We asked the billion-dollar question

Research report, 1 October 2026. Prepared for the founder and a prospective co-founder/sponsor. Synthesised from six research notes in docs/strategy/research_notes/Arena billion dollar thesis/; the team's earlier MARKET-CENTRAL-ASIA.md (12 Aug 2026) and GTM-PLAN.md (19 Aug 2026) are referenced where the new evidence confirms or contradicts them. Every figure is cited to a source; estimates and judgements are labelled as such.

Bottom line. Arena as the founder describes it today (a Kyrgyzstan-first, sponsor-funded, cash-prized intellectual sports league) has a probability of becoming a billion-dollar company that we put well under 1%; no sponsorship-, prize- or competition-centred platform in the entire comparable set has ever crossed $1B, and Kyrgyzstan's whole reachable 15–34 online cohort (~2.1M people) is smaller than one week of Chess.com's new registrations. The same idea, re-aimed, has a real but still slim shot, which we estimate at 2–5% over ten years, if three things become true: Arena enters Uzbekistan (the only Central Asian market with the youth volume, ~10.9M online 15–34-year-olds, and a proven 2.5M-person appetite for free mass coding programmes) by year two; the rated league is treated as the top of a funnel whose paying customers are employers and Western clients, not learners (the Turing/Andela route, which produced the only emerging-market unicorns in this category, at $300M revenue and $2.2B valuation for Turing); and the product runs on Duolingo-class retention economics (72% gross margin, ~11% of revenue on marketing, 9% of monthly actives paying) rather than on prize money. The far more probable good outcome, perhaps a 15–25% chance, is a $30–100M regional talent-and-skills company that would already be the largest technology business ever built out of Bishkek. Along the way the evidence overturns several of the founder's working assumptions: the Jyldyz Academy figures of 15,000 learners and 9,000 women cannot be verified anywhere public (the verifiable number is ~3,000); Kyrgyz sponsor budgets for skill contests run 100,000–300,000 som per event, not the six figures a season needs; the current iOS-only TestFlight build excludes 73–84% of Kyrgyz phones; MIT OpenCourseWare and Khan Academy are both CC BY-NC-SA and cannot legally be hosted or translated inside a sponsor-funded product; and Kyrgyz lottery law lets the regulator reclassify any marketing event with "lottery elements" as a promotional lottery, which would make under-18 cash prizes illegal and expose the organiser to confiscation of all proceeds. The league is still worth building, but for a different purpose.

The verdict: under 1% as framed, 2–5% as a talent exchange

The billion-dollar question has a clean empirical answer because the comparable set is unusually complete. Across roughly thirty gamified-learning, skill-competition, ranking and community platforms, nine reached a documented $1B valuation, and every one of them did it through one of three revenue lines: a consumer subscription bought by 5–10% of a very large monthly-active base (Duolingo, Chess.com, Quizlet), an institutional subscription attached to a free viral product (Kahoot!, Coursera/Udemy), or B2B talent and AI-data services sold to deep-pocketed buyers (Turing, Andela, Toptal). Duolingo closed FY2025 with $1,037.6M revenue, 133.1M MAU and 12.2M paid subscribers, a 72.2% gross margin and sales-and-marketing spend of only ~11.4% of revenue (Duolingo Q4/FY2025 shareholder letter). Chess.com was described as a billion-dollar company in 2023 on $100M+ of revenue from premium memberships, advertising and broadcast rights (Inc.; Sherwood). Turing tripled from $120M to $300M revenue in 2024, turned profitable, and raised at $2.2B in March 2025 by re-pointing its 4M-developer network at frontier AI labs (Sacra).

The platforms that look most like the founder's description, built on competitions, leaderboards, prizes and sponsors, are all absent from that list. Kaggle was sold to Google in 2017 for an undisclosed and reportedly small sum (Wikipedia: Kaggle; HN); HackerRank topped out around $500M in 2022 (SiliconANGLE); CodeSignal at $488M (Forge); Topcoder was folded into a $500M deal for all of Appirio (Staffing Industry Analysts); Devpost raised a $4.6M Series A and never more (Wikipedia: Devpost); Gitcoin distributed ~$11M in 2024 and missed its own target (Gitcoin governance forum); Codeforces runs 11,000-participant weekly rounds on T-shirts and a $125,000 community donation (Wikipedia: Codeforces; Codeforces blog). Even in chess, where prize money is most visible, Chess.com's entire 2025–26 Champions Chess Tour pays about $1.03M, roughly 1% of revenue; prizes are a marketing line, not a business (Wikipedia: CCT 2025–26).

The scale arithmetic is what kills the consumer route for a Central Asian company. Observed revenue per monthly active user clusters at $4–10 for subscription platforms (Duolingo ≈ $7.80 per MAU, ≈ $19.70 per DAU) and under $1 per registrant for nonprofits and competition sites (Khan Academy spends ≈ $0.46 per registered user per year) (Duolingo Q4/FY2025; ProPublica: Khan Academy). A $1B consumer business at Duolingo's 9% paid penetration and ~$7/month blended price needs on the order of 50M DAU and 130M MAU; at Central Asian wages ($491–508/month in Kyrgyzstan) the sustainable price is $2–5/month, so the user base required is three to five times larger again (One.uz; AKIpress). The research's estimate of Arena's whole serviceable market (Kyrgyzstan plus urban Uzbek and Kazakh youth plus the Russian-reading diaspora) is 11–12M people, of whom Kyrgyzstan is about 2.1M or 18% (DataReportal Digital 2026: Kyrgyzstan; PopulationPyramid Kyrgyzstan 2025). A product that captured every one of them at Kyrgyz price points would still be a sub-$500M-revenue company.

Valuation multiples make the bar higher still. Coursera and Udemy IPO'd at a combined ~$7.5B and merged in 2025–26 at ~$2.5B, later described as ~$1.8B, on roughly $1.55B of combined revenue, about 1.5x revenue for sub-10% growth, loss-making learning marketplaces (Wamda; Class Central). Kahoot! was taken private at $1.72B on $163.5M ARR, about 10.5x, after a 75% markdown from its 2021 peak (Kahoot! Q2 2023 report; Wikipedia: Kahoot!). Turing's $2.2B on $300M is ~7x (Sacra). So "a billion" means roughly $100–150M of fast-growing, high-margin ARR, or $300–700M of slower revenue. Time-to-$1B for the consumer winners was 8–10 years (Duolingo 8, Kahoot 8, Coursera 9, Udemy 10, Quizlet 15, Chess.com 16), and only the B2B services route was faster (Turing, 3 years) (Wikipedia: Duolingo; Wikipedia: Quizlet; Chess.com GA announcement). Capital is the final constraint: all Central Asian startups together raised a record $320M in 2025 (Kazakhstan $209M, Uzbekistan $99.3M, Kyrgyzstan not even itemised), and Kyrgyzstan's own High Technology Park lists "produce one unicorn" as an aspiration, not a track record (Times of Central Asia; Times of Central Asia: HTP).

What would have to be true, then, is specific. Arena would need to be in Uzbekistan by year two and in Kazakhstan and the Russian diaspora by year three, because Kyrgyzstan alone cannot produce the user base. It would need a demand-side customer (employers, IT-park exporters, Western automation clients, eventually AI labs) paying for access to the ranked top of the ladder, because that is the only revenue line in the comparable set that produced a billion-dollar emerging-market company at a user base Arena can plausibly reach (Turing: $300M revenue on 4M developers, ≈ $75 per registered developer, all of it paid by a few dozen clients) (Sacra). It would need a free rated loop with Duolingo-grade retention so that growth comes from product mechanics rather than paid acquisition. It would need outside capital at a scale the region has never supplied to a Kyrgyz company. And it would need eight to ten years. Each of those is individually achievable; their conjunction is what produces our low single-digit estimate. The table below sets out the three proven routes and what each would demand of Arena.

Sources: Wikipedia: Duolingo; Inc.; PitchBook; Wikipedia: Kahoot!; SiliconANGLE; Sacra; Wikipedia: Andela; FutureProofing.

Kyrgyzstan is the right laboratory and the wrong market to stop in

The founder's instinct to start at home is defensible on the numbers, provided everyone understands it is a laboratory. Kyrgyzstan has 7.32M people with a median age of 25.4, and roughly one in three (≈2.4M) is aged 15–34 (DataReportal Digital 2026: Kyrgyzstan; PopulationPyramid). The pressure that makes a skills league relevant is acute: official youth NEET is 16.7% (2024) and UNICEF puts it near 20%, with women at 29.4% against men at 12.0%; an estimated 79% of employed youth work without formal contracts; informal employment is 66% of the total; the average wage is ~42,954 som, about $508/month, the lowest minimum wage in the Eurasian Economic Union (Kyrgyzstan SDG 8.6.1; UNICEF Generation 2050; 24.kg; AKIpress; Times of Central Asia). Remittances were 17.6% of GDP in 2024 ($2.99bn, 93% from Russia), down from 26.9% in 2022, and the Russian escape valve is closing: the Kyrgyz migrant stock in Russia fell from 1.063M in 2022 to ~352,000–380,000, 128,497 Kyrgyz citizens sit on Russia's register of "controlled persons", and entries to Russia fell 24% year on year in Q3 2025 (World Bank API; Akchabar; Asia-Plus; IOM Kyrgyzstan H2-2025). A returnee migrant with a smartphone and no formal job is exactly the person a free, ranked, phone-based route into remote work serves. The team's August market memo quoted remittances at ~27% of GDP; that was the 2022 figure, and the 2024 number is 17.6%. The dependence is real but falling, which strengthens the "re-skill at home" story (World Bank API).

Connectivity is a solved problem, and distribution is highly concentrated. Internet penetration is 88.5%, there are 11.7M mobile connections (159% of population), smartphone adoption is 83%, 4G covers 99.2% of the population, and mobile data costs about $0.17 per GB, the second cheapest in the world (DataReportal; GSMA Mobile Economy Eurasia 2025; Kabar). Instagram alone reaches 3.90M Kyrgyz users (53.3% of the population, +32% in a year), more than the entire 15–34 cohort, while LinkedIn has 350,000 and X 68,800; WhatsApp is the dominant messenger (DataReportal; Central Asia Barometer). The one hard fact the current build ignores is the operating system: StatCounter puts Android at 73.4% of Kyrgyz mobile OS share in November 2025 and 84.4% in September 2026, with iOS at 15.5–26.5% (StatCounter, Nov 2025; StatCounter, Sep 2026). An iOS-only TestFlight launch therefore addresses between a sixth and a quarter of the market, and that minority skews toward Bishkek and higher incomes, the opposite of the NEET, rural and female cohorts the mission claims. Uzbekistan is 82.3% Android; only Kazakhstan, at 54.4% Android and 45.2% iOS, rewards iOS parity (StatCounter Uzbekistan; StatCounter Kazakhstan).

Language is a three-layer problem that the founder's trilingual plan handles correctly but that content sourcing does not yet. Eighty percent of the population speaks Russian and 70–80% of university programmes are Russian-medium, yet 75% of schoolchildren attend Kyrgyz-medium schools and in many rural communities Kyrgyz is the only language spoken (24.kg: Minister Kendirbaeva; Times of Central Asia; 24.kg; Internet Society). English proficiency is "very low" across the region (EF ranks Kyrgyzstan 101st, Uzbekistan 104th and Kazakhstan 107th of 123), so "English for AI tools" is itself a track, and PISA baselines (Uzbekistan reading 336 against an OECD average of 476) mean tasks must be scaffolded rather than olympiad-hard (Akchabar; OECD PISA 2022: Uzbekistan). The practical formula is Russian-first for content depth, Kyrgyz for UI, onboarding and the typing/reading disciplines, English as a learned skill.

On the sponsor side, Kyrgyzstan is small. The High Technology Park's 463 residents earned $195.4M in 2025 (+50%), 95% of it exported, and employ about 3,000 people against a stated goal of 50,000 trained specialists, a 10–15x gap that is Arena's best institutional argument (Kaktus.media; Times of Central Asia). But every Kyrgyz skill-contest prize found in 2025 was small: 100,000 som for a Bishkek hackathon, 200,000 som from the Ministry of Agriculture, "from 300,000 som" for Startup Nation, $15,000 for an earlier Startup Nation, $20,000 plus a trip for ETH Bishkek, 200,000–300,000 som for the esports federation's student leagues (Kaktus; Vesti.kg; Economist.kg; NSU.kg; esf.kg). No bank-run skill contest by Optima, MBank or Kompanion was found for 2024–26; MBank runs consumer prize lotteries, not skill events. The team's market memo lists "Kaspi, MBank, Optima, Ipoteka" as a sponsor set, but the only banks verifiably sponsoring skills competitions are Kazakh: Kaspi.kz, Freedom Bank and Samruk-Kazyna as Gold partners and Halyk as Silver on Kazakhstan's $1.1M HackAlem AI (Trend.az). Donor money has also thinned: USAID cancelled 78% of its Kyrgyzstan programmes (another source says 69%) and effectively 100% in Kazakhstan, Uzbekistan and Turkmenistan in 2025, and UNDP small grants are capped at $6,000 (The Diplomat; Times of Central Asia; UNDP). A realistic first-year Kyrgyz sponsor pool is $50,000–150,000, which is already larger than any single 2025 hackathon prize and is the right ambition for a Season 1.

Scale lives next door. Uzbekistan has 37.2M people, median age 27, 89% internet penetration, social media use growing 41.7% a year, and a Telegram-first culture (70% prefer it) (DataReportal Digital 2026: Uzbekistan; Central Asia Barometer). Its appetite for free mass skills programmes is proven at a scale nothing in Kyrgyzstan approaches: One Million Uzbek Coders registered 2,503,060 people and issued 1,175,933 completion certificates, 47% of them to girls (CICT.uz); the President Tech Award drew 6,000 participants for a $1M pool and is rising to $5M in 2026 with a new $1M AI award (Times of Central Asia); IT Park exports reached $940M in 2025 against a $5bn 2030 target, with 3,500 resident companies (UzDaily; HKTDC); and Uzbekistan ranks first in the world for Coursera learners as a share of its labour force (Eurasianet). Kazakhstan is the money market rather than the pressure market: youth unemployment is 3.8%, wages are $790–900, the top-90 EdTech companies earned ₸95.4bn ($200M) in 2024, the state pays private IT schools 400,000–500,000 tenge per learner through Tech Orda, and Astana Hub participants booked $1.5bn of revenue in 2025 (World Bank API; Times of Central Asia; TAdviser; Astana Hub Tech Orda; Astana Times). Tajikistan (56.8% internet) and Turkmenistan (46.1%) are not near-term markets (DataReportal: Tajikistan; DataReportal: Turkmenistan).

Putting it together, the research estimates a TAM of ~25M online 15–34-year-olds (≈22M in-region plus ~3M Central Asian migrants in Russia), a SAM of 11–12M, and a serviceable obtainable market of 25,000–60,000 registered and 5,000–15,000 monthly competitors in year one (Kyrgyzstan only), 300,000–600,000 registered by year three (with Uzbekistan, Kazakhstan and the diaspora), and 1.5–3M registered with 300,000–700,000 monthly competitors by year five, with sponsor pools of $50–150K, $0.5–1.5M and $1–5M respectively, the top end matching the President Tech Award's 2026 fund and Kazakh esports event pools. These are estimates built from the cited inputs, not sourced forecasts. They are two to five times higher than the 60,000–120,000 registered users the team's own memo projected for 36 months, mostly because the memo treated Kyrgyzstan as the main market and the research treats it as the beachhead. The year-one monetisation these numbers imply, $3–10 of sponsor money per monthly active per year, is ad-model economics, not subscription economics; it can fund a league but not a company, which appears only when Uzbekistan and the demand side arrive.

Comparables: the winners sold to 5–10% of users or to institutions; the losers sold excitement

The benchmark table below is the single most useful artefact for a sponsor conversation, because it shows where Arena's described model sits relative to every documented outcome.

Sources: Duolingo Q4/FY2025; Sherwood; Kahoot! Q2 2023; PitchBook; Wamda; Sacra; Wikipedia: Andela; Vendr; GetLatka: HackerRank (estimate); Wikipedia: Kaggle; GetLatka: Skool (estimate); ProPublica: Khan Academy.

Two comparables deserve a closer look because the founder has named them. Khan Academy is the aspiration, and its FY2025 Form 990 is sobering: $117.5M of revenue, of which $90.4M was contributions, $87.5M of expenses, and 189M registered users, roughly $0.46 of spending per registrant per year (ProPublica). Khan is a public good and, as a business, a $90M-a-year philanthropy whose scale is set by the Gates Foundation, not by learners; it only began earning meaningful revenue ($22.2M) through district licences and Khanmigo at $4/month (khanmigo.ai/pricing). The Skool community the founder cites, Nate Herk's AI Automation Society, is the clearest picture of the community-to-paid funnel: 440,000+ free members and about 3,500 paying $129/month, a 0.8% conversion that grosses roughly $451K/month for the creator, on which Skool itself takes $99/month plus 2.9–3.9% (skoolmakers; skool.com about page; RevenueGeeks). Herk's audience is global and English-speaking; a $129/month tier is 26% of a Kyrgyz average salary. The community mechanics transfer; the price point does not.

QuizUp reached 70–100M users, could not monetise, and was sold for $7.5M (productmint). HQ Trivia hit 2.3M concurrent players in February 2018 and was dead by February 2020; its board blamed the CEO for "upping cash prizes instead of introducing new game formats", bots reached ~90% accuracy, and prize payouts were delayed for weeks (productmint; TechCrunch). Triplebyte ran profitably at ~$10M, pivoted four times under venture pressure, and shut in 2023 because "we never solved candidate acquisition at scale" and because the candidates it attracted (junior, international) were not the ones clients wanted (Otherbranch postmortem). Indian consumer edtech was a wipe-out (BYJU'S $22B to zero, Unacademy $3.5B to ~$206–500M) on bookings-led growth that reversed with demand (Wikipedia: Byju's; TechCrunch). Google shut Code Jam after twenty years within months of reclassifying it as a cost centre (Pragmatic Engineer). The pattern across all of them: cash prizes are a weak retention lever and a strong cheating magnet; a ranking that promises jobs must pre-solve who buys the top of the leaderboard; and a league with a single corporate patron dies when that patron's hiring need falls. Even Duolingo now shows the limits of pure consumer gamification: in Q2 2026 DAU grew 23% but bookings only 8%, and its EBITDA margin slipped from 31.2% to 25.9% (Duolingo Q2 2026 letter).

The revenue ladder, and the laws that gate every rung

The sequence the evidence supports runs sponsor-funded competitions first (cash now, lowest legal risk if skill-only and free to enter), then recruiter seats and placement fees on the rankings, then state and institutional money (Kazakhstan first), and consumer premium last and lowest. Each rung has a known price card and a known legal gate.

Sponsors globally pay in three stackable ways: they fund the prize pool (pass-through, zero margin), they pay a platform or services fee (Gitcoin charges 10% of the pool; Devpost and Kaggle sell tiered services at unpublished prices), and they buy branding tiers that range from $1,000–4,000 at a small hackathon to $2,500–15,000 at AAAI's and $10,000–50,000 at Berkeley's (Gitcoin; Devpost; AAAI; Hack Berkeley). Kazakhstan already buys this: HackAlem AI's $1.1M pool carries Gold and Silver bank partners, and Decentrathon 3.0 put ₸16M in front of 2,670 participants from 19 countries with a built-in job fair (Trend.az; Astana Hub). A defensible regional rate card (an inference from these tiers, not a sourced price) is a 15–25% services fee on the pool plus $2,500–15,000 branding tiers, which puts a plausible first-year band at $100,000–500,000 if Kazakh sponsors are in the mix and nearer the bottom of that range if Kyrgyz-only. The team's GTM plan already specifies a 15% platform fee waived for year-one anchors, which is consistent with the Gitcoin precedent and with the finding that the second sponsor is sold with the first sponsor's logo.

The hiring rung is where the money is, and it has two products. Recruiter and assessment seats sell at HackerRank's $1,990 (Starter) to $4,490 (Pro) per year list, with enterprise quotes of $200–600 per seat per month (Vendr; InterviewCost). Placement and staffing keep 30–55% of client billings: Toptal marks up 30–50%, Turing's margin is put at 50–55%, Andela charges $6,000–15,000 per month and a $50,000 direct-hire fee (FutureProofing; Flexiple). The skill the founder wants to teach is the one with the clearest external price: on Upwork, n8n is the most-requested automation tool (534 postings, 16.1% of AI-automation posts in May 2026, ahead of Zapier at 11.9% and Make at 10.7%), the median hourly rate is $29.50 with a $20–40 interquartile band and $75–120+ for agent-architecture work, entry-level sellers advertise $10–35/hr, and seniors $125–250 (Upwatcher; Upwork n8n experts; Ciphernutz). Upwork's own 2026 skills report shows AI-application earnings up 109% year on year, with AI integration +178% and chatbot development +71% (Upwork via GlobeNewswire). Against a Kyrgyz average wage of ~$3/hour and platform-worker earnings of ~$14.92/hour, a staffing arm billing Western clients $40–100/hour and paying regional talent $15–40 sits comfortably inside the Turing margin band (Academia: platform work in Kyrgyzstan). Local demand is thin (hh.kz shows only 14 n8n vacancies inside Kazakhstan, Habr Career two, and Russian boards pay 150,000–300,000 RUB/month), so the job the league leads to is remote, not local (hh.kz, area 40; Habr Career; hh.ru vacancy). Two cautions: prompt engineering "as a standalone job title has largely contracted" and belongs inside the builder path, not as a destination (ilinmaks); and no employer-recognised automation certification yet exists (n8n's exam is still "coming soon"), so a platform-issued badge tied to a public GitHub portfolio is as credible as anything on the market (n8n Education).

The institutional rung is Kazakh first. Tech Orda has paid private IT schools to train 12,500+ people since 2021, reports 88–90% employment and an average graduate salary of 505,000 tenge, and accredited 79 schools in 2025, with 1,287 employers including Kaspi, Halyk, Freedom, EPAM and Yandex hiring from it (Astana Hub Tech Orda; finance.kz). Becoming an accredited provider turns completion and employment data, which a league produces anyway, into state-paid revenue of a few hundred dollars per learner. Kyrgyzstan has no equivalent voucher; its digital-skills money sits inside World Bank and ADB education loans (36,000 teachers trained, 1,200 schools equipped) procured by ministries, which means tendering as a subcontractor, not selling software (World Bank results brief). Uzbekistan's IT Park grants residency to education companies only if more than half their adult graduates are hired by exporters, a rule that rewards the outcome Arena would track (buxgalter.uz).

Consumer premium comes last because the numbers are small. Super Duolingo's annual plan works out to $5/month; Boot.dev prices itself in Kyrgyzstan at **KGS 8,999 per year ($100), which marks the ceiling of what the market bears**; a Bishkek bootcamp costs 7,000–30,000 som per month (DealNews; Boot.dev pricing; Economist.kg). The research's inference is a sustainable Kyrgyz ARPU of $2–5/month at 1–2% conversion, with Kazakhstan the only in-region market where a $10–20 tier has evidence behind it. The team's GTM plan models Altai Pro at $9/month bought by 10% of monthly actives; the evidence supports neither number in Kyrgyzstan, and the Pro line should be re-based or moved to Kazakhstan. Because 73–84% of users are on Android, Google Play's 15% on the first $1M and on all subscriptions is the effective platform tax, and Apple's Small Business Program matches it at 15% under $1M (Google Play Console Help; Apple). Apple Guideline 3.1.1 forbids using local rails or crypto wallets to unlock features inside the iOS app, so MBank, O!Dengi and ELQR checkout belongs on the web (Apple App Review Guidelines; ELQR).

The legal gate deserves to be read in full, because it is the one place where the founder's "no age limit" and "cash prizes" collide. Kyrgyzstan's Law on Lotteries (No. 78 of 10 May 2017) defines a win as something that "является исключительно делом случая" (exclusively a matter of chance), so a contest decided by demonstrated skill is, on the statute's own words, not a lottery and needs no permit (FSA Kyrgyzstan, law text). But three articles bite. Article 5(8) lets the regulator declare that any marketing or promotional event containing "lottery elements" is a promotional lottery "вне зависимости от занимаемой по этому вопросу позиции ее организаторов", that is, regardless of the organiser's view. Article 1(17) restricts lottery participants to those aged eighteen and over. Article 20(1) punishes an unpermitted promotional lottery with confiscation of the entire proceeds to the republican budget plus criminal liability for organisers and officials. Article 5(11) separately bans lotteries inside mobile-operator networks, a trap for any carrier-billed promotion with O!, Beeline or MegaCom (same source). The Law on Gambling (No. 50 of 30 June 2022) is irrelevant unless Arena ever takes stakes (Kyrgyz citizens may not gamble at all), but it is the law a hostile reading would reach for (Continent Online; Vitvet). Prize income is taxed at 10% in Kyrgyzstan, withheld and remitted by the paying organisation; Kazakhstan withholds 10% for residents and 20% for non-residents with a 6-MRP (25,950 tenge) de minimis; Uzbekistan taxes olympiad and contest prizes at source (Zakon.kg; MyBuh.kz; Norma.uz). The app stores add their own layer: Apple 5.3.1–5.3.2 require contests to be sponsored by the developer with official rules in-app naming Apple as uninvolved, 5.3.3 bans in-app purchase of anything used "in conjunction with real money gaming of any kind", and Google Play prohibits "games that accept money in exchange for an opportunity to win a physical or monetary prize" and any in-app call to action to join real-money contests (Apple; Google Play policy). Apple's new 13+/16+/18+ age ratings took effect with mandatory questionnaires by January 2026, and Kyrgyz banks must KYC every top-up "будь то 500 сомов или 50 000", with minors' accounts requiring guardian consent (MacRumors; Vesti.kg).

The design rule that satisfies all three regimes simultaneously is: free entry with no consideration of any kind; winners determined by skill alone, with no random drop, lucky draw, loot box or spin anywhere near a real prize; prizes funded by sponsors and paid from a local legal entity that withholds tax; official rules in-app naming Arena, not Apple or Google, as sponsor; and no in-app currency that converts into prize eligibility. On age, the workable posture is one ladder open to all for learning, ranking and non-cash rewards, cash prizes restricted to verified adults or paid to a verified guardian, scholarship, device and internship prizes for minors, a 13+ or 16+ rating, and no third-party ad SDKs if under-13s are admitted. Uzbekistan's own IT Park runs a cash-prize hackathon for 16–20-year-olds, so minors receiving prizes has precedent in the region, but the Kyrgyz 18+ lottery rule and KYC make it the exception to design around rather than the default (IT Park Uzbekistan). Cross-border payouts should run through Payoneer or Wise, both available to Kyrgyz residents; Kyrgyzstan's $31bn-a-year, 100-plus-VASP stablecoin corridor is legal but under US and UK sanctions scrutiny, and bank sponsors will not accept crypto-paid prize programmes, so keep it out (iGen Payments; OSW, May 2026). One more regulatory watch item: Kyrgyzstan is considering licensing online courses, coaches and mentors, which is a reason to position Arena legally as a sport and federation partner rather than as a course provider (newsroom.kz). All of this should be confirmed with Kyrgyz counsel; the statute leaves Finnadzor discretion and no enforcement case against a skill contest was found.

Free content: OCW and Khan cannot be hosted, but a permissive AI corpus exists in Russian

The founder's plan to build the learning layer on OpenCourseWare runs into a licence wall. MIT OpenCourseWare is CC BY-NC-SA 4.0, and MIT's own gloss is that "non-commercial use means that users may not sell, profit from, or commercialize OCW materials or works derived from them" (MIT OCW Terms of Use). Harvard CS50, Stanford Engineering Everywhere, The Odin Project, OpenStax's Python book and Project Euler carry the same NC-SA licence (CS50 License; Stanford SEE; Odin Project licence; OpenStax; Project Euler). Khan Academy is CC BY-NC-SA too, with SAT content excluded entirely, though Khan explicitly "encourages linking to its materials and embedding videos" (Khan Academy ToS). Creative Commons defines NonCommercial as use "not primarily intended for or directed towards commercial advantage or monetary compensation"; a platform that places this content inside sponsored challenges is, on the plain reading, using it commercially (CC BY-NC-SA 4.0 legal code). Translation is an adaptation, so a Kyrgyz OCW or Khan would need written permission. The right treatment for all of it is deep-link and embed, never host, remix or translate. Two further surprises: freeCodeCamp's current README no longer grants any Creative Commons licence over its curriculum, only a bare copyright notice, so it should be treated as proprietary until confirmed otherwise (freeCodeCamp README); and Elements of AI, the Finnish national-AI-literacy model, is "all rights reserved" with no Russian or Central Asian edition; a Kyrgyz version would be a licensing deal with MinnaLearn and the University of Helsinki, whose EU editions were paid for by the European Commission (Elements of AI FAQ; Wikipedia: Elements of AI). Microsoft Learn's hosted site, DeepLearning.AI's short courses, fast.ai's prose and Anthropic's Academy are likewise non-reusable; n8n's documentation sits under its Sustainable Use License (Microsoft Learn Terms; DeepLearning.AI Terms; fastai/fastbook; n8n-docs).

The content the founder wants (AI, automation, agents) is also where the permissive supply is. Google's developer-site content, including the 12-module Machine Learning Crash Course with its "Intro to Large Language Models" unit, is CC BY 4.0 with Apache-2.0 code (Google Developers Site Policies; Google MLCC). Microsoft's Generative AI for Beginners (21 lessons) and ML for Beginners (26 lessons, 52 quizzes) are MIT-licensed on GitHub and already translated into Russian (microsoft/generative-ai-for-beginners; microsoft/ML-For-Beginners). Hugging Face's NLP/LLM course and Agents course are Apache-2.0 with Russian translations (huggingface/course; huggingface/agents-course). Kaggle Learn's notebooks are Apache-2.0, Anthropic's prompt-engineering and tool-use courses repo is MIT (archived read-only on 15 September 2026, so frozen but usable), and Exercism's problem-specifications (over a hundred exercises, each shipping a description, metadata and canonical test data) are MIT (Kaggle/learntools; anthropics/courses; exercism/problem-specifications). CS Unplugged is CC BY-SA, usable with the share-alike obligation (CS Unplugged). No surveyed source has a Kyrgyz translation, which means a Kyrgyz-language AI curriculum built by translating the MIT, Apache and CC BY corpus from its Russian versions is both legal for a commercial platform and a first-mover asset nobody else holds.

The two entry disciplines the founder specified, typing and reading, need almost no third-party content. Monkeytype is GPL-3.0 and already ships word lists for Kyrgyz (~300 words), Kazakh, Uzbek and Russian; keybr is AGPL-3.0 with no Central Asian layouts, and the AGPL's network clause makes Monkeytype the lower-friction base, provided the modified engine is released as a separately served component (monkeytypegame/monkeytype; kyrgyz.json; aradzie/keybr.com). The Kyrgyz list is thin and must be replaced with a frequency-ranked list of several thousand words built from an open Kyrgyz corpus. The harder problem is the keyboard itself: Windows ships a Kyrgyz layout, macOS does not, and on iOS the letters ө, ү and ң appear only after long-pressing о, у and н, so a Kyrgyz typing challenge must first teach the layout, and Kyrgyz leaderboards must be kept separate from Russian ones because long-press latency depresses words per minute (kbdlayout.info; latysh/kyrgyz-keyboard-layout-mac; Kloop). Reading-speed tests need only original or CC BY passages in Kyrgyz and Russian, a 5–10 question comprehension check, and effective-WPM scoring (raw speed times comprehension), against an adult average of 200–250 WPM (ReadMinutes; ScholarWithin).

Duels and challenge banks have equally clear boundaries. Advent of Code is "free to use, not free to copy" but explicitly permits linking "even in commercial contexts"; Codeforces allows republishing problem texts with attribution but forbids publishing them "in open sources supporting automatic testing (such as online judges or similar resources)"; Project Euler is NC (Advent of Code About; Codeforces blog entry 967; Project Euler). The legal duel pattern is to show a link and title, have players submit a proof artefact (an accepted-submission URL verified through Codeforces' public API, an Advent of Code star), and score on-platform; Exercism's MIT test cases are the one bank that can be imported, translated and auto-graded outright. For data duels, filter Hugging Face and Kaggle datasets to cc0, cc-by-4.0, odc-by, pddl, cdla-permissive-2.0, apache-2.0 and mit, and exclude anything nc, nd, unknown or other (Hugging Face Hub licences; Kaggle licences discussion). YouTube embedding of creators like Nate Herk is permitted, but YouTube's terms forbid selling sponsorships on any page where YouTube content "is the primary basis for such sales", so embedded video must supplement a sponsored challenge, never be it (YouTube Terms of Service).

The founder's three prerequisites verify at different depths. Google OpenID Connect returns email and email_verified, and the absence of an hd claim indicates a consumer rather than Workspace account; GitHub OAuth with read:user plus the public REST API yields account age, repositories and recent commits for a genuine "builder" badge; LinkedIn's self-serve OpenID Connect returns only name, picture and email and, in LinkedIn's own words, "does not verify user identities and should not be marketed as such", so the LinkedIn gate proves only that an account exists (Google OpenID Connect; GitHub OAuth scopes; LinkedIn Sign In with OpenID Connect).

Finally, the free-content model's economics are visible in two Form 990s and one near-death. freeCodeCamp earned $2.99M in 2024 against $1.65M of expenses, 46.9% from donations and 44.2% from royalties (YouTube and publishing), and holds $9.75M in reserves: a free top-of-funnel monetised by third parties, not learners (ProPublica: freeCodeCamp). Exercism, with 2M users and 45M submissions, nearly ran out of money in 2024 on $7,500 a month of donations; its founder's diagnosis was that "keeping something free for everyone relies on either the user being the product, or on significant donations" (Exercism blog). A Central Asian league should not plan on donor income at all; the sponsor and employer are the "user who is the product", which is why NC content is incompatible with the model.

White space is real: nobody runs an open-age ranked league, and the partners already exist

The competitor scan across Kyrgyzstan, Kazakhstan, Uzbekistan and Russian-language platforms found many bootcamps, state mass-coding programmes, school olympiad pipelines and one-off hackathons, and no entity that combines open-age entry, persistent rankings and sponsor-funded prizes on one platform. The closest analogues are each missing a leg: Yandex Cup 2025 was open-age with a 12M RUB pool and 21,000 participants but is an annual event, not a league (Yandex); Decentrathon and HackAlem AI are open-age with real money (₸16M; 2,415 participants and a Guinness attempt) but episodic (Astana Hub; Inform.kz); the IT Cup of Kyrgyzstan is open to "everyone interested", drew 500+ participants and is run by KRSU with the Federation of Sports Programming and HTP, but on ICPC rules with small prizes (KRSU; KRSU); cpfed.kz runs a ₸3M championship (cpfed); Codeforces and Stepik rank without prizes. The nearest thing to an Elo-rated league in the region, MindZan, proved the mechanic works for schoolchildren's mental arithmetic across 46 schools and 3.4M problems, and sold for $10,000 in April 2026, a signal of demand rather than a competitor (Digital Business).

In Kyrgyzstan the paid alternative is a classroom. Twenty-two Bishkek schools charge 7,000–30,000 som per month and none publishes placement rates; Geeks claims 2,000+ graduates employed since 2018 and charges 19,000 som/month for a three-month QA course (Economist.kg; 2GIS). The free alternatives are course-shaped: Beeline has run free IT courses for citizens 17+ since 2021, Yandex's school graduated 130 teenagers in July 2026, a "One Million Coders" Kyrgyzstan with Udacity content launched in October 2025, and the state's Kyrgyz-language edtech (Bilim Bulagy, kitep.edu.gov.kg, BilimAi) is K-9 only (24.kg; Economist.kg; Kaktus; edu.gov.kg). The elite pipeline is small but better than the earlier market memo implied: Kyrgyz schoolchildren won four medals at IOI 2026 (20th of 97 countries) and six golds at the Eurasian Olympiad in Informatics, almost all from Sapat, Seitek, Akylman and similar lyceums, the olympiad culture the founder wants to open to everyone else (Gazeta.kg; OANA). All six gaps the research tested (Kyrgyz-language adult content, mobile-first competition, league-shaped free tier, community, verifiable credential, cross-country rankings) are open in Kyrgyzstan and only partially filled in Kazakhstan (events) and Uzbekistan (certificates).

The partner map is the asset Kyrgyzstan does have. The Federation of Sports Programming KR and olympiads.kg supply the "national ranking" legitimacy and already co-run the IT Cup with HTP and KRSU; the Esports Federation (ESF.KG) is national-team partner for the $20M Esports Nations Cup 2026 and runs student leagues, the most institutionally recognised "league" template in the country (esf.kg). HTP gives access to its residents as sponsors and employers; GDG Bishkek's DevFest reaches 1,000+ developers a year and is openly seeking sponsors (DevKG); Beeline has a four-year free-course record and sells a distance-learning data tariff, the only zero-rating precedent found (beeline.kg); AUCA, KRSU and Ala-Too host the hackathons and own the student bodies. Cross-border, Astana Hub's Silkway Accelerator explicitly recruits Kyrgyz startups and its Tech Orda accreditation is the Kazakh revenue door; IT Park Uzbekistan's residency rule for education companies is the Uzbek one; Daryn.online, at ~$3M revenue and 20,000 monthly learners across Kazakhstan, Uzbekistan, Azerbaijan and Kyrgyzstan, shows a Central Asian edtech can operate in four countries (Economist.kg; Digital Business). Bootcamps are ambivalent: partners as prize and talent sponsors, competitors for paid-learner wallets.

Two names in the founder's brief need a plain statement of what is publicly verifiable. On Jyldyz Academy, the co-owner's LinkedIn states that the "Kelechek" project provided free HTML/CSS/JS and web-design training to 3,000 students from across Kyrgyzstan and an earlier social project trained 130; Instagram shows ~16,000 followers; the academy's websites were unreachable (DNS not found, parked, or missing); and no public source for "15,000 learners" or "9,000 women trained" was found (LinkedIn: Olzhobai Iskander; Instagram). Those larger figures may be true, but until the academy publishes cohort data they should be presented to sponsors and investors as "3,000+ verified; 15,000/9,000 internal", and the discrepancy should be resolved before it appears in a deck. On Aitim, the entity found is AITIM Holding LLC, a Bishkek consulting and digitalisation firm, AmCham member and registered HTP resident that says it has "supported over 500 companies", with a sister AITIM Group (est. 2015, Bishkek and Chicago) serving US trucking companies, and a founder, Timur Tilenov, who also runs "AITIM Awards"; no public link between AITIM and edtech or competitions was found (AmCham: AITIM Holding; AmCham: AITIM Group; NY Weekly). That profile makes Aitim a credible anchor sponsor and, through its HTP residency and US client base, a plausible first demand-side partner for placing automation talent. But a Kyrgyz business group's prize budget should be assumed to be in the tens of thousands of dollars a year, consistent with every in-market benchmark, and the co-founder question should be settled on the basis of what Aitim brings beyond cash: HTP access, US clients, and legal entity.

Design lessons: rated loops retain, creators acquire, cash corrodes

The strongest precedent for an "intellectual sport" at mass scale is chess, and its anatomy is instructive because the founder's UFC framing gets it backwards. Chess.com ended 2025 with 243M members, 8.7M daily active users and 2.5 billion games in a single quarter; its growth engine is participation, not spectating (Chess.com Q4 2025 Board Report). Watchability was marketing into that loop: PogChamps put Twitch streamers with $50,000–100,000 pools and grandmaster coaches on screen, hit 375,000 concurrent viewers (within 2x of the 613,000 record for the actual World Championship), and Chess.com credited it alongside The Queen's Gambit (62M households, a ~500% jump in sign-ups, 100,000 new players a day) with a dramatic increase in registrations (Wikipedia: PogChamps; Chess Watch; CNN). The design insight is that amateurs with stakes and an admired expert coaching them turn a skill most viewers lack into relatable drama. Two costs come with a rating that carries status: Chess.com handled 3M cheating reports and closed 367,900 accounts in one quarter, so anti-cheat must be in the architecture from the first season; and it runs youth in a separate ChessKid product (14.3M registered, no adult chat) while sharing one rating system; "one ladder, segmented social surface" is how open-age works in practice (Chess.com Q4 2025).

Esports supplies the cautionary half. FaZe Clan went from a $725M SPAC valuation to a ~$17M sale; the Overwatch League's city-franchise model was dismantled in 2023; and ESL FACEIT Group's own executive concluded that "everyone was trying to mimic traditional sports… that might not have been the winning formula" because audiences follow creators, not teams (Esports Trade Association; Esports Insider). What survived is publisher-controlled ecosystems, sovereign-backed operators, and the one non-sponsor revenue line that scaled: Riot's Valorant Champions skin bundle made $16M+ with a cut to teams (Naavik). Central Asia is not a viewing market: the 2024 PUBG Mobile Central Asia qualifier peaked at 10,704 viewers, against 4.1M for Indonesia's MLBB league (Esports Charts; Esports Charts). So Arena's "alternative to UFC" should be understood as an alternative for the fighter, not the fan: a participation sport where Central Asians are players, with creator-led showcase events as marketing and a season finale as the only live spectacle. Do not sell team slots, and do not build the product around an appointment live show.

Competitive programming and Kaggle show that status alone sustains participation. Codeforces runs rated rounds roughly weekly with colour-coded tiers (Pupil, Specialist, Expert, Master and up), four rating-gated divisions and a "hack" mechanic, averaging more than 11,000 participants per round across 1,000+ rounds with essentially no prize money (Wikipedia: Codeforces). ICPC selected 140 teams from 63,294 students at 3,307 universities in 2025 (ICPC fact sheet). Kaggle sustains a 23.29M-account community in which only 612 people have ever reached Grandmaster; its five-tier progression runs across four tracks (Competitions, Datasets, Notebooks and Discussions), so the 99% who never medal still advance by teaching and sharing, and its Playground tier gives beginners a low-stakes lane (Wikipedia: Kaggle; Kaggle Progression). The warning from this world is Code Jam: 300,000 engineers a year, twenty years old, shut down within months of Google's January 2023 layoffs because its only rationale was a single sponsor's hiring (Pragmatic Engineer). Competitive programming also never became watchable because problems are illegible to viewers in real time; chess solved that with evaluation bars and commentary, and Arena will need its own legibility layer, with live scoreboards and a host who explains the task, if anything is to be streamed.

Trivia shows what dies and what lives. HQ Trivia's live, one-stream, cash-split format peaked at 2.3M concurrent players and collapsed; QuizUp's 100M users evaporated; asynchronous, always-available formats (Trivia Crack at 600M downloads, Kahoot! at 12 billion cumulative participants) endured (productmint; Quizzy; Wikipedia: Trivia Crack). The regionally native template is older and better than any of them: Что? Где? Когда? and Brain Ring are a fifty-year-old post-Soviet team sport with a formal MAK rating, "synchronous" tournaments in which the same question pack is played simultaneously in many cities and pooled into one rating, Uzbekistan's Zakovat TV show since 2001 and a national championship since 2004, and a 2022 Central Asia Cup with 48 teams from four countries (ЧГК wiki: Рейтинг; Hook.report). Six-player teams solve the "individual humiliation" problem of open-age competition, and AUCA's Brain Ring for Kyrgyz 11th-graders reveals the locally valued prize: the three finalists win tuition scholarships of 100%, 80% and 75%, not cash (AUCA). The World Cube Association adds the cheapest federation design in the survey (volunteer delegates, standard regulations, no age or gender divisions, parental consent under 18, one global ranking), and it already runs national championships in Bishkek and Kazakhstan (WCA Regulations; WCA: Kyrgyzstan Nationals 2026). Nitro Type's 7M monthly students and 37,000 teachers came from safety defaults (no chat, teacher-controlled matchmaking) plus a virtual economy (Nitro Type; Nitro Type news).

The retention mechanics with measured effect sizes all come from Duolingo. Weekly XP leagues of similar-activity peers with promotion and demotion zones raised total learning time 17% and tripled the number of highly engaged learners in A/B tests; users who reached a 10-day streak showed sharply reduced dropout; and a 21% improvement in current-user retention over four years drove 4.5x DAU growth (Lenny's Newsletter / Jorge Mazal). In June 2026 a single streak-restoration event drew 15.4M learners, nearly 8M of them with no active streak, and retention hit an all-time high of 84% (Duolingo Q2 2026). The caveat Duolingo itself reports is that leagues mostly moved already-engaged users, so they are a retention tool rather than an acquisition tool. Acquisition in this category comes from one-voice creators: Nate Herk went from a first video in September 2024 to 230,000 subscribers in nine months and over a million by 2026 on a weekly tutorial cadence (Apify blog; nateherk.com). Skool Games, which made community owners' net new MRR the score with $30,000+ first prizes and Hormozi dinners, appeared paused by July 2026; when the scoreboard is revenue, the league selects for aggressive selling (SkoolPrep; RevenueGeeks). The GTM plan's "solo voice first" decision is the single founder choice most strongly confirmed by the evidence.

Translated into a build list: an Elo/Glicko rating with colour tiers (Codeforces); weekly-reset peer leagues of about thirty similar-activity players with promotion and demotion (Duolingo); streaks with freezes; multi-track medals that reward helpers and content creators (Kaggle); synchronous "same pack, many venues" team tournaments pooled into one rating (ЧГК); delegate-run local competitions with one national ranking and no age divisions (WCA); a separate youth surface with no open chat (ChessKid, Nitro Type); amateur-with-coach showcase events featuring Kyrgyz and Russian-language creators (PogChamps); a physical season finale rather than a live-show core; and anti-cheat from day one. Avoid franchise slots, cash split thinly across many winners, dependence on one sponsor's hiring budget, appointment-only live formats, and any rating that lacks integrity tooling.

Recommended strategy: keep the league, change the prize, the platform and the market order

Most of the founder's vision survives contact with the evidence. Keep the open-age rated league with free entry: it is the only format with legal cover in all three countries and the only one every durable precedent shares. Keep sponsor-funded prizes, but as marketing and as the sponsor's wedge into employing, never as the engagement core. Keep the trilingual commitment, the Flutter mobile-first build, the typing and reading entry disciplines, the FIFA-style ranking, medals, challenge history and portfolio, the season-plus-one-off cadence, the learning-encouraged community, the founder-voice weekly challenge, and the end-state of jobs and startups. Keep the Gmail and GitHub gates as real signals; keep the LinkedIn gate as an existence check and say no more. Keep "an army of AI creators" as the mission, because the demand data (Upwork AI skills +109%, n8n the most-requested tool) says it is pointed at the right skill.

Change seven things. First, ship Android, and ship it first: the current iOS-only TestFlight path excludes 73–84% of Kyrgyz and 82% of Uzbek phones and selects for exactly the Bishkek elite the mission is not about; Flutter makes this a build-and-signing task, not a rewrite (StatCounter; StatCounter Uzbekistan). Second, redesign the prize: lead with tuition, laptops, internships, HTP placements and Upwork onboarding, which are the prizes with demonstrated local value (AUCA scholarships, Beeline's laptops, Decentrathon's job fair); restrict cash to verified adults, pay it from a Kyrgyz entity withholding 10%, and keep every random element away from anything of value so Article 5(8) never triggers. Third, replace the content plan: build the hosted curriculum on Microsoft's MIT courses, Hugging Face's Apache courses, Google's CC BY pages, Exercism's MIT test banks and Kaggle Learn, using their Russian versions as the pivot for a Kyrgyz translation that nobody else owns; deep-link to OCW, CS50 and Khan and never host or translate them. Fourth, re-read the Skool reference: take the community mechanics (weekly cadence, free tier, levels) and leave the price architecture; a 0.8% conversion at $129 is a Western creator's economics, and in Kyrgyzstan the paying "member" is the sponsor and employer, not the learner. Fifth, move the consumer tier: re-base Altai Pro from $9/month at 10% of MAU to $2–5/month at 1–2%, or reserve a premium tier for Kazakhstan where paid EdTech is a $200M market. Sixth, plan Uzbekistan for year two as a Telegram mini-app with Uzbek UI, because Kyrgyzstan is 18% of the serviceable market and Uzbekistan has already proven 2.5M-person elasticity for free coding. Seventh, and most important, build the demand side from Season 1: sign HTP residents, Aitim's US clients, IT Park exporters and a handful of Western automation buyers as "first look at the top 20" partners now, because every precedent says the ranking only becomes a business when someone is waiting to buy its top.

The table below records where the evidence disagrees with the founder's working assumptions, so that a co-founder or sponsor sees the corrections in one place.

The sequencing that follows is dated from October 2026 and sized to the evidence rather than to ambition. Season 0 (months 0–6, Kyrgyzstan): Android and iOS builds live; founder-authored weekly challenge in Russian with Kyrgyz UI; typing and reading disciplines on a Monkeytype fork with a new Kyrgyz frequency list and separate ky/ru ladders; Exercism-based coding duels and Microsoft/Hugging Face AI modules in Russian; Elo rating, weekly leagues, streaks and anti-cheat from launch; prizes in kind only (devices, tuition, internships from Aitim, HTP residents and Beeline); legal opinion from Kyrgyz counsel on Article 5(8) and prize tax; a Kyrgyz legal entity ready to act as tax agent. The GTM plan's own threshold (2,000–5,000 weekly active competitors and 40%+ weekly participation) is the right one, and the research's year-one envelope of 25,000–60,000 registrations is the ceiling to measure against. Season 1 (months 6–18): first sponsor packages at a 15–25% services fee plus branding tiers, with Kazakh sponsors courted from the start; first adult cash prizes paid locally with 10% withheld; a ЧГК-style synchronous team tournament with university and lyceum venues as WCA-style delegates; a "first look at the top 20" recruiter pilot with HTP residents and Aitim's US clients, priced at HackerRank-like seats; a creator-led showcase event in the PogChamps mould with Kyrgyz-language streamers; the Kyrgyz AI curriculum translation published as an open asset. Year 2: Uzbekistan via a Telegram mini-app and Uzbek UI, aiming at the One Million Uzbek Coders alumni and President Tech Award culture; application for IT Park residency on the graduate-placement rule; Kazakhstan launch with iOS parity and a Tech Orda accreditation application; Country Cup standings across the three. Years 2–3: a staffing arm placing ranked automation talent with Western clients at $40–100/hour and 30–50% gross margin, which is the line that can reach $10M+ ARR and which, per every comparable, is the only one with a path past $100M. Year 5: 1.5–3M registered, 300,000–700,000 monthly competitors, $1–5M of sponsor pools, and a talent business whose revenue, not its prize fund, determines whether the billion-dollar question is still open. Kill criteria matter as much as targets: if Season 0 cannot reach 2,000 weekly competitors with in-kind prizes alone, cash will not fix it; if no employer will pay for first look at the top 20 by month 18, the ranking has no buyer and the strategy must change before Uzbekistan, not after.

Conclusion

The research reframes Arena's question from "can a league be worth a billion dollars" to "what is the league for". Every mass intellectual sport that worked (chess, competitive programming, Kaggle, ЧГК) retained people with rating and status, acquired them through named personalities and cultural moments, and treated prize money as a marketing expense of about one percent; every one that bet on prizes, spectacle or a single patron is gone. The durable value Arena can build is a verified, ranked, trilingual dataset of Central Asian builders at a moment when the region's governments are spending real money to produce exactly that (Tech Orda, IT Park, HTP's 50,000-specialist goal) and when global demand for automation skills is doubling annually. The company that owns that dataset and sells access to it (to employers, exporters, Western clients and eventually the AI labs that made Turing worth $2.2B) is the only version of Arena the evidence can connect to a billion-dollar outcome, and the odds are still low single digits.

What is new in this analysis is how tightly the constraints fit together. The licence problem, the lottery problem, the app-store problem and the monetisation problem all point to the same design: free entry, skill-only outcomes, sponsor-funded prizes paid locally to adults, permissively licensed content translated into Kyrgyz, and a demand-side buyer signed before the ranking is asked to carry value. Kyrgyzstan is small enough to get this right in a year and too small to be the answer; Uzbekistan is where the users are and Kazakhstan is where the money is, and the sequencing has to respect both. If the founder and sponsor act on nothing else, they should act on three items in the next quarter: ship Android, resolve the Jyldyz Academy numbers, and sign the first employer who will pay to see the top of the ladder.