DECLASSIFIED is grounded in primary-source methodology. So are its improvement recommendations. The following 15 findings draw from peer-reviewed research across learning science, persuasion psychology, cognitive psychology, political science, and structured data standards. Where the research contradicts intuition, the research wins. Where the research is contested, that is noted.
One thing is unambiguous: the project's core architecture — a quiz format applying analytical categories to primary-source evidence — is research-validated. The testing effect (Roediger & Karpicke, 2006) demonstrates that retrieval practice improves retention significantly more than re-reading or passive exposure. Every game session is doing cognitive work that a page of text cannot. The question is how to make it do more.
Current state: Players see a claim → evaluate it → learn the Phillips Pattern. The order is wrong.
DECLASSIFIED's game currently inverts this sequence. Players encounter the claim cold, make their assessment, then see the pattern revealed as an annotation. The inoculation mechanism requires exposure to the technique before the claim, not after. Prebunking outperforms debunking in the literature consistently. This is the most significant structural gap between what DECLASSIFIED does and what the research says works.
Current state: Players enter the game directly. No activation of analytical processing.
This intervention is one sentence. The mechanism is activation of System 2 processing before the first claim loads. Research replicated across multiple studies in political misinformation, health misinformation, and COVID-19 contexts. Three seconds of reading time. Documented efficacy. The phrase "Think about accuracy" before the first claim is enough to measurably shift analytical engagement.
Current state: Each claim is presented large, prominently, in the claimant's own voice, before evaluation.
DECLASSIFIED's current design does prescriptions 2 and 3 reasonably well. It fails prescription 1 — the claim is displayed first, prominently, unframed. This is the exact pattern that strengthens misinformation through processing fluency: familiar statements feel more true because they are processed more easily. Every game session is fighting the continued influence effect with the subsequent correction. The correction wins — but the game is doing unnecessary work.
Current state: Players complete an Act and move on. Claims answered incorrectly are not revisited. Retention decays.
DECLASSIFIED is currently a one-session tool. The research says one-session tools produce short-term knowledge gains that decay rapidly. Spaced retrieval — surfacing missed claims at 24-hour, 72-hour, and 7-day intervals — converts a game session into a learning system. It also drives return visits, which drives time-on-site metrics that improve every other goal the site has.
Current state: From the first claim, players assess four simultaneous dimensions: verdict (4 options), citation tier (4 options), pattern (10+ options), and weighted score. New players face approximately 160 possible combinations before their first feedback.
The four-dimension assessment is the right level of rigor for a player who understands what they are doing. It is a significant friction point for a player who is encountering the Bogost Citation Scale and the Phillips Pattern Library for the first time simultaneously. Scaffolding reduces initial cognitive load without reducing intellectual depth for returning players.
Current state: The Phillips Pattern Library is positioned as an analytical reference — something you consult after learning the verdicts. It is not framed as a prebunking instrument.
The 14 named patterns are the most distinctive intellectual contribution this project makes. They are also the least prominently surfaced. Adding the threat component — explicit framing that these patterns are being deployed against the user right now — activates the inoculation mechanism. "Naming them is the first defense" is not a rhetorical flourish; it is the mechanism that the research says makes technique-based inoculation effective.
Current state: Players arrive and begin evaluating claims without any framing of the analytical task as distinct from political opinion.
DECLASSIFIED's game format is already partially addressing this by framing the task as analytical skill — the scoring system, the pattern recognition mechanic, the Bogost Citation Scale. This is smart design. But it is not made explicit. One sentence in onboarding can activate a different identity frame — the analytical identity rather than the partisan identity — before the claims begin.
Current state: Annotations operate entirely on the Care/Harm and Fairness/Cheating moral foundations — the analytical register of accuracy, documentation, and institutional process.
The annotation for "Mexico will pay for the wall" that cites appropriations records and bilateral treaty absence is maximally persuasive to someone operating from a Fairness foundation. It is less persuasive to someone operating from a Loyalty/Authority foundation. The record doesn't change. The framing can. This is not relativism — it is applying communication research to a documented evidentiary record.
Current state: After verdict reveal, players see whether they were correct. No information about how other players responded.
The caveat is important: social proof must not be manufactured. The site's entire premise is honesty. Invented percentages would undermine everything. This recommendation is contingent on implementing lightweight analytics to generate real session data. Once that data exists, social norm descriptors in verdict feedback are low-cost and research-validated.
Current state: Each claim is displayed multiple times — as the question, in the scoring summary, in the Hall of Shame, and in review contexts.
The game makes corrections — but it also repeats myths. The corrections recover the ground lost to the illusory truth effect, but the game is running a deficit it does not need to run. Eliminating unframed myth repetition in summaries and navigation costs nothing.
Current state: No structured data on any game page. 60 fact-checked claims across six Acts with zero discoverability signaling to search infrastructure.
Google deprecated the visual display of ClaimReview rich results in standard search in 2025. However, as of the March 2026 core update, ClaimReview remains a trust signal for AI Mode citation — which is increasingly how people discover information. This is a discoverability window that currently sits unused across 60 documented fact-checked claims.
<script type="application/ld+json"> ClaimReview blocks to each game file for each claim. Required: claimReviewed, reviewRating (textual: "FALSE"/"MISLEADING"), url, author, datePublished, itemReviewed. Six Acts × 10 claims = 60 structured data entries. All templatable from the existing CLAIMS JavaScript arrays.
Current state: No documented accessibility review. Amber-on-dark palette is unaudited for contrast ratios. Color-blind users may have difficulty with amber/red/green verdict coding.
WCAG 2.1 AA requires a minimum contrast ratio of 4.5:1 for normal text. Approximately 8% of males have some form of color-vision deficiency. Alt text on SVG infographics is absent. Keyboard navigation through verdict selection is unverified. These are not cosmetic concerns — they determine whether the site's content is accessible to a meaningful portion of its potential audience.
Current state: The homepage leads with the game (top of funnel), then cross-sells the book. The methodology page and pattern library are tertiary navigation.
The most effective funnel for educational persuasion tools sequences as: hook → credibility → investment → commitment. Right now the site asks for investment (playing a complex game) before establishing credibility (the methodology page, the primary source standard, the research basis). A skeptical first-time visitor needs to trust the methodology before they will trust the verdicts.
DECLASSIFIED's voice — dry, prosecutorial, Gen X, primary sources — is maximally effective for its existing audience. Social Identity Theory predicts that the same voice is less effective for the audience that most needs the content. This is not a fixable problem. It is a structural feature of identity-based information processing. The recommendation is not to change the voice — the voice is the site's competitive advantage. The recommendation is to build a second entry point.
Current state: localStorage tracks some progress data. No aggregate analytics. No ability to generate the real session data that Social Norm descriptors (Recommendation 9) require.
The site currently cannot answer: What percentage of players get the Raffensperger verdict correct? Which patterns are most frequently misidentified? Which Acts have the highest drop-off? These questions have direct implications for content prioritization, difficulty calibration, and the social proof data that Recommendation 9 requires. Without measurement, optimization is guesswork.