The public forecast log: live (non-backtest) forecasts, newest first, with their question text and resolution when one exists. A forecast whose question was later voided stays in the log (the forecast is part of the record) and carries question_status = "voided" plus the recorded voided_reason so it is never mistaken for an open call. question is the exact wording that was forecast and timestamped; title is a reader-friendly rendering of the same question (see foresight_api.titles) and falls back to the text.

entity_kind is entity.kind for the question's entity — the same column /v1/entities/{id} already publishes — and is null when the question has no entity (a market question) or the entity row is gone. It is here because entity_id alone cannot tell a dependency question from one about a company: an SEC Form D filer has a real entity_id exactly like tox-dev/tox does, and only the kind separates them ("company" vs "repo"). A client that is about dependencies -- the homepage's two lists are -- filters on it; see SOFTWARE_ENTITY_KINDS in static/index.html.

This endpoint itself does NOT filter. The archive is append-only and public by design, and an API that silently drops rows from the log is a worse answer than a page that curates its own two lists: the record has to stay complete for anyone checking it. The field is added so the curation can happen where the curation claim is made.

resolved=true returns only forecasts whose question has an outcome, newest resolution first. That slice is what lets a reader check the record against what happened -- including the calls we got wrong, which are in it by construction rather than by the luck of a window ordered on issue date.