DecisionGraph traces the evidence behind AI decisions, blocks stale approvals, and requires a fresh review when data changes. Try the no-signup recall demo. Built with GPT-6 Astra in Codex; open source, with DataHub integration.

What real task does your product handle with GPT-6 Astra?

Hi Product Hunt — I'm Thomas, the maker of DecisionGraph. An approval can be perfectly reasonable when it is made and become wrong when its evidence changes. DecisionGraph keeps the evidence behind a decision, blocks reuse of stale approvals, and creates a replacement that needs fresh human approval. The earlier record stays available for comparison. Try the guided demo: calculate a fictional retail recommendation, approve it, change the forecast, and watch the stale approval get rejected. Then review and approve the replacement. No signup, purchases, external writes, or model calls occur in the browser sandbox. For this challenge, I used GPT-6 Astra in Codex to build the guided walkthrough, improve the landing page, add lifecycle checks, and implement an experimental structured evidence reviewer with citation validation. The reviewer cannot approve decisions or change the ledger. Live API verification remains pending after HTTP 429 responses; no successful Astra API run is claimed. This is a revision of my existing open-source project. Its DataHub core and linked integration video predate this challenge. The repository documents the new work and the earlier recording's provenance. Where would stale approvals cause the most trouble in your workflow?

What real task does your product handle with GPT-6 Astra?

DecisionGraph

Give AI decisions an evidence trail and a recall button

Give AI decisions an evidence trail and a recall button