On-chain data · independently reproducible

Loot Survivor — player performance data

Descriptive statistics from 231,717 games played by 2,741 players on Starknet mainnet (Sep 2025 – May 2026). Each chart is rendered from data recorded on a public blockchain and links to a query that returns the same figures.

Data source and how to reproduce it

Every Loot Survivor game is a non-fungible token (NFT) minted on Starknet mainnet, and the final score of each game is written to the blockchain by the game's smart contract. The figures below are read directly from that on-chain record.

Method applied to every chart: official game minter only, standard ruleset (settings_id = 1). Figures are medians (the middle / typical value), which are not skewed by extreme outliers.

1 · Median score by player experience

Each game grouped by the player's experience at the time it was played — their 1st game, 2nd–5th, 6th–20th, and so on — shown against the median (typical) score in each group.

↗ Run this query on live on-chain data
Show the exact SQL
-- Loot Survivor — Claim 1: Median score by player experience
-- Source: https://api.cartridge.gg/x/pg-mainnet-10/torii/sql  (SQLite)
-- Each game is grouped by the player's experience AT THE TIME it was played
-- (i.e. where the game falls in that player's own chronological sequence), then we
-- take the median score per experience tier. This measures experience-at-time-of-play
-- and is monotonic; it avoids the confound of bucketing by a player's lifetime game
-- total (which mixes in selection effects and produces a spurious mid-range dip).
WITH g AS (
  SELECT ( (instr('0123456789abcdef',substr(s.score,7,1))-1)*17592186044416
         + (instr('0123456789abcdef',substr(s.score,8,1))-1)*1099511627776
         + (instr('0123456789abcdef',substr(s.score,9,1))-1)*68719476736
         + (instr('0123456789abcdef',substr(s.score,10,1))-1)*4294967296
         + (instr('0123456789abcdef',substr(s.score,11,1))-1)*268435456
         + (instr('0123456789abcdef',substr(s.score,12,1))-1)*16777216
         + (instr('0123456789abcdef',substr(s.score,13,1))-1)*1048576
         + (instr('0123456789abcdef',substr(s.score,14,1))-1)*65536
         + (instr('0123456789abcdef',substr(s.score,15,1))-1)*4096
         + (instr('0123456789abcdef',substr(s.score,16,1))-1)*256
         + (instr('0123456789abcdef',substr(s.score,17,1))-1)*16
         + (instr('0123456789abcdef',substr(s.score,18,1))-1) ) AS sc,
         ROW_NUMBER() OVER (PARTITION BY o.owner ORDER BY s.internal_executed_at) AS game_no
  FROM "relayer_0_0_1-TokenMetadataUpdate" m
  JOIN "relayer_0_0_1-TokenScoreUpdate"   s ON s.id = m.id
  JOIN "relayer_0_0_1-OwnersUpdate"       o ON o.token_id = m.id
  WHERE m.settings_id = 1 AND m.minted_by = '0x0000000000000006'
),
b AS (
  SELECT CASE WHEN game_no=1 THEN '1st game'
              WHEN game_no<=5 THEN '2nd-5th game'
              WHEN game_no<=20 THEN '6th-20th game'
              WHEN game_no<=50 THEN '21st-50th game'
              ELSE '51st+ game' END AS experience,
         CASE WHEN game_no=1 THEN 1 WHEN game_no<=5 THEN 2 WHEN game_no<=20 THEN 3
              WHEN game_no<=50 THEN 4 ELSE 5 END AS ord,
         sc
  FROM g WHERE sc BETWEEN 1 AND 3000
),
r AS (SELECT experience, ord, sc,
             ROW_NUMBER() OVER (PARTITION BY experience ORDER BY sc) rn,
             COUNT(*)     OVER (PARTITION BY experience) c FROM b)
SELECT experience,
       c AS games,
       ROUND(AVG(CASE WHEN rn IN ((c+1)/2,(c+2)/2) THEN sc END),1) AS median_score
FROM r GROUP BY experience ORDER BY MIN(ord);

2 · Skill persists: later score, grouped by skill in the first 10 games

Players with 20+ games are split into four tiers by their median score over their first 10 games, then each tier's median score is tracked across their later games. The tiers stay ordered and separated across 100+ games — they never cross. A game decided by chance would show the four lines converging.

r = 0.55correlation between a player's first-10-game median and their later-game median, across 763 players (0 = pure chance, 1 = fully determined by skill)
128 vs 181median score of the lowest vs highest early-skill tier, measured 70+ games later — a gap that holds the entire time
↗ Run this query on live on-chain data
Show the exact SQL
-- Loot Survivor — Claim: skill persists (skilled players consistently outscore weaker players over many games)
-- Source: https://api.cartridge.gg/x/pg-mainnet-10/torii/sql  (SQLite)
-- Players (>=20 games) are split into four tiers by their MEDIAN score over their FIRST 10 games,
-- then each tier's median score is tracked across later game windows. If outcomes were driven by
-- chance the tiers would converge; instead they stay ordered and separated across 100+ games.
WITH g AS (
  SELECT o.owner AS player,
         ( (instr('0123456789abcdef',substr(s.score,7,1))-1)*17592186044416
         + (instr('0123456789abcdef',substr(s.score,8,1))-1)*1099511627776
         + (instr('0123456789abcdef',substr(s.score,9,1))-1)*68719476736
         + (instr('0123456789abcdef',substr(s.score,10,1))-1)*4294967296
         + (instr('0123456789abcdef',substr(s.score,11,1))-1)*268435456
         + (instr('0123456789abcdef',substr(s.score,12,1))-1)*16777216
         + (instr('0123456789abcdef',substr(s.score,13,1))-1)*1048576
         + (instr('0123456789abcdef',substr(s.score,14,1))-1)*65536
         + (instr('0123456789abcdef',substr(s.score,15,1))-1)*4096
         + (instr('0123456789abcdef',substr(s.score,16,1))-1)*256
         + (instr('0123456789abcdef',substr(s.score,17,1))-1)*16
         + (instr('0123456789abcdef',substr(s.score,18,1))-1) ) AS sc,
         ROW_NUMBER() OVER (PARTITION BY o.owner ORDER BY s.internal_executed_at) AS gn,
         COUNT(*)     OVER (PARTITION BY o.owner) AS n
  FROM "relayer_0_0_1-TokenMetadataUpdate" m
  JOIN "relayer_0_0_1-TokenScoreUpdate"   s ON s.id = m.id
  JOIN "relayer_0_0_1-OwnersUpdate"       o ON o.token_id = m.id
  WHERE m.settings_id = 1 AND m.minted_by = '0x0000000000000006'
),
e AS (SELECT player, sc, gn FROM g WHERE n >= 20 AND sc BETWEEN 1 AND 3000),
em AS (  -- each player's median over their first 10 games
  SELECT player, AVG(CASE WHEN rn IN ((c+1)/2,(c+2)/2) THEN sc END) AS e10
  FROM (SELECT player, sc, ROW_NUMBER() OVER (PARTITION BY player ORDER BY sc) rn,
               COUNT(*) OVER (PARTITION BY player) c FROM e WHERE gn <= 10) GROUP BY player),
q AS (SELECT player, NTILE(4) OVER (ORDER BY e10) AS skill_tier FROM em),
later AS (
  SELECT q.skill_tier, e.sc,
         CASE WHEN e.gn<=20 THEN 1 WHEN e.gn<=40 THEN 2 WHEN e.gn<=70 THEN 3 ELSE 4 END AS later_window
  FROM e JOIN q USING(player) WHERE e.gn > 10 AND e.gn <= 120),
r AS (SELECT skill_tier, later_window, sc,
             ROW_NUMBER() OVER (PARTITION BY skill_tier,later_window ORDER BY sc) rn,
             COUNT(*)     OVER (PARTITION BY skill_tier,later_window) c FROM later)
SELECT skill_tier, later_window, c AS games,
       ROUND(AVG(CASE WHEN rn IN ((c+1)/2,(c+2)/2) THEN sc END),1) AS median_score
FROM r GROUP BY skill_tier, later_window ORDER BY skill_tier, later_window;

3 · First-games score: players new to the franchise vs. returning players

Median score over each player's first three games, comparing players with no prior Loot Survivor history against players who had previously played the original title (similar game mechanics).

↗ Run this query  ·  ↗ Nth-game query, returning players excluded
Show the exact SQL
-- Claim 3: Prior skill transfers — OG veterans outscore new players from game 1
-- Source: https://api.cartridge.gg/x/pg-mainnet-10/torii/sql  (SQLite)
-- COHORT-AWARE: separates genuinely-new players from OG "Loot Survivor" veterans
-- (contract 0x018108b3...618c4dc4). The og_veterans list below is the set of new-game
-- wallets that ALSO minted an OG adventurer, derived from on-chain mint events
-- (see scripts/enumerate_og_players.py). Exact-address match => a conservative LOWER bound.
WITH og_veterans(addr) AS (VALUES
  ('0x0550212d3f13a373dfe9e3ef6aa41fba4124bde63fd7955393f879de19f3f47f'
  ... (96 returning-player addresses embedded) ...