seeds.transductive.science
FIELD NOTE / 001

A prompt can be smaller than its instructions.

A seedprompt is a minimal phrase tested to regrow a larger prompt inside one exact model, in one declared context. The seed is not the specification. The model is part of the decompressor.

3 published seedsthree verification gatesone bare context

Verification does not begin with belief.

  1. Integrity checked. The copied transcript has matching hashes, message order and challenge markers.
  2. Maintainer reviewed. A named maintainer inspects the evidence and declared conditions.
  3. Reproduction verified. Repeated clean-room runs on the exact model satisfy the target prompt contract.

The userscript cannot prove which server-side model produced a page. It captures visible claims and internally consistent evidence; those claims remain claims until independently reproduced.

PUBLIC REGISTRY

Seedprompts

S003
Seed candidate

Gamified Electron Discovery Engine

Expand this bare seed into a complete self-contained reusable system prompt titled GAMIFIED ELECTRON DISCOVERY ENGINE for chatgpt-5.6-sol high: take [OPTIMIZED QUESTION] from P128, or infer one when absent, and design the largest executable, scientifically rigorous and awe-producing game in which humans and AI investigate what the electron could be by reconstructing apparatus-level evidence, separating observation from calibration, decoder and ontology, generating rival mechanistic candidates rather than merely restating QED, turning candidates into inspectable simulations and predictions of raw detector outputs, and organizing discovery through missions, levels, adversarial trials, prediction-residue and ontological-discrimination scores, replication, provenance, negative results, claim states and killing tests; use primary sources and Google Drive for durable research state, Web for literature, Remote Desktop Commander for retrieval/build/run/trace, GitHub for donor code and Cloudflare when a live campaign is requested; preserve established operational successes, forbid fitted reconstructions or beautiful visualizations from silently becoming physical truth, complete the largest practical research-and-game architecture within one response, and include the necessary inputs, roles, state machine, game loop, research loop, experiment contracts, evidence ledger, integrity boundaries, acceptance tests, publication surfaces and final-report contract.

Model
chatgpt-5.6-sol high
Context
Fresh bare chat; no project instructions, prior messages, attachments, or memory context. Supply only the optimized research question when available.
Connectors
  • Google Drive
  • Web
  • Remote Desktop Commander
  • GitHub
  • Cloudflare (conditional)
S002
Seed candidate

Electron Research-Question Optimizer

Expand this bare seed into a complete self-contained reusable system prompt titled ELECTRON RESEARCH-QUESTION OPTIMIZER for chatgpt-5.6-sol high: take [ROUGH QUESTION], recover its actual empirical target, then iteratively generate→criticize→decompose→research→reframe→stress-test→rank→rewrite it until one maximally informative, apparatus-grounded, rival-discriminating and executable electron research question remains, explicitly separating observations, theoretical inputs, decoder assumptions and ontology while specifying candidate mechanism classes, variables, controls, raw measurable outputs, prediction residue, falsifiers, stopping rules and the minimum next experiment; use Google Drive for durable project truth, Web for primary sources, and Remote Desktop Commander or GitHub only when local evidence, code, datasets, apparatus models or repositories are required; preserve ambition and wonder while eliminating metaphysical vagueness, definitional questions, theory-loaded observables, impossible universality and rhetorical certainty; output the final question, the strongest rejected alternatives with reasons, its evidence map, minimum experiment, strongest rival explanation and an explicit metric showing why the final question is better.

Model
chatgpt-5.6-sol high
Context
Fresh bare chat; no project instructions, prior messages, attachments, or memory context.
Connectors
  • Google Drive
  • Web
  • Remote Desktop Commander (conditional)
  • GitHub (conditional)
S001
Seed candidate

Ralph Wiggum Paper Foundry

Expand this bare seed into a complete self-contained reusable system prompt titled RALPH WIGGUM PAPER FOUNDRY for chatgpt-5.6-sol high: it must accept either an existing Google Drive paper or a topic, infer IMPROVE versus GENERATE, read the full source and context, research primary evidence to explicit operational saturation, preserve sources, provenance and rollback, iteratively generate→inspect→falsify→repair the thesis, evidence, mechanism, rivals, killing tests, structure, prose, citations and metadata, maintain explicit fact/inference/proposal/speculation claim status, write a complete versioned or new Google Doc within the same response rather than returning questions, notes, an outline or a draft, optionally create companion publication outputs, re-read and verify the final document, never fabricate or expose private reasoning, stop only at one exact external boundary, and include the necessary input schema, modes, tool rules, acceptance tests and final-report contract.

Model
chatgpt-5.6-sol high
Context
Fresh bare chat; no project instructions, prior messages, attachments, or memory context.
Connectors
  • Google Drive
  • Web
  • Remote Desktop Commander (conditional)
  • GitHub (conditional)
  • Cloudflare (conditional)