Candle Service Fixes: Inclusive Timestamps, User Agents, and Discovery

Respecting Inclusive Timestamps

I fixed the Gecko client to respect its inclusive before_timestamp semantics (9596a616). An off-by-one on an inclusive boundary means you either re-fetch the same candle forever or skip one. I also fixed admin queue accuracy in the same commit—the queue view was reporting counts that didn’t match what the workers actually had pending.

Custom User Agents

I set a custom user agent on the HTTP clients and marked an upstream as exhausted when it returns errors (c0439b6c). A proper user agent is polite and often the difference between being served and being blocked. Treating repeated upstream errors as exhaustion stops a worker from hammering a dead endpoint.

Discovery and Drag Curves

I expanded the discovery queries and added a drag-curve worker plus API (e5291416). Drag curves track how a token’s price behaves after an initial move, which is useful for characterizing follow-through versus fade. Adding it as a worker means it runs continuously and exposes results through the same API surface as everything else.

API Examples

I updated the API docs to use an actual dev key in the examples (4a361094) so consumers can copy-paste a working call instead of a placeholder that fails.

Integration Work Is the Work

A data aggregation service lives or dies on its upstream integrations. Today was all boundary handling: inclusive semantics, rate limits, user agents, and error classification. None of it is glamorous, but it’s what makes the data trustworthy.