Fact Checker · Veritas

Purpose#

The Veritas Fact Checker — a full-screen overlay that surfaces fact-check results, detail dossiers, source credibility analysis, an AI evidence ranker, a conversational "Ask Veritas" surface backed by a chat-like message list, and a cross-reference view that compares claim treatment across sources.

Entry points#

• /domains/veritas shell surface — More verification tools exposes a "Fact Checker" link (data-veritas-secondary-tool-link="fact-check") to /domains/veritas/fact-check
• /veritas broadsheet subnav — "Fact checker" (data-lilith-subnav-item-id="fact-check") links to /domains/veritas/fact-check
• Direct URL / bookmark — yes (auth required)

The current Claim Tracker route does not include a direct handoff to the Fact Checker route; if V1 later requires that transition, it should be wired as its own visible affordance and covered separately.

Layout regions#

page.tsx is a 'use client' component that renders an sr-only <h1> ("Fact Checker · Veritas · OSHUN") then mounts <VeritasFactChecker onClose={() => router.back()} />. Overlay is fixed full-screen.

• Header: title "Veritas Fact Checker", responsive wrap behavior, and a 44px-minimum Close button (aria-label="Close Fact Checker").
• Nav tablist: six tabs with role="tab", aria-selected, and stable data-veritas-fact-check-tab hooks.
• Body: focusable scrollable region labelled "Fact checker workspace"; current view is driven by view: FactCheckView.
  • Claim Results (results) — list of ClaimVerification rows with verdict badges (VerdictBadge) and filter by verdict
  • Claim Detail (detail) — dossier for selectedClaim (a single ClaimVerification), reached via handleSelectClaim → setView('detail')
  • Source Credibility (credibility) — list / detail for selectedSource (SourceCredibility)
  • AI Evidence Ranker (ai-ranker) — ranked evidence panels
  • Ask Veritas (conversational) — chat-style message log backed by conversationMessages: ConversationalMessage[] (initial state from SAMPLE_CONVERSATION)
  • Cross-Reference (cross-reference) — comparison of how multiple sources treat the same claim

State drivers: view, selectedClaim, filterVerdict, showBothSides, conversationMessages, credibilitySearch, and selectedSource.

States#

• Loading — overlay renders synchronously; no async fetch.
• Results view, populated — fixture verifications render with VerdictBadges (verified | likely_true | inconclusive | likely_false | debunked) and result-count hooks.
• Results view, filtered to no matches — search branch renders "No claims match your search criteria."
• Detail view — reached from a keyboardable list row; direct tab also falls back to the first fixture claim when selectedClaim is null.
• Credibility view — source search and keyboard source-detail expansion are verified.
• AI Evidence Ranker — ranked evidence, credibility analysis, verdict generation, reasoning chain, and source verification chain render.
• Ask Veritas view — SAMPLE_CONVERSATION renders, follow-up prompts reuse the input, and send appends a local user/assistant pair.
• Cross-Reference view — related-claim matrix renders and keyboard drill-in opens the referenced claim detail.
• Confidence gauge animation — animates by default and now removes the arc transition under prefers-reduced-motion: reduce.
• Offline — client-only search/detail/ranker interactions remain usable after context.setOffline(true).
• Standalone PWA — /?surface=pwa restores to /domains/veritas/fact-check and matches standalone display-mode.
• Reduced motion — E2E emulates reduced motion and verifies the confidence arc transition is effectively zero.

Interactions#

Header#

• Close Fact Checker (button, aria-label="Close Fact Checker", X icon)
  • Function: keyboard Enter calls onClose() → router.back()
• Nav tablist (navItems):
  • results — FileText, "Claim Results"
  • detail — Eye, "Claim Detail"
  • credibility — Shield, "Source Credibility"
  • ai-ranker — Brain, "AI Evidence Ranker"
  • conversational — MessageSquare, "Ask Veritas"
  • cross-reference — Link2, "Cross-Reference"

Results view#

• Search input — filters claim text and exposes a no-results state.
• Verdict filter (filterVerdict) — all | verified | likely_true | inconclusive | likely_false | debunked.
• Claim row — click/Enter/Space calls setSelectedClaim and switches view to detail.
• VerdictBadge — colour and icon per VERDICT_CONFIG[verdict].

Detail view#

• Confidence gauge (ConfidenceGauge) — half-circle arc animates to score unless reduced motion is requested; colour ramp ≥80 ok, ≥60 warn, ≥40 alert, else alert.
• See Both Sides toggle — opens supporting and contradicting argument columns.
• Evidence card — click/Enter/Space expands ranking explanation, relevance, credibility, author, and date details.
• Back to results — tab click returns to the results view.
• Trust rule callout — built via buildFactCheckTrustRule (uses resolveWebVeritasTrustRule with surface: 'assistant-evidence').

Credibility view#

• Source search — filters by source name/domain.
• Source row — click/Enter/Space sets selectedSource and shows the credibility detail.
• Source detail — renders description, factual reporting, bias lean, and fact-check count.

AI Evidence Ranker view#

• Ranked panels with evidence items — fixture rank order, overall trust, AI verdict generation, limitations, reasoning chain, and source-chain steps render.

Ask Veritas (conversational) view#

• Message log — renders conversationMessages initialised from SAMPLE_CONVERSATION.
• Follow-up prompt — clicking an assistant prompt copies it into the compose input.
• Compose input + send button — sending appends local user and assistant entries via setConversationMessages.
• Each message — exposes role-coded user vs assistant hooks.

Cross-Reference view#

• Cross-reference matrix renders claims with relationship labels and similarity values; click/Enter/Space on a related claim opens its detail.

Current E2E evidence#

• apps/oshun/web/e2e/veritas-fact-checker.spec.ts — opens from Veritas shell tools and the broadsheet subnav covers opening the route from /domains/veritas via More verification tools, Close/browser history return to /domains/veritas, and opening the same checker from the /veritas broadsheet subnav.
• apps/oshun/web/e2e/veritas-fact-checker.spec.ts — filters fixture claims, opens a dossier by keyboard, expands evidence, and closes by keyboard covers signed-in navigation from /library, initial result counts, close target size/name, no-results search, text search, debunked verdict filtering, keyboard row open, detail claim state, confidence gauge/reduced-motion behavior, evidence-chain counts, both-sides expansion, evidence-card expansion, and keyboard close/back behavior.
• apps/oshun/web/e2e/veritas-fact-checker.spec.ts — audits credibility, AI evidence ranking, and cross-reference drill-in views covers the credibility tab, source search, keyboard source-detail expansion, AI evidence ranker, credibility analysis, verdict generation, source verification chain, cross-reference matrix, relationship hooks, and keyboard related-claim drill-in.
• apps/oshun/web/e2e/veritas-fact-checker.spec.ts — sends a local Ask Veritas message and reuses follow-up prompts covers the conversation tab, fixture message count, follow-up prompt reuse, send action, and appended user/assistant messages.
• apps/oshun/web/e2e/veritas-fact-checker.spec.ts — restores standalone, remains usable offline, and avoids mobile overflow covers standalone PWA launch restoration, offline search/detail/ranker interactions, mobile no-horizontal-overflow checks, and the route-level serious axe scan.

Data & contracts#

• Reads: fixture imports — SAMPLE_CONVERSATION and other sample data (SAMPLE_CLAIMS, SAMPLE_CREDIBILITY, etc.; verify exact imports inside the file)
• Writes: none externally; local state mutations only. Ask Veritas appends local messages; no Sophia/BFF assistant request is made.
• Realtime: none (conversational view is local-only)
• Caching: not applicable
• Auth/role check: shell middleware (verify)
• Trust helpers: buildFactCheckTrustRule → resolveWebVeritasTrustRule({ surface: 'assistant-evidence', ... }) from ./veritasTrustLabels
• Telemetry: no track* imports in this file.
• Page metadata: not set (client component); sr-only <h1> reads "Fact Checker · Veritas · OSHUN"

Cross-references#

• Component: apps/oshun/web/src/components/domains/veritas/VeritasFactChecker.tsx
• Type: FactCheckView (lines 66–72) — six-option enum
• navItems declaration — six-option visible nav
• Route E2E: apps/oshun/web/e2e/veritas-fact-checker.spec.ts
• Sub-components: VerdictBadge (lines 93–121), ConfidenceGauge (lines 123–~189), BreakdownBar
• Sibling tools:
  • domains-veritas-claims.md, domains-veritas-articles.md, domains-veritas-headlines.md, domains-veritas-research.md
• Customer-facing Veritas claim views: veritas-claim.md
• Feature spec: V1/features.md

Open questions / known gaps#

• No telemetry on tab switches or claim selection — surprising given the assistant-evidence surface is exactly what resolveWebVeritasTrustRule is set up to measure
• Ask Veritas view writes to a local array; decide whether V1 wires this to a real Sophia-backed assistant endpoint
• Direct detail tab with no selected claim falls back to SAMPLE_CLAIMS[0] instead of disabling the tab or showing an explicit empty-detail state