
Gamified AI news aggregator for Web3 projects that lets you publish updates in an open, gamified feed and reach real users, with @DeFight_Game
- Tweets3K
- Following240
- Followers1.4K
- Likes4.7K

ALT Card 1 of 4. 2,000 impressions on X, 300 views on Telegram. Which post reached more people? The numbers cannot say: each platform counts a different thing.

ALT Card 2 of 4. Same word, different events. X counts every time the post appears on a screen, not unique. LinkedIn counts times displayed on screen, an estimate. Telegram counts about once per person for four days, forwarded copies included. Sources: X developer docs, LinkedIn Help, Telegram Channels FAQ.

ALT Card 3 of 4. Not a rule but a behaviour. A link asks the reader to leave. Some tap and go, and the like or reply never happens. Same LinkedIn author, same format: a link in a text post, 2% fewer impressions; the preview card as the only visual, 35%. Source: MagicPost, 865,980 LinkedIn posts; vendor data, no cause claimed.

ALT Card 3 of 4. Same scoring, different follow graph. The ranking files read show no separate rule for company accounts. But original posts from mutual follows get +15 on the reply weight, on top of 5. Reposts and replies from accounts a reader does not follow are filtered out. Source: xai-org/x-algorithm, commit of 14 November 2026.

ALT Card 1 of 4. Why is this person number 31? A ranking of experts is a claim about real people, and a claim should be possible to check.

ALT Card 2 of 4. A published method made the outside check possible: 2nd to 18th. Columbia in the U.S. News ranking. A Columbia professor tested its self-reported figures against data it published elsewhere; in 2022 the university admitted submitting inaccurate data in earlier years. Sources: Michael Thaddeus, 2022 analysis; CBS New York, September 2022 report.

ALT Card 3 of 4. A breakdown for every entry. Bar chart for a fictional Analyst K at number 31: component scores out of 100, topic depth (weight 35%) 84, discussion (30%) 48, engagement (20%) 55, consistency (15%) 41; weighted contributions 29.4, 14.4, 11.0 and 6.2, total 61. Data window 90 days, no penalty applied. Fictional example.

ALT Card 1 of 4. What happens when a view count starts to pay? Some participants will find it cheaper to produce the number than to produce the work.

ALT Card 2 of 4. Quote: When a measure becomes a target, it ceases to be a good measure. Marilyn Strathern, writing about audits in British universities. YouTube's policy bans anything that artificially increases views, likes or comments, and holds creators responsible for methods used by someone they hired. Source: Strathern, 1997; YouTube fake engagement policy.

ALT Card 1 of 4. Nobody asked publishers to write angrier posts. Facebook's 2018 ranking rule asked for comments, and publishers delivered what produced comments.

ALT Card 2 of 4. Rewarding comments rewarded anger: 50% to 80%. One Polish party's estimate of how negative its Facebook posts were before and after the 2018 feed change: from 50/50 positive/negative to 80% negative. Source: internal Facebook report, April 2019, as reported by The Wall Street Journal, 2021; the figure is the party's own estimate.

ALT Card 3 of 4. Count it, and the work changes shape. Facebook counted comments, posts turned to outrage. Spotify counted streams past a minimum length, noise tracks were cut to 30 seconds. X apps paid for posts, replies filled with spam. YouTube moved from clicks to time watched. Six documented cases, 2012 to 2026. Usually without anyone breaking a rule.

ALT Card 4 of 4. Change the incentive, not the people. Where platforms responded, they changed the rule: YouTube changed the measure, Spotify added a floor, X cut off the apps. Before you launch a programme, imagine the cheapest piece of work that maximises the count. That is what you will get more of. Read it on Dapp.Expert.

ALT Card 1 of 4. Twenty short posts or four guides people still read? Both are output. A count of posts cannot tell them apart. A count of views cannot either.

ALT Card 2 of 4. Output is a gate, not the measure. Twitch Affiliate, 2017: 500 minutes broadcast and 7 broadcast days in the last 30 days, plus an average of 3 concurrent viewers and 50 followers. YouTube's Partner Program counts subscribers and watch hours, not uploads. Source: Twitch Blog, 2017; YouTube Help, read 15 November 2026 (official).

ALT Card 3 of 4. Tested again, many effects shrank. Many Labs 2: 28 findings, 15,305 participants, 36 countries. About half replicated, and the median effect size fell from 0.60 in the originals to 0.15 in the replications. "Professor" priming: 13% better in 1998, a 0.14% difference across 23 labs. Sources: Klein et al., 2018, Many Labs 2; O'Donnell et al., 2018, Registered Replication Report.

ALT Card 4 of 4. Four questions: a tendency or a law? 1, what was measured, words or actions? 2, how big was the effect, against what? 3, has an independent team repeated it? 4, how close are the conditions to yours? A claim that cannot answer them is a hypothesis, not a law. Read it on Dapp.Expert.

ALT Card 1 of 4. A fine for late pickup made lateness worse. Six day-care centres in Israel introduced a small fine for arriving more than ten minutes late. Removing it did not bring lateness back down.

ALT Card 2 of 4. Late arrivals per centre per week: 8.0 before the fine, 16.4 over the twelve weeks with it, 18.7 after the fine was removed. Four centres without a fine stayed between about 8 and 10. Source: Gneezy and Rustichini, Journal of Legal Studies, 2000; averages calculated from the paper's Table 2.

ALT Card 3 of 4. "A fine is a price". Before, a late parent was imposing on a teacher. After, the parent was buying extra time at a posted rate. The money did not add to the reason people had for being on time. It replaced it. Source: Gneezy and Rustichini, Journal of Legal Studies, 2000.

ALT Card 4 of 4. Can you reward with information instead? Across 128 experiments, expected rewards tied to a task reduced free-choice interest. Positive feedback increased it. Feedback and credit cost less and do not set a price. Source: Deci, Koestner and Ryan, meta-analysis, 1999. Read it on Dapp.Expert.

ALT Card 1 of 4. Same cookies. Why were the scarce ones rated higher? In a 1975 study, cookies from a jar that held only a few were rated as more desirable than the same cookies from a well-stocked jar.

ALT Card 2 of 4. Scarcity works, so it gets faked: 632 low-stock messages found on about 11,000 shopping sites, plus 393 countdown timers and 88 "limited time" messages with no deadline. Not every claim was false, but a reader cannot tell which. Source: Mathur et al., Princeton, 2019, crawl of about 53,000 product pages.

ALT Card 1 of 4. Can a free postcard make people give more? A Zurich charity sent 9,846 letters asking for money. Some contained one or four postcards drawn by a child, as a gift.

ALT Card 2 of 4. Four postcards: 75% more donors. Bar chart: in the Christmas mailing of 2001, 12% of households donated with no gift, 14% with one postcard, 21% with four postcards. In the next mailing, with no gifts for anyone, 9.6%, 8.9% and 8.6%: differences small and not significant. Source: Falk, Econometrica, 2007, 9,846 letters, gift assigned at random.

ALT Card 3 of 4. But free tokens do not buy loyalty. A study of nine major airdrops on Ethereum and Layer-2 networks found that in some airdrops up to 66% of the distributed tokens were sold quickly, often in the recipient's first transaction after claiming. Source: Messias, Yaish and Livshits, arXiv preprint, not yet peer-reviewed.

ALT Card 3 of 4. And the crowd can be manufactured. In a study of crypto manipulation, 20% of the Telegram channels examined were involved in pump-and-dump or Ponzi schemes. More than 56% of the Twitter accounts sharing Telegram and Discord invite links were bots or suspended accounts. Source: Nizzoli et al., IEEE Access, 2020, more than 50 million messages over three months.

ALT Card 1 of 4. When is a claim about writing style worth believing? "Her sentences are short." "He hedges a lot." Each claim hides two questions: compared to what, and measured how?

ALT Card 2 of 4. Two coders, same fragments: did they agree? Bar chart of Cohen's kappa: automated rules against an AI reading, Mollick 0.505, Evans 0.618, White 0.514; AI reading against a blind AI reading, 0.859, 0.91 and 0.937. Target 0.6. Internal Dapp.Expert study, not peer-reviewed. Two machine readings can share the same bias.

ALT Card 3 of 4: Evidence first, verdict last. Step 1, the testimony, attributed. Step 2, the exhibits, shown. Step 3, her reading, marked as hers. Step 4, the innocent explanation, tested. Step 5, the verdict, in the conditional: "That would be fraud." Citation Needed, FTX trial day four, 9 October 2023.

ALT Card 1 of 4. Why are Ethan Mollick's AI posts easy to trust? After reading one, you usually know what was tested, what was found, how far to trust it, and what you could try tomorrow.

ALT Card 2 of 4. Result first: consultants using AI in a 2023 study with BCG were 25.1% faster. They also finished 12.2% more tasks. Then, before any explanation, the limits: a new working paper, not yet peer-reviewed (since published in Organization Science, 2026). Source: Mollick, One Useful Thing, 16 Sep 2023; HBS working paper.
















































