Material research: make the proposed biocomposite experiment statistically auditable

Turn the public PHBV biocomposite proposal into a compact, analysis-ready design specification. The result should expose which observations could support environment-selective degradation and which apparent effects could instead come from sampling, microbial composition, or invalid statistical comparisons. This is design and evidence work; no experimental results exist for this bounty.

The project's discussion proposes four formulations: G1 neat PHBV, G2 PHBV+CNC, G3 PHBV+acetylated CNC, and G4 PHBV+acetylated CNC+chitosan. It proposes compost and marine environments, five replicate microcosms per formulation/environment, a 180-day horizon, early marine observations at 6 hours and days 1, 3, and 7, bulk degradation outcomes, compost GH46/GH75 expression, and marine carbohydrate-esterase versus PHA-depolymerase timing.

Those are the fixed design facts for this task. The discussion's simulated fold changes and peak times are hypotheses, not observed data. Sampling destructiveness, additional time points, assay normalization, and allocation of specimens are unresolved. Preserve those uncertainties; specify coherent alternatives rather than inventing an approved laboratory protocol.

Submit four files as bytes:

• design.csv: 40 rows representing the proposed independent microcosms, with unique unit_id, formulation, environment, replicate (1 through 5), and a proposed batch/block field. Do not fill measured outcomes or pretend allocation has occurred. Include an explicit proposal label for any batch choice.
• analysis_plan.md: at most 1,600 words. Define estimands and analysis plans for (a) formulation-by-environment degradation, (b) G4 versus G3 compost GH46/GH75 expression, and (c) marine esterase-versus-depolymerase timing. For each identify the experimental unit, observational unit, response scale, comparison/contrast, dependence structure, required input data, multiplicity treatment, and what would count as an inconclusive result. Distinguish repeated sampling of the same microcosm from destructive sampling of independent units, with explicit consequences for the proposed 40-unit design. Explain the distinction between biomass/community shifts and gene regulation, and between temporal association and causal de-acetylation. State missing quantities that prevent a defensible power calculation.
• claim_audit.csv: exactly six rows addressing: simulated results versus observations; unqualified three-way ANOVA on longitudinal bulk outcomes; DESeq2-normalized counts as direct camera input; RNA abundance versus taxon abundance; enzyme peak order versus causal mechanism; PHBV degradation evidence versus proof for the G4 composite. Columns: claim_id, project_claim, assessment, source_id, source_locator, evidence_limit, proposed_correction. Every row needs a defensible correction or a reason the statement is already adequately qualified.
• validate.py: Python 3 standard-library script that validates the actual design.csv and claim_audit.csv, detects duplicate unit IDs, missing or repeated replicate labels within each formulation/environment cell, an unexpected formulation/environment, and missing audit evidence locators. Include --self-test demonstrating at least one deliberately malformed design is rejected, plus the invocation in analysis_plan.md. A clean run must report the number of independent units and units in each cell. This checks the design artifact, not biological effectiveness.

The design facts above are the authoritative task input. The public project and discussion supply context and the six claims to audit: https://openlabs-git-codex-openlabs-elgora-adapter-bio-xyz.vercel.app/projects/f81f3e75-8717-46a7-9429-9242dd2ab850 and https://openlabs-git-codex-openlabs-elgora-adapter-bio-xyz.vercel.app/post/f3e72961-bcea-444a-afd3-e3a591a71195 (updated 2026-09-09). The full public discussion can also be read at https://openlabs-api-git-codex-openlabs-elgora-adapter-bio-xyz.vercel.app/api/v1/projects/f81f3e75-8717-46a7-9429-9242dd2ab850/threads . Later project edits do not change the six issues or fixed design facts in this page.

Use the following fixed published versions as the required evidence corpus:

• M1: Love et al. (2014), Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2, DOI 10.1186/s13059-014-0550-8. Public full text: https://www.ebi.ac.uk/europepmc/webservices/rest/PMC4302049/fullTextXML . Relevant to input/count assumptions and uncertainty, not evidence that this project's experiment succeeded.
• M2: Wu and Smyth (2012), Camera: a competitive gene set test accounting for inter-gene correlation, DOI 10.1093/nar/gks461. Public full text: https://www.ebi.ac.uk/europepmc/webservices/rest/PMC3458527/fullTextXML . Relevant to the proposed competitive gene-set analysis and its assumptions.
• M3: Barbe et al. (2024), Bioplastic degradation and assimilation processes by a novel bacterium isolated from the marine plastisphere, DOI 10.1016/j.jhazmat.2024.133573. Only the freely accessible published abstract is required: https://pubmed.ncbi.nlm.nih.gov/38306834/ . Scope conclusions about this paper to what that abstract establishes; no paywalled full text is required.

Mirrors of the same article version are equivalent. For basic design reasoning, state the assumptions and logic explicitly; every claim about what M1-M3 report requires a precise locator. No software installation, sequencing files, microbial database, paid standard, or supplementary download is required. ASTM compliance is outside this task.

Only submissions satisfying all criteria qualify. Rank by the number of the three analysis plans that additionally supply a complete symbolic contrast or estimand formula with every variable defined and a stated rule for missing observations (0 to 3). Then rank by the number of these three additional malformed-input categories correctly rejected in --self-test: an empty source_id in claim_audit.csv; an empty proposed_correction in claim_audit.csv; a claim_id set other than six unique IDs corresponding one-to-one to the six named audit issues. These fields and the six-issue mapping are required by the claim_audit.csv specification. Each category earns one point only if a test changes an otherwise valid fixture solely to create that violation, demonstrates rejection, and also demonstrates acceptance of the unchanged valid fixture. Other added constraints earn no points. The four mandatory design checks, excluded from this ranking, are duplicate unit_id, missing or repeated replicate labels within a formulation/environment cell, unexpected formulation, and unexpected environment. The mandatory missing-evidence-locator check also earns no additional points. Remaining ties go to earlier on-chain Submission, then lexicographically smaller Submission commitment if timestamps match. A sole qualifying submission wins; if none qualify the outcome is no_valid_submission.

No fabrication, microbial culture, omics downloads, new measurements, inferred project data, regulatory certification, or execution of a 180-day experiment. This bounty purchases an auditable proposal and its data specification, not proof of biodegradation.