← Methodology page · Trust & Transparency · Validation data
Version 0.2 · September 2026 Platform: https://www.gailabai.com
GaiaLab is a research-use software platform for turning a biological question, gene panel, and stated context into an inspectable hypothesis record. It brings together retrieved source material, scored and filtered outputs, cited evidence where available, uncertainty, limitations, and proposed next research questions.
GaiaLab does not establish biological truth, causal validity, clinical utility, safety, efficacy, or patient benefit. It is not a medical device and must not be used for diagnosis, treatment selection, or therapeutic recommendation. Every output requires independent scientific and experimental validation.
This document describes the public technical and evidentiary boundary of the platform as of September 2026. It deliberately distinguishes implementation controls and computational checks from scientific validation.
Biological research questions often require a researcher to reconcile disparate source types: gene annotation, pathway memberships, interaction networks, literature, drug-target records, and trial registries. GaiaLab is designed to make this synthesis process more inspectable.
The intended output is not a conclusion. It is a research record that can be reviewed, challenged, refined, or rejected by qualified researchers.
A typical analysis proceeds through the following stages:
1. Input record — a gene panel and biological context are supplied. 2. Normalization and retrieval — gene symbols are normalized where possible; configured data sources are queried with partial-failure handling. 3. Aggregation and scoring — returned records are organized into evidence channels and ranked using declared computational rules. 4. Evidence presentation — cited records, source context, contradictions, missing information, and limitations are surfaced where available. 5. Optional structured AI synthesis — a deployment-dependent synthesis layer may organize the retrieved record into hypotheses and questions for human review. 6. Record preservation — analysis identifiers, snapshots, and selected verification artifacts are retained where the relevant feature is enabled.
Source availability varies by deployment, API credentials, upstream service availability, and source-specific access conditions. Missing or unavailable sources must not be represented as supporting evidence.
GaiaLab’s accountability standard is that scientific claims should increasingly arrive with receipts.
For an inspectable record, the platform aims to preserve:
- the input panel, stated context, and recorded source-availability state;
These records support inspection, reproduction attempts, and critique. They do not establish correctness, biological validity, clinical usefulness, ownership of a discovery, or independent validation.
For GaiaWorld pre-reveal benchmarks, the public record may additionally include declared protocols, freezes, scoring locks, prediction commitments, and published scoring artifacts. A pre-commitment helps make a later evaluation auditable; it does not itself establish a model’s scientific merit.
Citation links and automated checks are aids to review, not proof.
- Retrievability checks: identifiers such as PMIDs can be checked against an external index. A non-returned identifier is unresolved or potentially invalid; it is not automatically proof of intentional fabrication.
Where source records or citations are absent, outputs should be interpreted as hypotheses rather than database-backed assertions.
Some deployments can use a configurable, multi-role AI synthesis component to organize a retrieved record into a hypothesis, critique, evidence summary, risk considerations, and synthesis. The number of roles, provider routing, feature flags, and availability are deployment-dependent.
This component is not an independent scientific review panel, does not constitute multiple independent validations, and does not make its output scientifically correct. It may be unavailable, reduced to a single-provider path, or omitted entirely. Any generated synthesis remains subject to the evidence and limitations shown in the record and to independent expert review.
GaiaLab can retain analysis identifiers and snapshots that record inputs and selected outputs. Reproduction of an analysis may still differ when external databases change, APIs fail, model/provider behavior changes, source licenses change, or configuration differs.
Some prediction commitment batches are organized as SHA-256 Merkle roots and can be timestamped through OpenTimestamps and Bitcoin. A cryptographically verified receipt establishes that the committed bytes existed no later than the verified Bitcoin block time. It does not establish the exact internal timestamp, scientific correctness, clinical efficacy, ownership, or that the committed prediction was made before every potentially relevant external event.
See the public verification page for the exact proof boundary: https://www.gailabai.com/verify.
Engineering checks can test documented behavior such as deterministic controls, input handling, replay behavior, hashes, declared scoring operations, and abstention. These checks are necessary for trustworthy infrastructure but are not evidence of biological validity.
GaiaWorld benchmark results are retrospective, bounded evaluations with experiment-specific protocols and limitations. They must be read from their published records, including negative results, source rejections, abstentions, and pre-reveal eligibility closures. A positive score on one retrospective boundary does not demonstrate general biological prediction ability, prospective performance, causal validity, or clinical utility.
- The platform is limited by the quality, scope, licensing, and availability of its underlying sources.
If referring to GaiaLab, cite the specific analysis or benchmark record when possible, together with its date, identifier, source boundary, and limitations.
*This document reflects GaiaLab v0.2 as of September 2026. For current implementation details and public research records, see the repository, Methodology page, and GaiaWorld registry.*