Monolithic vs microservices PAM: choose the boundary, not the fashion
For iGaming operators, the right player account architecture depends on where change, failure and audit evidence need to be contained.
A ledger, replayable ETL and a disciplined exception queue can turn wallet reconciliation into a controlled operating process.
A wallet operator that holds customer balances in several fiat currencies and digital assets has to answer two questions repeatedly: does its internal ledger balance, and does that balance agree with the relevant bank, payment provider or chain data? The hard part is not producing a daily report. It is making differences visible quickly without silently changing the record that explains them.
This is a practical reference design, not a report of a named customer deployment. Autonix Lab describes its work as designing, building and integrating business-critical systems, including payment infrastructure and tailored B2B software. In a wallet project, that kind of work can connect the ledger, data pipeline and operational controls. The exact components depend on the operator’s custody model, transaction volume and accounting policy.
Use a transactional ledger database that can commit a complete posting atomically. Represent every movement with balanced debit and credit entries. A customer deposit, for example, should affect the appropriate customer liability and settlement or custody account; a fee should be recorded as its own movement, not hidden in a net amount. Keep the original transaction reference, asset, amount, event time, posting time and source system with the entries.
Do not use a mutable wallet balance as the primary record. Derive balances from posted entries, or maintain a balance projection that can be rebuilt from them. This costs more storage and requires careful indexing, but it makes corrections auditable: a reversal or adjustment adds new entries instead of rewriting history.
Use fixed-precision decimal representations for fiat and asset amounts. Store the asset identifier and its permitted precision with each posting. Do not treat USD, EUR, BTC and a token as interchangeable numeric units. Rates used for valuation belong in a separate, timestamped record; they should not alter the underlying quantity held.
Collect events from wallet services, payment providers, bank statements and chain observers into an immutable landing area. Preserve the source payload or a faithful normalized copy, along with its origin, retrieval time and stable source identifier. That gives operators a way to distinguish a missing upstream event from a mapping bug in their own pipeline.
Run batch ETL jobs to validate, normalize and map those records to ledger accounts. Schedule jobs by source and reconciliation window rather than forcing every feed into one cadence. A bank file may arrive once a day; internal wallet events may arrive continuously; chain observations may need confirmation rules before they count as settled. Keep both event time and ingestion time, because late arrivals are normal.
Make each job safe to retry. Use a stable key built from the source and source event identifier, and enforce uniqueness at the point where events become ledger postings. Track the last successfully processed position or time window, plus a checksum or count of records read and accepted. A failed run should resume or replay without creating duplicate postings. The same principle underpins durable state and idempotent retries in transaction pipelines.
Compare internal totals with external evidence at distinct levels: per asset and account, per provider or custody venue, and per transaction where identifiers allow it. A daily control can compare opening balance plus posted movements against closing balance. Separately, match individual deposits, withdrawals, fees and settlement batches. Aggregate agreement alone can conceal offsetting errors.
For digital assets, distinguish observed, confirmed and settled activity according to the operator’s policy. A chain reorganization or delayed confirmation can change the evidence available after an initial match. For fiat, settlement timing, provider fees and statement cutoffs can create expected timing differences. Record the rule and source behind each status rather than treating every unmatched item as a ledger defect.
An exception tool can be a queue backed by a database table or an operational case system; the important parts are the data and ownership. Each item should carry the account and asset, amount, source references, first-seen time, reason code, severity, assigned owner and resolution evidence. Useful categories include missing internal posting, unmatched provider transaction, amount mismatch, duplicate event and timing difference.
Automate flags, not explanations. Rules can group repeated failures, prioritize material differences and suppress known timing cases until a deadline. They should not close an exception merely because two totals happen to net out. Require a recorded decision for adjustments, and retain the original exception and its resolution trail.
Set a freshness target for each source and measure ingestion delay, reconciliation lag, unmatched value, exception age and replay counts. Alert on a broken feed or a growing backlog, not just on a failed job. Run periodic rebuilds or independent ledger checks to test that projections still match posted entries.
The trade-off is straightforward. More frequent reconciliation shortens the time a defect can hide, but increases compute, provider dependence and operational noise. A batch-first design is easier to reason about and replay; it may be too slow for products promising near-real-time balance assurance. Start with explicit freshness and escalation rules, then tighten them where the business and source data support it.
A sound multi currency wallet reconciliation stack is therefore not one database or one dashboard. It is a ledger with immutable postings, ETL that can be replayed safely, source evidence that survives transformation, and exception handling that assigns decisions to people. Build those boundaries first. Tune throughput after the numbers can be explained.
For iGaming operators, the right player account architecture depends on where change, failure and audit evidence need to be contained.
No verified release or pricing moves stand out; the practical work is making agent decisions traceable, retrieval measurable and personal data governable.
A queue can absorb live-event surges, but durable bet state, idempotent retries and clear acceptance rules keep the pipeline honest.