Design the evidence gates for a reproducible STAT6 compound shortlist

Deliver a decision-ready validation design for a STAT6 small-molecule screen: separate evidence of direct binding, cellular pathway modulation and disease-relevant activity, and specify how ambiguous or contradictory results affect advancement. Ground the design in two fixed public studies of AS1517499. This is a remote experimental-design deliverable; no experiment, novel binder or proprietary prediction is required or claimed.

The OpenLabs project combines LULA predictions and experimental testing and explicitly seeks a focused screening campaign, reproducibility checks, controls and validation methods. Its predictions and candidate structures are not public inputs here. The concrete decision is what validation evidence should be required before a predicted compound is called a STAT6 binder and advanced beyond a pathway-response screen.

Use AS1517499 as a worked reference-evidence case, without assuming that an author's designation as a selective inhibitor establishes direct binding or clinical efficacy. Compare the two published biological contexts below. The output must turn evidence gaps into a staged, implementable decision design; a literature summary alone is insufficient. Choose and justify appropriate assay technologies using additional publicly accessible primary methodological sources where needed. Do not invent new experimental observations or claim that this design has been validated.

• evidence.csv: an assay-level evidence matrix covering every labeled panel in Figures 1, 3 and 8 of Lee et al. and Figures 1, 4, 5 and 8 of the renal-fibrosis study. Record panel, biological system, perturbation, dose/time, comparator, endpoint, reported experimental unit and n, independence/replication information, qualitative result and source location. Panel groups may share a row only when these fields are genuinely identical. Report undisclosed fields as such. Classify what each result can establish about binding, target engagement, pathway activity and phenotype; do not invent exact numbers from a plot.
• validation-design.md: a staged design with an explicit initial screen, orthogonal confirmation, interference/cytotoxicity counterscreens, target-dependence validation and context-relevant functional confirmation. For every stage specify the decision addressed, assay principle, required reagents or sample types, positive/negative/vehicle controls, concentration/time rationale, independent versus technical replication, normalization, exclusion rules, batch reproducibility checks, advancement/hold/rejection criteria and the source or declared assumption behind each choice. The design must distinguish a measured affinity or target engagement from downstream signaling. Include foreseeable assay interference and biological alternative explanations, including how the two study contexts change interpretation.
• decision-rules.csv: machine-readable gates with IDs, stage, required measurements, pass condition, fail condition, indeterminate condition and resulting advance/hold/reject action. Conditions must be operational, not phrases such as “looks convincing.” Numerical thresholds are proposed design assumptions requiring rationale, not facts supposedly established by the two studies. Missing measurements cannot count as a pass.
• evaluate.py or evaluate.R, cases.csv, and case-results.csv: executable application of the proposed gates to at least eight clearly labeled hypothetical cases. Cover all pathway stages and include discordant binding/pathway evidence, cytotoxicity, assay interference, non-reproducible batches and missing evidence. The cases are logic tests of the proposed design, not scientific measurements or proof of sensitivity/specificity. At least one plausible progression, one rejection and one hold must be exercised. If the model has branches, include enough additional cases to exercise every terminal action and gate failure/indeterminate route.
• decision-brief.md and README.md: recommend the minimum defensible evidence package for a shortlist, identify which evidence is already supplied by the fixed studies and which must be collected, and list unresolved feasibility dependencies such as protein construct, assay availability, compound solubility and target biological context. Explain the most consequential alternative outcome and how it would change the decision. Include source citations, dependencies and one command to reproduce the rule evaluations. All required artifacts must be included as file bytes, not external links.

Fixed biological evidence is the public PMC article versions available on 17 September 2026:

1. Lee et al. (2022), *Inhibition of STAT6 Activation by AS1517499 Inhibits Expression and Activity of PPARγ in Macrophages to Resolve Acute Inflammation in Mice*, DOI 10.3390/biom12030447, PMC8946515: https://pmc.ncbi.nlm.nih.gov/articles/PMC8946515/ . Required panel scope: Figures 1, 3, 8 and their relevant methods/results. Other sections may clarify interpretation.
2. *Pharmacological Inhibition of STAT6 Ameliorates Myeloid Fibroblast Activation and Alternative Macrophage Polarization in Renal Fibrosis* (2021), DOI 10.3389/fimmu.2021.735014, PMC8426438: https://pmc.ncbi.nlm.nih.gov/articles/PMC8426438/ . Required panel scope: Figures 1, 4, 5, 8 and relevant methods/results. Other sections may clarify interpretation.

Additional primary methodological sources supporting the proposed assay choices must have a publicly accessible full methods description, cited by stable identifier and URL with exact section/figure. Record the accessed version/date. No paid data, private LULA output, undisclosed molecule structures, raw author-held data, animal work or laboratory execution is needed for this design task.

Project context: https://openlabs-git-codex-openlabs-elgora-adapter-bio-xyz.vercel.app/projects/48647efd-8f4f-4c87-a5bf-a1f61832ae05 . Independent contribution; no project-owner endorsement is claimed.

Only submissions satisfying every criterion are eligible. Prefer fewer material scientific or decision-logic errors, then the strongest defensible ability of the proposed controls and orthogonal measurements to resolve the stated alternatives, then clearer operational reproducibility and provenance. Extra assays without decision value do not improve rank. Remaining ties go to the earlier on-chain submission timestamp, then lower numeric submission ID. No winner is required if no submission meets the criteria.