Changelog

never answer one read model's request with another model's read on the same dataset

health lists each dataset's freshness, Runtime Cache usage is counted, v1 usage is exact

• /api/health reads a per-dataset freshness rollup from the snapshot index and
• Every Runtime Cache read, hit and write is counted per key namespace and
• /v1/usage reads the last 24 h through keyset pages on (created_at, id): past

serve the pump-dump and insider-selling boards through one read model

• One signal read model behind the free and v1 routes of both boards: the
• A fresh board is cached 600/1,200 s, a stale one 30/30; a failed read is 503
• The stale bound is 1 hour everywhere: API, v1 freshness_sla_seconds (3600),

serve the /global overview through one read model

• Both /global overview routes run on one model: the crypto snapshot while it
• A fallback to an old snapshot is flagged stale and no longer cached (was
• Registry prose, docs, the contract and OpenAPI describe the new fallback.

serve correlation, market-cap search, price prediction and coin compare through read models

• Eight research read models behind the 11 routes; one CoinGecko client for all
• Unknown correlation ids: partial success with `unresolved` on both routes
• The coin picker's hour-cached CoinGecko list gets its own dataset
• Outages answer 503 problem+json on the free routes (v1 502); contracts,

score crypto assets only, from each venue's primary contract, and give dropped coins no score

• The pump-dump and insider-pressure scorers are engine metrics pump_dump.score
• Their 72 h and daily funding inputs read each venue's primary crypto
• A coin that stops scoring is written with score NULL and its sub-scores and

one freshness bound for listing events, and rollup reads that cannot lose buckets

• The events payload marks the persisted rows partial once the newest completed
• The hub's weekly and monthly rollup reads are ordered by their unique key and