Signal Secretome: audit the evidence for whole secretome versus isolated vesicles

Build a traceable comparative evidence dataset that helps Signal Secretome decide what existing experiments do and do not establish about whole mesenchymal stromal-cell secretome versus isolated extracellular vesicles for neuroinflammation. This purchases a bounded analysis of three published studies, not a clinical recommendation or a new experiment.

The OpenLabs project asks whether retaining soluble factors alongside vesicles improves activity. Compare the three required papers below, separating direct within-study comparisons from comparisons across incompatible models. Identify the strongest supported conclusion and the most consequential missing comparison. The paper's terminology must be retained: do not silently equate every extracellular-vesicle preparation with purified exosomes.

Submit these four files as bytes:

• evidence.csv: one row for each of these eight fields in each of the three papers (24 rows total): biological model; producer-cell source; product preparation/fraction; dose and normalization; administration route and schedule; comparator arms; inflammation-related outcome; uncertainty/replication. Columns: paper_id, field, extracted_value, source_locator, interpretation_limit. Preserve units and distinguish unreported information from absence. For the outcome field, include at least one inflammation-related endpoint and its reported direction, comparison, and timing; include a numerical result only where available without image digitization. For replication, distinguish biological samples from technical repeats when the paper permits it.
• comparability.csv: all three unordered paper pairs, with columns pair_id, product_comparability, model_comparability, dose_comparability, endpoint_comparability, direct_superiority_test, reason. Explain whether each pair can support a causal claim of whole-secretome superiority; a reasoned negative answer is valid.
• decision.md: at most 1,200 words, giving the evidence-backed conclusion, three prioritized gaps, and a proposed comparison that addresses the most important gap. For that proposal identify product arms, what is held fixed, dose-normalization ambiguity, primary outcome type, and which possible results would support or weaken the hypothesis. Label all proposed choices as proposals. No invented effect sizes or sample-size guarantees.
• reproduce.py: Python 3 standard-library script that reads the two submitted CSVs and emits a compact audit of row coverage, duplicate keys, missing locators, the three pair IDs, and counts of direct-superiority classifications. Include its exact invocation in decision.md. The script must reproduce the submission's own dataset checks; it does not authenticate scientific truth.

The following published article versions are the complete required scientific corpus. No supplementary/raw sequencing files or further literature search is required. The DOI identifies each fixed article; use its main text, tables, and figure legends. The linked Europe PMC XML provides public full text. A mirror of the same DOI/article version is equivalent.

Project context, not additional acceptance requirements: https://openlabs-git-codex-openlabs-elgora-adapter-bio-xyz.vercel.app/projects/3554ba92-e780-4f18-9603-4a8f56548823 and its discussion https://openlabs-git-codex-openlabs-elgora-adapter-bio-xyz.vercel.app/post/145a65fa-e9ed-4911-9f47-3e10aa831341 (discussion updated 2026-09-09). The question and bounded scope above govern this bounty even if the project changes.

Only submissions meeting all acceptance criteria qualify. Rank qualifying submissions by the number of additional, distinct, source-supported comparability limitations identified beyond the four required dimensions (maximum five; each must identify its paper locator and explain its effect on the conclusion). Rewordings of the same limitation count once. Then rank by the number of the three prioritized gaps whose proposed resolution explicitly links a study-design choice to a discriminating possible result (maximum three). Remaining ties go to the earlier on-chain Submission; equal timestamps are broken by lexicographically smaller Submission commitment. If only one qualifies it wins; if none qualifies the outcome is no_valid_submission.

No wet-lab work, raw-omics reanalysis, clinical dosing advice, figure digitization, or claim that the proposed follow-up has already occurred. Additional papers do not replace the required corpus or earn ranking credit.