# Dredge Echo Astra — Product Hunt submission draft

## Product name
Dredge Echo Astra

## Tagline
Ask once. Dredge until the answer holds.

## Short description
Adaptive reasoning that builds the workflow your question deserves. Watch Dredge deepen an investigation, challenge conflicting evidence, preserve useful work when you redirect it, and return a conclusion with an inspectable evidence trail.

## Maker’s first comment — Fi
Hey Product Hunt, I’m Fi, the builder behind Dredge Echo Astra.

I kept coming back to one question: how should intelligence organize itself when the answer is not obvious?

That question became Dredge.

The work starts with what the problem needs. When evidence conflicts, the investigation deepens. New branches appear. Competing explanations get their own space. When I change direction, useful work stays with us. And when the evidence holds, Dredge knows when to bring the machinery back down.

That ability to reorganize is the heart of what I’m building.

In the demo, a national launch checklist looks ready—but a regional acceptance requirement challenges the conclusion. You can keep the broad investigation or redirect it toward Missouri. Watch what survives, what retires, and how the answer changes with its scope.

The demo is a transparent recorded run of the actual orchestration runtime using sample evidence and deterministic providers. It does not make live Astra calls. The separate live integration is implemented, with model-access validation still ahead.

My vision is an intelligence layer for answers that have to survive scrutiny. I want the evidence, the uncertainty, and the changes in direction to remain inspectable.

Ask once. Dredge until the answer holds.

I’d love to hear where you would put that to work—and what an answer would need to show before you trusted it.

— Fi

## What makes this different

- The controller chooses reasoning effort and orchestration shape from observed signals.
- A mutable execution graph turns those choices into actual units of work.
- Conflict can add independent hypothesis construction, challenge, and arbitration.
- User steering preserves unaffected evidence and invalidates conclusions that depend on retired work.
- Evidence convergence can collapse the graph and reduce reasoning effort.
- The response carries claims, uncertainty, evidence state, graph history, and decision telemetry.

## Astra’s role

The isolated Astra implementation uses the OpenAI Responses API with per-call reasoning effort. Dredge assigns different efforts to scouting, investigation, challenge, judging, and arbitration. The default configured model is `gpt-6-astra`.

Do not claim a successful live Astra API execution until it has been observed. The hosted replay uses deterministic providers and is clearly labeled. This draft must retain that disclosure unless the demo is replaced with a verified live experience.

## Demo gallery plan

1. **Conflict changes the shape:** graph expansion with the conflicting regional requirement visible.
2. **New direction. Useful work stays:** Missouri steering with technical evidence preserved and national work retired.
3. **A smaller launch. A stronger conclusion:** scoped evidence check and grounded conclusion.

Capture these from the actual demo after visual review. No screenshots are fabricated or attached yet.

## Challenge facts checked September 11, 2026

Source: https://www.producthunt.com/contests/gpt-6-astra-challenge

The page says “Build something ambitious with GPT‑6 Astra,” lists September 18, 2026,
and links to the Product Hunt launch form with the contest identifier. It advertises
five winners, each receiving one year of ChatGPT Pro, $10,000 in API credits, and
promotion by OpenAI.

The rendered page also returned a zeroed countdown; do not interpret that as a
verified precise cutoff or closure. Timezone, full eligibility terms, judging rubric,
and any contest-specific questionnaire were not available in the page text checked.
Reconcile those in the live submission flow before final submission.

Submission entry: https://www.producthunt.com/posts/new?ref=contest_preview&contest=gpt-6-astra-challenge

## Submission readiness

Prepared: product positioning, tagline, description, maker comment, interactive
recorded demo, voiceover script, and reproducible event export.

Still needed before an honest final submission: confirm the complete challenge
requirements; validate live Astra access if required; review the demo visually;
make the reviewed demo accessible to judges; record/upload a demo video if the
launch form requires one; and complete the actual Product Hunt launch form.

Nothing has been submitted to Product Hunt. The hosted review copy is private.
The Nebius submission and deployment remain untouched.