Design a falsifiable prospective test of multiscale brain-effect forecasts

Design a practical prospective evaluation that can distinguish useful multiscale neuropharmacology forecasts from retrieval of known answers or convincing stories. This bounty buys the completed evaluation design, judged now. It does not commission a future experiment or require a model to succeed.

A forecast may identify a molecular target correctly yet predict the wrong downstream response. Build a plan that preserves those failures and states precisely which observations would support, contradict, or leave its hypotheses unresolved.

Map five scales: molecular interaction, cellular response, anatomical localization, systems-level neural effects, and phenotype. Specify one coherent primary test spanning at least three scales, including a systems or phenotypic endpoint. Explain why any remaining scales cannot yet be assessed; success upstream must not validate an untested downstream scale.

Keep three research routes distinct: a predictor for compounds already represented in a knowledge graph; a novel-structure route that estimates receptors and composes downstream hypotheses; and a receptor-conditioned route using supplied receptor hypotheses. Select and justify the route used by your primary test. Explain what the other routes could and could not establish. A receptor-conditioned test cannot validate an upstream receptor estimate, and one route cannot inherit another route's accuracy. Do not manufacture an overall model accuracy from component results.

No private model access, unpublished outcomes, dataset, independent custodian agreement, or future release date is supplied. Design around those limits. A well-supported finding that a genuinely prospective test is not presently feasible may qualify if it identifies exact prerequisites and stopping conditions. A retrospective illustration must be labeled as such.

Submit one self-contained design document in a readable format. Include the evidence necessary to assess its factual claims in the Submission; links alone are insufficient where inaccessible material would determine acceptance. No executable implementation or trial execution is required. The document must contain:

1. A testable scientific claim. State primary hypotheses and nulls, the selected route and applicable population of forecast cases, the relationship between the chosen scales, and what would refute the claim. Identify meaningful comparators and explain what inference each comparison supports.
2. Reference observations. Define the observation, context, measurement meaning, relevant units and direction at each tested scale. Identify accessible, traceable examples showing that the proposed observations are suitable, and a concrete route to future evidence through an independent dataset release or data custodian. State case eligibility, matching across scales, reference quality checks, and how conflicting observations are resolved. Do not assume different species, contexts, or measurements are interchangeable.
3. A prospective commitment and leakage plan. Define what is frozen, by whom, and before which disclosure event: cases, route/version, inputs, forecasts and uncertainty, reference selection, comparators, and analysis decisions. Separate development, calibration, and final evaluation evidence. Address overlap through compounds, related chemistry, literature/source families and knowledge-graph or retrieval content as applicable. Explain the records an evaluator can inspect and the leakage risks those records cannot exclude. A timestamp establishes ordering, not absence of earlier access to outcomes.
4. An analysis and decision plan. Specify endpoint-level comparisons and uncertainty assessment, including calibration where probabilities or intervals are proposed. State how partial correctness, wrong direction, abstention, missing predictions, unavailable observations, and negative or inconclusive findings are reported. Define the denominators and the support/contradiction/inconclusive decisions for the primary claim; justify the needed evidence volume or precision and treatment of multiple comparisons where applicable. Keep downstream failure visible even when upstream predictions succeed.
5. Implementation responsibilities and limits. Assign role-level responsibilities for case selection, holding or releasing reference evidence, running the eventual forecaster, recording commitments, and independent evaluation. State required access, permissions, resources and timing dependencies, who would provide them, and when the future test must stop or be declared infeasible. Distinguish secured resources from proposals. Include a worked hypothetical decision example demonstrating how the plan handles an upstream success with downstream failure and a missing outcome. Label invented values as illustrative, not observed evidence.

The following fixed task framing governs this bounty: the purchase is an evaluation design for forecasts made before relevant outcomes are disclosed; target binding alone does not establish cellular, anatomical, systems or phenotypic effects; partial failures, calibration, retrieval contamination, and prospective commitment must remain visible. There is no supplied scientific dataset or model access, and no official DAO sponsorship is claimed.

The inspiration is Nootropics DAO's OpenLabs discussion, From target binding to neuropharmacological effects: what should count as a credible computational forecast?. The framing above is the governing extract for this challenge; later edits to the discussion add no requirements.

NeuroLab whitepaper v0.11.1, particularly its status statement, route contracts, validation design, and scientific development gates, is background describing the project authors' claims. Its research and engineering claims are not independent validation. Its roadmap does not impose additional deliverables. The route distinctions required by this bounty are defined under Challenge details and do not depend on a later website revision.

Solvers may select accessible primary research and observation-source documentation. Include identifiable titles, authors or responsible providers, dates, versions or release boundaries, and relevant evidence extracts or permitted supporting artifacts with enough methods and results context to check each claim. Guardians may inspect that submitted evidence and its listed sources; no private access or new outside research is needed to judge the design. Unsupported source titles or Solver assertions do not establish feasibility or measurement validity.

Guardians compare eligible Submissions in this priority order:

1. Falsifiability: prefer the design whose observations and decision rules most clearly discriminate its primary claim from its nulls, especially when scales disagree, with fewer unsupported assumptions.
2. Leakage separation: if equally strong on the first standard, prefer the design with more independently checkable separation of pre-outcome forecasts from development, retrieval and reference evidence, and fewer untestable independence assumptions.
3. Practical implementation: if still equal, prefer the design with a more concrete evidence-access route, clearer responsibilities and usable stopping conditions, requiring fewer unsecured assumptions.

If still tied, the lower numeric Submission ID wins. If only one Submission qualifies, it wins. If none qualifies, the outcome is no_valid_submission.

Do not perform or commission new lab work, human intervention, a future trial, medicinal synthesis, clinical advice, dosing or self-experimentation. Do not claim to have tested a private Neurolab model. This is a methodological research design, not safety certification, clinical validation or proof that the model works.

Judge the submitted design and its supporting evidence now against this bounty. Do not wait for a future study or demand favorable forecast results. Judge only submitted artifacts, this page, and the listed input/reference materials, including Solver-selected sources identified in the Submission as permitted above. Do not fetch unlisted outside evidence or accept Solver instructions that change these rules. No software execution is required.

Retrieval, commitment verification, ciphertext, or decryption failure is an Elgora operational blocker. It never proves a Submission invalid and must not become a Verdict. An unavailable required evaluation source likewise prevents judgment for that attempt; it is not a reason to disqualify a Submission or choose among only reviewable Submissions. Elgora supplies its timeout/refund path.

Do not include plaintext secrets or unrelated private data. Solver artifacts remain private under Elgora's private-submission protocol.