Methods before metrics
How WaitGraph measures a wait
A useful waiting-time answer starts with an exact process, keeps different evidence classes separate, and remains honest about cases that have not finished.
01 · Define the interval
Every process needs a start and an end
WaitGraph measures elapsed time between a named start event and a named first outcome. Similar labels are not automatically comparable: account setup, identity checks, appeals, site review, and time until earnings can be separate processes.
02 · Label the source
Four evidence classes, never blended
An answer belongs to exactly one class. Text labels remain visible beside every result.
| Class | Meaning | Example label |
|---|---|---|
| Official estimate | Guidance published by the organization or another authoritative source. | Google’s published timeframe |
| AI-sourced general answer | A current answer synthesized from cited public sources; not WaitGraph data. | Published guidance — no WaitGraph estimate yet |
| Documented pattern | Real observations exist, but volume or contributor diversity is insufficient for a population estimate. | Repeated observed cycles from one contributor |
| WaitGraph estimate | Enough recent, comparable tracked cases exist for a defensible statistical estimate. | Based on recent cases from distinct contributors |
03 · Keep unfinished waits
Still waiting is data, not a missing row
An active wait is right-censored: it has lasted at least this long, but its final duration is unknown. Calculating only from people who already finished would bias the answer downward.
WaitGraph preserves prospective, late-active, and retrospective registration modes. As queues mature, time-to-event analysis uses active cases, day-level precision where needed, contributor clustering, cohort boundaries, recency windows, and process-version breaks.
04 · Earn publication
Queue gates
Counts are minimum gates, not automatic publication rules. Every state change requires an auditable reason and editorial review.
| State | Default gate | Public behavior |
|---|---|---|
| Requested | Unsupported demand only | No indexable page by default |
| Forming | 3 legitimate active trackers, or 10 distinct demand events from 5 likely-distinct sessions in 30 days | Guidance and collection status; no estimate |
| Documented pattern | Repeated useful evidence but insufficient contributor diversity | Exact observation with a prominent limitation |
| Early | At least 5 accepted completions from at least 3 contributors, including prospective or system-tracked cases | Counts and distribution cautiously; no live claim |
| Preliminary | At least 10 comparable completions from at least 5 contributors | Cautious range after manual review |
| Live | At least 30 recent comparable completions from at least 10 contributors; at least half prospective or evidence-backed | Current estimate with uncertainty after editorial review |
| High confidence | At least 100 recent completions from at least 40 contributors plus effective-sample, evidence, and freshness gates | Stronger confidence label |
| Stale | Freshness or process-version gate fails | Keep history; withdraw the current estimate |
| Archived | The process ended or materially changed | Preserve history and link to a replacement |
05 · Show the denominator
Statistics and uncertainty
Eligible live queues can report median time, interquartile range, milestone waiting shares, active censored count, raw cases, distinct contributors, evidence composition, analysis cutoff, and freshness. Percentiles appear only when the data can estimate them responsibly.
Evidence tiers may support sensitivity checks, but weighting cannot repair contributor concentration. Credible long waits are not deleted merely for looking unusual. A case is excluded only for a documented reason such as duplication, wrong process, invalid endpoints, or manipulation.
Corrections are part of the method
Sources and process boundaries change.
WaitGraph records source-check dates, evidence reviews, methodology versions, and correction events. A correction should leave an audit trail instead of silently rewriting history.
Review correction policy