fix: an unknown exchange rate is unknown, not zero - #8

Fixes #7.

buildPriceInfo fell back to multiplying by zero when the token was not in the exchange rate cache:

priceUSD = roundTo(priceUSD != null ? priceUSD
    : price * (cache.containsKey(token) ? cache.get(token) : 0));

So a price in a token we have no rate for was published as a confident $0.00 — indistinguishable from a genuinely free item. _PriceInfo consumers already render a null usd as "n/a" (see BaseTransformer), so null is the value this should produce; the convention already exists and this converges onto it.

Scope, measured

49 activity rows in soonmarket_realtime_event carry a non-zero price and a zero usd in a token with no exchange rate at all — FOOBAR (21), EASY (17), CYPHR (6), RDM (3), PIXEL (2).

The other 9737 zero-usd rows are not this bug and must not be "repaired" away. XPR, XUSDC and LOAN all have rates, so the fallback never fired for them. Their zeros come from roundTo defaulting to 2 decimals with amounts genuinely below half a cent — every one of the 9724 XPR rows is between 0.0001 and 5 XPR, and XPR is $0.00286. A bad cached rate would not respect that ceiling, which is what rules it out. This is spelled out in #7 along with two hypotheses that were tested and disproved.

The part that would bite a careless fix

roundTo unboxes its Double argument, so the derived values cannot simply be computed and rounded once priceUSD may be null. Three of them are now guarded:

• royaltyUSD and marketFeeUSD — were guarded on royalty/marketFee but not on priceUSD
• sellerReceivedPriceUSD — dereferenced priceUSD unguarded, so swapping the : 0 for null without touching this would have turned the fix into a new NPE on exactly the rows it is meant to fix

Also note cache.get(token) replaces containsKey(...) ? get(...) : 0. Besides being the actual fix, that removes an unboxing NPE if the map ever held a null value for a present key.

No repair

The 49 rows cannot be repaired — no rate exists for those tokens, so null is the correct value and this produces it going forward. The other 9737 are already correct.

Verification

This repo has no build of its own (CI checks line endings only), so it was verified through the consumers:

• event-processor-contract — mvn test passes with this source vendored in
• soon-market-api — mvn compile passes. Its mvn test fails locally on a Quarkus startup TimeoutException from the DB connection pool, which is an environment issue unrelated to this change

Consumers need the usual submodule bump afterwards; at minimum event-processor-contract and soon-market-api, which are the two that render prices.

🤖 Generated with Claude Code