Inspiration

False emergency alerts, recycled flood photos, truncated pollution graphs, and typosquatted disaster-aid links spread fastest when people are stressed and scrolling quickly. Research from the Stanford History Education Group (Civic Online Reasoning) shows that professional fact-checkers do not just read a post closely from top to bottom. Instead, they practice lateral reading: immediately opening parallel verification checks on the domain registry, the original image or chart baseline, and primary agency feeds before deciding whether to trust or share a claim.

SourceDesk turns lateral reading into an interactive newsroom verification game built for the Girls Who Code HHS Hackathon 2026 ("Level up your gameplay, level up the world. Build a game for social good").

What it does

Players step onto an evening civic verification desk where six high-velocity dispatches arrive across public safety, clinical health claims, emergency relief, severe weather, environmental data, and transit alerts:

Emergency Nighttime Dam Release (SD-2026-01): A viral spillway photo paired with an urgent evacuation rumor that turns out to be a cropped August 2017 archive photo (MISLEADING CONTEXT).
NeuroSpark 300% Recall Study (SD-2026-02): A viral energy-drink headline extrapolated from an unblinded n = 11 manufacturer-funded poster where raw recall shifted from 10.0 to 10.4 words (MISLEADING CONTEXT).
48-Hour Storm Relief Direct Deposit Portal (SD-2026-03): A mass-forwarded SMS link demanding Social Security and bank routing numbers on a 36-hour-old typosquatted .org domain (FABRICATED / IMPOSTER).
National Weather Service Heat Warning & Cooling Centers (SD-2026-04): An authentic municipal heat bulletin corroborated by primary .gov ASOS station telemetry (VERIFIED BULLETIN).
Truncated Particulate Pollution Graph (SD-2026-05): A bar chart claiming a 90% overnight drop in PM2.5 by starting the vertical axis at 88.0 instead of 0.0 (MISLEADING CONTEXT).
Synthetic Transit Shutdown Voice Note (SD-2026-06): A forwarded audio clip with a hard 8.0 kHz text-to-speech spectral shelf and zero transit agency corroboration (FABRICATED / IMPOSTER).

For every dispatch, players operate four lateral-reading instruments ([1] Domain & WHOIS Registry, [2] Media & Chart Forensics, [3] Primary Source Cross-Check, and [4] Urgency & Framing Scanner), inspect the interactive side-by-side SVG forensic diagrams, and press their editorial verdict stamp ([V] Verified Bulletin, [M] Misleading Context, or [F] Fabricated / Imposter).

SourceDesk also includes a live Custom Claim & Headline Verifier Checklist so students can paste real-world headlines or forwarded links outside the six dispatches and compute a deterministic 0-100 Credibility Risk Score with actionable verification steps.

How we built it

Zero-Dependency ES Module Engine (js/sourcedesk-engine.js): Pure, deterministic state machine managing dispatches, probe inspections, Evidence Efficiency bonuses (rewarding players who identify the 2 highest-signal forensic probes rather than clicking blindly), streak multipliers, civic reach protected, and custom claim risk evaluation.
Swiss Editorial Newsroom Interface (index.html, styles.css, js/app.js): Designed with warm archival newsprint (#F6F3EC), carbon ink (#121619), Instrument Serif + Space Grotesk + JetBrains Mono (tabular-nums), custom inline SVG forensic diagrams, full keyboard shortcuts (1-4, V/M/F, N), synthesized Web Audio API mechanical stamp and probe feedback, and a printable 1-page Lateral Reading Field Guide (@media print).
Automated Verification & Bot Playtest Harness (tests/sourcedesk.test.js & scripts/verify-playtest-matrix.js): Verified with 48/48 unit tests (node --test), 20/20 (100%) planted-defect mutation catches, and an exhaustive automated bot playtest simulation across all 592,704 possible clue-inspection and stamp combinations.

Challenges we ran into

Balancing the scoring model so players are genuinely rewarded for thoughtful lateral reading rather than either blind-guessing or mindlessly clicking every button required careful tuning. By weighting each dispatch's two primary forensic probes and applying an Evidence Efficiency bonus (+10 pts) alongside a Blind Stamp penalty (-25 pts when stamping with zero probes inspected), the scoring mechanics reinforce real-world verification habits.

Accomplishments that we're proud of

Exhaustively verifying all 592,704 possible gameplay state transitions in scripts/verify-playtest-matrix.js with zero reachable NaN states or scoring overflows.
Building interactive SVG forensic comparisons right into the dossier card (such as comparing a truncated 88.0-92.0 axis against a true 0.0-100.0 zero baseline, and visualizing an 8.0 kHz synthetic voice clone spectral cutoff).

What we learned

Designing for media literacy works best when the interface mirrors the actual workflow of investigative fact-checkers: stepping outside the viral post to inspect domain registration age, primary telemetry feeds, and uncropped baselines before amplifying a claim.

What's next for SourceDesk

Expanding the dispatch library with community-contributed verification puzzles, adding a two-player cooperative copy-desk timed shift mode, and exporting classroom-ready printable lateral-reading case cards.

Verification & AI Disclosure

Playtesting & Verification Note: Playtesting and scoring balance were verified via an automated bot simulation harness across 592,704 simulated clue-inspection and decision combinations (scripts/verify-playtest-matrix.js), not human subjects.
Hackathon Rule 2 AI Disclosure: Built with AI coding and editorial illustration assistance per Girls Who Code HHS Hackathon Rule 2; all game rules, dispatches, lateral-reading probes, and automated test suites are open-source and deterministically verified in the repository.