Time-to-Event Explorer: test evidence Experimental

TTE-CURV-001 The curves render as step functions with stepped: 'after' — the value holds until the next event, the only correct KM interpolation — carrying the product-limit values per group.

TTE-CURV-005 A curve whose largest observed time is a censoring extends flat to that time and stops — the flat tail is the absence of information, not evidence of safety, and the guide says so.

TTE-STAT-001 One estimator pass per group feeds everything shown: the curve, the band, the censor marks, and every number in the strip table come from the same km.js estimate — there is no second derivation to disagree with the first.

TTE-DERIV-001 Each participant's observation is their earliest qualifying event day, ties broken by input order; later events for the same participant never displace the first.

TTE-CURV-002 Censoring tick marks sit on the curve at each censored time, one per distinct time with the survival level in force, following their group's color and visibility.

TTE-DERIV-002 An event-free participant censors at the population follow-up-end day, carrying the population censoring description into the censor-mark tooltip; without a usable follow-up day the participant is excluded with a counted reason, never silently.

TTE-CURV-003 The pointwise band is drawn as exact step rectangles per inter-event interval from the same estimate, at low opacity in the group color, and visibly gaps where the interval is undefined rather than extrapolating.

TTE-STAT-004 The estimator is cross-validated against R survival::survfit on a frozen ADTTE-shaped fixture: for every endpoint × arm, times, risk sets, event counts, survival and both bounds agree to 1e-10 — including agreement on where the interval is undefined.

TTE-ANIM-001 The chart renders without intro animation: transition frames would pair an interpolated line with a band and risk table already at final geometry — transient states that are not estimates of anything. Every render, including each live filter change, shows one truthful frame.

TTE-USER-003 A CI toggle shows or hides the pointwise band without touching the curves or the strip table.

TTE-CURV-004 The y-axis always names the estimator — "Cumulative incidence (1 − KM)" or "Event-free probability (KM)" — never a bare percentage, in both orientations.

TTE-USER-002 An orientation control flips between cumulative incidence and event-free probability; the curves, band and axis title flip together.

TTE-RISK-001 The strip table shows number at risk and cumulative events per group beneath the time axis — the ST&F guide's mandated furniture — with row labels naming each group in its color.

TTE-RISK-002 Every strip number is centered on its tick's live pixel position via the chart's own scale, over ticks generated by the module — aligned by construction, never by a parallel layout.

TTE-RISK-003 Hiding a group via the legend hides its curve, censor marks, band and strip rows together; the remaining display stays consistent.

TTE-FILT-001 Multiselect filters over the event dataset compose the endpoint live: an event qualifies when, for every filter with an active selection, its value is in the selected set — the whole population stays the denominator, and participants whose events no longer qualify censor at follow-up end. No endpoint list is hard-coded.

TTE-STAT-006 An entirely-censored group stays at S = 1 with censor marks and no confidence band — true, and visibly different from "no data"; an empty group returns an empty estimate rather than throwing.

TTE-FILT-002 A filter with no active selection qualifies every event; an empty selection qualifies none and draws the honest all-censored display rather than failing. Which columns get a filter is configurable (event_filters), with absent columns dropped with a console warning.

TTE-FILT-003 The notes name the composed endpoint (endpoint_label) and the active filter selection — "All recorded events qualify" or the per-filter selected counts — so the display always states what qualifies an event right now.

TTE-USER-004 Configured single-select population filters constrain the denominator before derivation; a filter whose column is absent from the population is dropped with a console warning, per the house pattern.

TTE-DATA-002 Event rows with a missing / non-numeric / non-positive onset day, or a participant absent from the population data, are excluded with a named reason, counted in the notes, and exportable as CSV — nothing is dropped silently, in either dataset.

TTE-DATA-003 One population row per participant: later duplicates are dropped with a named, counted reason; the first row is kept.

TTE-GUIDE-001 The in-app notes state, on every render: the derivation rule (time to first qualifying event, censored at end of follow-up), the bands are pointwise 95% CIs (not simultaneous), and 1 − KM can overestimate absolute risk when competing events are present — with the clinical guide carrying the full account.

TTE-USER-008 Clicking an event step dispatches the shared participantsSelected event with the ids whose event occurred at that time, and names them in the footnote — the profile drill-down seam every explorer keeps open (D6).

TTE-USER-007 A selection with no usable records renders the standard empty-state message and no chart — never an empty confident curve.

TTE-DEMO-001 The built demo page composes the endpoint from the vendored adae.csv + adsl.csv — 254 safety participants, three arms, 217 first events with every event qualifying — through the committed bundle with no console errors.

TTE-GUIDE-002 The clinical guide states the estimator's assumptions and limits plainly: the competing-risks overestimation worked through the demo's own censored deaths, the uninformative-censoring assumption, the pointwise-vs-simultaneous distinction, the sparse-endpoint display, and the right-tail at-risk caveat.

TTE-CFG-001 The default settings carry the events + population mapping: participant id, group, follow-up-end day and censoring description on the population side; onset day, event description and the default multiselect filter columns on the event side.

TTE-CFG-002 The display orientation is a validated direction setting: survival is kept, anything else falls back to incidence — the cumulative-incidence safety default (D2).

TTE-CFG-003 The pointwise-95%-band toggle is a boolean ci setting defaulting ON (D3): switching the mandated interval furniture off is the deliberate act.

TTE-DATA-001 The module validates both datasets against the mapping in one error naming every missing dataset and column (events: id + onset day; population: id + follow-up day); the group and description columns stay optional.

TTE-DATA-002 Event rows with a missing / non-numeric / non-positive onset day, or a participant absent from the population data, are excluded with a named reason, counted in the notes, and exportable as CSV — nothing is dropped silently, in either dataset.

TTE-FILT-001 Multiselect filters over the event dataset compose the endpoint live: an event qualifies when, for every filter with an active selection, its value is in the selected set — the whole population stays the denominator, and participants whose events no longer qualify censor at follow-up end. No endpoint list is hard-coded.

TTE-FILT-002 A filter with no active selection qualifies every event; an empty selection qualifies none and draws the honest all-censored display rather than failing. Which columns get a filter is configurable (event_filters), with absent columns dropped with a console warning.

TTE-FILT-003 The notes name the composed endpoint (endpoint_label) and the active filter selection — "All recorded events qualify" or the per-filter selected counts — so the display always states what qualifies an event right now.

TTE-DERIV-001 Each participant's observation is their earliest qualifying event day, ties broken by input order; later events for the same participant never displace the first.

TTE-DERIV-002 An event-free participant censors at the population follow-up-end day, carrying the population censoring description into the censor-mark tooltip; without a usable follow-up day the participant is excluded with a counted reason, never silently.

TTE-DATA-003 One population row per participant: later duplicates are dropped with a named, counted reason; the first row is kept.

TTE-DATA-004 A population with no group column draws one pooled curve labelled "All participants" rather than failing or guessing a grouping.

TTE-STAT-001 One estimator pass per group feeds everything shown: the curve, the band, the censor marks, and every number in the strip table come from the same km.js estimate — there is no second derivation to disagree with the first.

TTE-STAT-002 The product-limit estimator is computed exactly as design §3.1 states, including tie conventions: at risk means observed time ≥ t, and a participant censored at an event time is still in that event's risk set (the survfit convention). Verified digit-for-digit against hand-computed values and the Freireich 6-MP textbook example.

TTE-STAT-003 The variance is Greenwood's formula and the pointwise 95% interval uses the complementary log-log transform, so bounds respect [0, 1] structurally; where the transform is undefined (S = 1 before the first event, S = 0, an exhausted risk set) the bound is null — reported absent, never extrapolated.

TTE-STAT-004 The estimator is cross-validated against R survival::survfit on a frozen ADTTE-shaped fixture: for every endpoint × arm, times, risk sets, event counts, survival and both bounds agree to 1e-10 — including agreement on where the interval is undefined.

TTE-STAT-005 The strip-table numbers are defined exactly: n at risk at t = participants with observed time ≥ t; cumulative events at t = events at times ≤ t — read from the same sorted arrays the estimator walked.

TTE-STAT-006 An entirely-censored group stays at S = 1 with censor marks and no confidence band — true, and visibly different from "no data"; an empty group returns an empty estimate rather than throwing.

TTE-CURV-005 A curve whose largest observed time is a censoring extends flat to that time and stops — the flat tail is the absence of information, not evidence of safety, and the guide says so.

TTE-CURV-006 Group identity is color plus a per-group dash pattern in fixed order — identity, never rank: filters and legend toggles never repaint surviving groups, and the dash is the non-color channel for colorblind readers and monochrome print.

TTE-RISK-001 The strip table shows number at risk and cumulative events per group beneath the time axis — the ST&F guide's mandated furniture — with row labels naming each group in its color.

TTE-RISK-002 Every strip number is centered on its tick's live pixel position via the chart's own scale, over ticks generated by the module — aligned by construction, never by a parallel layout.

TTE-USER-005 An event step's tooltip names the time, the group, the display value with its interval labelled "pointwise 95% CI", and the step's own arithmetic (events over at-risk, censorings sharing the time).

TTE-USER-006 A censor mark's tooltip names the time, the count censored there, and the censoring reasons when the data carries them (CNSDTDSC).

TTE-DEMO-002 The committed population extract is guarded against silent upstream drift: 254 safety participants across the three arms, every follow-up day a positive integer in the measured 1–213 range, each with an end-of-study status.