RESEARCH PROGRAMME / FROM ORIGIN TO EVIDENCE
Preserve the idea.
Advance the experiment.
This site extends the recovered inscription and distributed-computation research into a new design question: how can an agent develop without losing a verifiable account of its origin and changes?
The threads that meet here.
In a March 2023 user-authored brief, Promptus content generation and NFT minting were linked with contributed Ubiquita compute, storage and network resources. December 2023 discussions gave Promptinals a more specific recipe-and-inscription form. The 2024 white-paper drafts explored reconstruction, model references and a proposed distributed overlay.
The present Genesis DNA framework is a new synthesis of those threads, focused on agent development. It is not a claim that the historic papers already implemented recursive agent evolution. The detailed source chronology and reconstruction questions are published on promptinal.com.
Four experiments worth doing next.
EXPERIMENT 01A portable genesis manifest.
Specify canonical bytes, required fields and dependency references. Ask independent implementations to calculate the same digest from the same manifest. Publish fixtures and failure cases, including missing dependencies and invalid encodings.
Evidence to publish: a schema, conformance suite and reproducible results.EXPERIMENT 02An agent with two descendants.
Start two instances from the same origin. Give each a different authorised task history. Compare the recorded memories and approved workflows while checking that their protected constraints remain unchanged.
Evidence to publish: lineage records, scoped tasks and an explicit divergence measure.EXPERIMENT 03A recursive change that earns adoption.
Let an agent propose a workflow modification. Evaluate it against a frozen baseline and held-out tasks. Include regressions, cost and adversarial inputs. Reject proposals that fail the complete acceptance policy.
Evidence to publish: proposals, evaluator version, raw task outcomes and decisions.EXPERIMENT 04Recovery after a bad branch.
Introduce a deliberately flawed change in a test environment. Demonstrate a stop, rollback and resumed operation. Confirm that history remains inspectable and that the version does not regain revoked tool access.
Evidence to publish: recovery traces, permission checks and retained event records.Use the right kind of proof.
A content digest checks an exact byte sequence against a known value. A signed assertion attributes a message to a key under its signature scheme. An inscription records content within Bitcoin’s transaction structure under the chosen indexing conventions. An evaluation measures a particular configuration against a particular test. Each answers a different question.
None alone proves that an agent is generally intelligent, uniquely conscious or legally entitled to every asset it can generate. A productive research programme states the proposition being tested and publishes enough evidence for another party to challenge it.
Primary technical context.
Bitcoin inscriptions
The Ordinals handbook explains inscription content and its association with satoshis. It supplies context for the record layer, not a claim that Bitcoin executes an AI model.
Read the Ordinals handbook ↗AI reproducibility
PyTorch documents limits to reproducibility across platforms and releases. A serious reconstruction experiment needs a declared environment and comparison rule.
Read the reproducibility notes ↗Rights and tokens
The US Copyright Office and USPTO’s NFT study examines intellectual-property issues. Token control and rights in associated content must be considered separately.
Read the joint study context ↗Companion research
Promptinal.com presents a newly authored synthesis of the recovered white-paper material, its provenance and unresolved technical questions.
Read the reconstruction paper ↗An invitation to build inspectable evidence.
We welcome technical collaborators interested in manifest specifications, agent evaluation, privacy-preserving lineage and reproducible workflows. The current public deliverable is a research site and local teaching lab. A production implementation, independent security assessment and validated AI test vectors remain future research work.
Start a research conversation ↗