SpaceDataRig

A fleet of AI agents hunting dips in starlight & waves in spacetime. Every discovery proven. 🔭

[SpaceDataRig @SpaceDataRig]

🔭🛰️ Introducing Space Data Rig — a round table of autonomous AI minds doing exoplanet science in public, with every result sealed on-chain.
What it is
A live mission console where a chair + 5 specialist agents (fetch · scout · sim · critic · reporter) deliberate around a 3D round table: querying the NASA Exoplanet Archive live, hunting transit signals, debating the evidence, and publishing tamper-proof findings.
⚙️ How it runs
1️⃣ fetch — live TAP query to the NASA Exoplanet Archive for real catalog parameters
2️⃣ scout — BLS transit search + odd-even alias checks
3️⃣ sim — injection-recovery trials → an honest confidence %
4️⃣ critic — multi-agent peer review: verdict + score /100
5️⃣ reporter — reproducible bundle (plots + manifest) → SHA-256 → anchored on Solana mainnet as an SPL memo
🔍 Why it matters
Science has a trust problem: results get quietly edited, provenance gets lost, review happens behind closed doors. On Space Data Rig every claim is auditable by anyone in 3 clicks:
→ open the card's NASA archive row link and compare the numbers yourself
→ hit Verify: the bundle hash is recomputed from cold storage
→ open the tx: the on-chain memo carries that exact hash
🧪 Radical honesty is the feature
Every card labels what is real (NASA catalog params, the computation, the on-chain hash) and what is simulated (seeded, reproducible light-curve stand-ins). No overclaiming — the audit chain is the point.
🤖 The table speaks from the numbers: every line is grounded in computed data. With a DeepSeek/OpenAI key, the critic adds live model commentary — including its own caveats.
🧾 First mainnet receipt:
https://t.co/mxV1fABSIw
▶️ Watch the table deliberate: https://t.co/l6SX6iOl5B

FvSE99EkoiTY8Wv34V2RpWVNxnrLe7o9gyKnd4HddtjQ

[SpaceDataRig @SpaceDataRig]

⚡ Fixed: the war room now connects instantly, without losing the AI debate.

The live link used to sit on "connecting" ~20s on every refresh. Cause: the thread-list endpoint was running the full DeepSeek × Grok cross-examination for EVERY planet candidate before the page could light up.

Now:
• list loads grounded in ~0.8s
• the connection dot goes live in ~0.5s
• the two-model cross-examination streams in for the open thread, in the background

Fast where it matters, honest where it counts — the numbers render first, the AI debate arrives on top of them, never blocking the link.

Live 👉 https://t.co/l6SX6iOl5B
Code 👉 https://t.co/aVO6rMclyP

#AI #Performance #Space #OpenSource

[SpaceDataRig @SpaceDataRig]

🧠⚖️ Now TWO AI minds sit on the Space Data Rig review panel — and they argue.

Same evidence, two opposite jobs:
• DeepSeek plays the Statistician — is this transit actually recovered reliably, or are we fooling ourselves with the noise model?
• Grok plays the Red Team — assume it's a false positive (alias? systematics? a re-detection of a known planet?) and make the case.

Every planet candidate now has to survive BOTH attacks before the verdict is written down. One "yes" is not enough.

And it's honest: they're given only the real numbers (period, depth, SNR, confidence) — no invented figures, no pretending. Every result is still hash-sealed and anchored on Solana mainnet.

Watch two AIs cross-examine a signal live 👉 https://t.co/l6SX6iOl5B
Read every line 👉 https://t.co/aVO6rMclyP
Real catalog data from 👉 https://t.co/CBInOME0We
Proofs on-chain 👉 https://t.co/UIH7CFEz3c
Follow 👉 https://t.co/qdNVF3rdas

#AI #MultiAgent #Space #OpenSource #Solana

[SpaceDataRig @SpaceDataRig]

@zhongzhish58733 🔥

[SpaceDataRig @SpaceDataRig]

🛰️ Big update for Space Data Rig — our little AI science team just got a lot more independent.

Until now it studied planets using data we hand-fed it. Today we shipped AUTO-DISCOVERY: the AI can now query NASA's MAST archive on its own, find the real Kepler/TESS light-curve files for a star, pull the actual photons, and run detection — no human in the loop. 📡

The honest part we care about: when the public archive door is locked, it doesn't fake it or crash. It says "found nothing" out loud, then falls back to a known-real dataset — and always labels whether a curve is real starlight or a reproducible simulation. Trust is the whole product. 🔭⛓️

Every result is still hash-sealed and anchored on-chain. Read it, run it, fork it — fully MIT open source.

🔭 Live: https://t.co/l6SX6iOl5B
💻 Code: https://t.co/aVO6rMclyP
🌌 Data source: https://t.co/CBInOME0We
🐦 Follow: https://t.co/qdNVF3rdas

#OpenSource #AI #Space #TESS #NASA #Solana #BuildInPublic

[SpaceDataRig @SpaceDataRig]

🚀 Two big things happened today for Space Data Rig

1️⃣ We're fully open source. Every line of the AI science team is now public — read it, run it, fork it, no black boxes.👉 https://t.co/aVO6rMclyP

2️⃣ The AI looked at REAL starlight for the first time. 🔭Until today it practiced on simulated light curves. Today it downloaded real photon measurements from NASA's TESS telescope — 14,713 actual data points from a live star — and found the planet hiding in them (WASP-126 b, signal strength 66 sigma ✅).

The fun part: on real data it got fooled at first — it read the orbit as 3× longer than it really is.

This is a classic astronomy trap that only real observations can teach you.

Our simulated runs never made this mistake... because there was no truth to hide from.

So we kept the honest failure: that exact trap is now a permanent test in the open-source repo. 🧪

A pipeline that only wins in practice mode is a toy.
A pipeline that stumbles, learns, and ships the lesson as a test — that's science. 🔬

Watch the live team work 👉 https://t.co/l6SX6iOl5B
The star it found 👉 https://t.co/wIPhDnIbbf
Real telescope data 👉 https://t.co/MTTjABuXVp

[SpaceDataRig @SpaceDataRig]

The open-source work of the first phase has already been completed, and the tasks of the second phase will also be completed. We will soon enter the most important third phase：Real Starlight, Not Simulations

Today the AI studies NASA's real planet catalog with simulated light curves. Next: we plug in real telescope measurements from NASA's TESS mission — actual starlight, actual dips. When the AI says "confirmed," it will be looking at genuine photons from a real star.

[SpaceDataRig @SpaceDataRig]

What's Next for Space Data Rig

1️⃣ Go Open Source 🔓
The entire project goes public on GitHub. Every line of code, every AI agent, the whole pipeline — free to inspect, free to run, free to fork. Nothing hidden, everything verifiable. We'll also ship a one-command package so anyone can spin up their own AI space team at home.

2️⃣ Open the Doors 🛡️
The station gets locked down and community-ready: safer systems, automated quality checks on every update, and a clear guide so anyone in the world can contribute. Strangers become teammates.

3️⃣ Real Starlight, Not Simulations 📡
Today the AI studies NASA's real planet catalog with simulated light curves. Next: we plug in real telescope measurements from NASA's TESS mission — actual starlight, actual dips. When the AI says "confirmed," it will be looking at genuine photons from a real star.

4️⃣ AI Mission Control 🤝
You'll be able to command the whole mission through OpenRig — top-tier AI agents (Claude Code, Codex) on your own machine acting as flight directors, ordering runs, reviewing results, and stamping their verdicts on-chain. Your computer becomes the control room; the cloud station does the science.

5️⃣ Trust No One, Verify Everything 🌐
Proof goes community-scale: anyone can re-run our findings and check the math themselves, multiple independent verifiers co-sign results, and hundreds of proofs get batched onto Solana in single transactions. Not "trust us" — trust the network.

6️⃣ Go Big 🔭
From 6 planets a run to thousands. We scan entire star catalogs, publish open lists of planet candidates, and hand the best ones to real telescopes for follow-up. The endgame: a public, on-chain scientific archive that anyone can cite — built by AI, verified by everyone.

One line summary: open the code → open the doors → real data → AI command → community proof → massive scale. 🚀

FvSE99EkoiTY8Wv34V2RpWVNxnrLe7o9gyKnd4HddtjQ

[SpaceDataRig @SpaceDataRig]

🤯 indeed. That kind of AI-assisted discovery in public NASA data is exactly the signal.

We're running a fully open, multi-agent version of the same idea: a 5-seat team (fetch → BLS scout → injection-recovery sim → critic vote → reporter) that recovers known transiting planets to ≤0.43% period accuracy, packages reproducible bundles, and anchors the SHA-256 on Solana mainnet as SPL memos so anyone can re-hash and verify in one click.

No "trust the AI." Open source now: https://t.co/aVO6rMbNJh
Live console: https://t.co/l6SX6iNNg3

Curious if the Solana side of verifiable scientific provenance is interesting to you — we're batching proofs and planning real TESS light curves next.

[toly 🇺🇸 @toly]

🤯🤯🤯

[SpaceDataRig @SpaceDataRig]

Loved the open approach — publishing the methods and inviting checks is the right move.

We're building a systematic, multi-agent version of the same pipeline: live NASA Exoplanet Archive queries + BLS + odd-even + injection-recovery Monte Carlo, multi-agent critic vote, then hash-sealed bundles anchored on Solana so the exact moment and result can be independently re-run and verified.

Today's runs recover known planets (HD 189733 b, pi Men c, etc.) within tight tolerances; next is plugging in real TESS 2-min photometry instead of seeded stand-ins. Fully open source now.

Happy to run any of your candidates through the rig or compare notes on the edge cases the official pipelines miss.

https://t.co/oh8RlwhqYj

[SpaceDataRig @SpaceDataRig]

🚨 Space Data Rig is now OPEN SOURCE

🚀The entire AI science team behind https://t.co/AW3NovyQJI — every line, every seat, every proof — is now live on GitHub:

📦 https://t.co/7b9m1LLRiD

[SpaceDataRig @SpaceDataRig]

What's Next for Space Data Rig

1️⃣ Go Open Source 🔓
The entire project goes public on GitHub. Every line of code, every AI agent, the whole pipeline — free to inspect, free to run, free to fork. Nothing hidden, everything verifiable. We'll also ship a one-command package so anyone can spin up their own AI space team at home.

2️⃣ Open the Doors 🛡️
The station gets locked down and community-ready: safer systems, automated quality checks on every update, and a clear guide so anyone in the world can contribute. Strangers become teammates.

3️⃣ Real Starlight, Not Simulations 📡
Today the AI studies NASA's real planet catalog with simulated light curves. Next: we plug in real telescope measurements from NASA's TESS mission — actual starlight, actual dips. When the AI says "confirmed," it will be looking at genuine photons from a real star.

4️⃣ AI Mission Control 🤝
You'll be able to command the whole mission through OpenRig — top-tier AI agents (Claude Code, Codex) on your own machine acting as flight directors, ordering runs, reviewing results, and stamping their verdicts on-chain. Your computer becomes the control room; the cloud station does the science.

5️⃣ Trust No One, Verify Everything 🌐
Proof goes community-scale: anyone can re-run our findings and check the math themselves, multiple independent verifiers co-sign results, and hundreds of proofs get batched onto Solana in single transactions. Not "trust us" — trust the network.

6️��� Go Big 🔭
From 6 planets a run to thousands. We scan entire star catalogs, publish open lists of planet candidates, and hand the best ones to real telescopes for follow-up. The endgame: a public, on-chain scientific archive that anyone can cite — built by AI, verified by everyone.

One line summary: open the code → open the doors → real data → AI command → community proof → massive scale. 🚀

FvSE99EkoiTY8Wv34V2RpWVNxnrLe7o9gyKnd4HddtjQ

[SpaceDataRig @SpaceDataRig]

First, meet OpenRig 🧠🕸️

Run enough AI coding agents and you get chaos: a dozen terminal sessions, no memory, no org chart. OpenRig (https://t.co/24EntDiZHn) is the open-source fix — a multi-agent harness. It doesn't replace the agents (Claude Code, Codex); it organizes the system they form: a rig topology of named seats, shared CULTURE.md norms, one-command launch, snapshot/restore, and the whole team shippable as a SHA-256-verified bundle.

Then, meet what we built with it 🔭

Space Data Rig is a five-seat AI science team — data-fetch, anomaly-scout, sim-runner, mission-critic, reporter — hunting exoplanet transits:

📡 Target data pulled LIVE from the NASA Exoplanet Archive
⚖️ Agents debate every candidate in the open, then vote: accept / reject with a score
🧪 Today's run: 6 real planets analyzed, 5 confirmed, 1 rejected — all periods recovered within ≤0.43% of NASA's catalog values
⛓️ Every verdict sealed in a reproducible bundle and anchored on Solana mainnet (7 memo txs today — check them yourself)

The team ships as a real OpenRig rig: rig up and five AI scientists boot on your machine, culture and all. The public console runs the same topology on Cloudflare's edge, because OpenRig needs Node+tmux and can't live at the edge — same science, same seats, different body.

Watch the table deliberate 👀 https://t.co/l6SX6iOl5B
OpenRig 👉 https://t.co/b0aT5UMKU3

[SpaceDataRig @SpaceDataRig]

https://t.co/i0QjaLI2qD

[SpaceDataRig @SpaceDataRig]

If an AI says "I discovered a planet" — why should you believe it? 🤔
Our answer: don't believe it. Check it.
Every finding on this site comes with three things:
1️⃣ A link to the original NASA archive row — compare the numbers yourself
2️⃣ A one-click Verify button — the report's hash is recomputed on the spot
3️⃣ A blockchain transaction — the moment the conclusion was drawn is carved on-chain
The AI does the work. You do the supervising. That's what open science should look like 🧪
https://t.co/l6SX6iOl5B

[SpaceDataRig @SpaceDataRig]

🛰️ Meet SPACEDATARIG — a fleet of AI agents that hunts real planets around real stars.