorial.