Skip to main content

Known limitations and next steps

AreaLimitationPossible next step
Tag coverageOnly a small share of the catalogue is tagged (65 of about 8,000 games at the time of writing), so "tagged alike" can only speak about those gamesMore AI tagging runs and review; the holdout test shows tags are the strongest lever
Tag typesTagged games are all slots, so tests on tagged games say nothing about table or live playersTag table and live games
Batch recencyThe batch profile decays by the last-played date, not per bet; heavy play 25 days ago plus one bet yesterday looks recentThe live path already fixes this with per-event decay
EvaluationAccuracy is measured on simulated customers whose rules we wroteRun the same holdout test on real activity
Player protectionA switch the recommender obeys, not a risk detectorFeed the list from the responsible-gambling system, possibly over the compacted suppression topic
LicensingNo country or licence rules; a game not allowed in a market could be recommendedAdd availability to the catalogue and filter candidates
CurrencyOnly EUR is acceptedConvert at ingestion or keep per-currency totals
Crowd scaleCo-play is computed in memory on a sample (250,000 customers by default)Batch job with ALS or sharded counting
Duplicate betsThe live path catches replays, not the same bet sent as two messagesBet-id dedup in the Lua script or upstream
Redis crowd samplingSCAN walks one nodePer-shard scan for Redis Cluster
Simulator writesPlain messages only; ingestions with a Schema Registry cannot be written by the simulatorRegistry-aware publisher