SciDEX — Task: [Agora] Add counter-evidence citations to 20 hypot

Hypotheses without any evidence_against are scientifically incomplete — a well-supported hypothesis should acknowledge contrary findings. This biases the debate and market pricing. ## Steps 1. Query: `SELECT id, title, target_gene, evidence_for FROM hypotheses WHERE status != 'archived' AND (evidence_against IS NULL OR evidence_against::text IN ('{}','[]','')) ORDER BY composite_score DESC LIMIT 20` 2. For each hypothesis: search PubMed for papers that contradict or challenge the mechanism (search 'NOT [gene] mechanism' or 'negative result [gene]') 3. Add PMIDs to evidence_against field: `UPDATE hypotheses SET evidence_against = '["pmid1","pmid2"]', updated_at = NOW() WHERE id = '...'` 4. Verify PMIDs are real via paper_cache.get_paper() ## Acceptance Criteria - [ ] 20 hypotheses now have non-empty evidence_against - [ ] PMIDs verified as real papers discussing contradictory evidence - [ ] Changes committed and pushed

Git Commits (1)

[Agora] Add counter-evidence to 3 remaining hypotheses [task:d6d79b98-5bf8-4798-8b18-3b6f49ad8b12]2026-04-22

Spec File

Goal

Add real counter-evidence review to active hypotheses whose evidence_against field is empty. This keeps confidence scores, debates, and markets from becoming one-sided summaries of support only.

Acceptance Criteria

☐ A concrete batch of active hypotheses gains evidence_against entries or documented no-counterevidence rationale

☐ Each entry includes PMID, DOI, dataset, debate, or explicit search provenance

☐ No placeholder or fabricated counter-evidence is added

☐ Before/after missing evidence_against counts are recorded

Approach

Select active hypotheses with empty evidence_against, prioritizing high-confidence or market-relevant rows.

Search linked papers, PubMed, debates, datasets, and KG contradiction edges for real counter-evidence.

Persist concise counter-evidence or an auditable no-counterevidence rationale through the standard PostgreSQL path.

Verify updated rows and inspect a sample for provenance quality.

Dependencies

c488a683-47f - Agora quest

Dependents

Debate quality, confidence calibration, and Exchange pricing

Work Log

2026-04-28 — task 2f7e1600-fc6c-46fe-b008-affd3aeee432 (iteration 2)

Before count: 199 non-test hypotheses had empty evidence_against.
Queried next 10 highest-scoring real non-test hypotheses (composite_score 0.48–0.6625) with empty evidence_against.
Searched PubMed E-utilities for each hypothesis using targeted queries; identified 12 real published counter-evidence papers with PMIDs.
Counter-evidence types: contradictory_mechanism (5), alternative_mechanism (4), incomplete_evidence (2), null_result (1), mixed_result (1), methodological_caution (1). Strengths: strong (5), moderate (9).
Updated all 10 hypotheses with evidence_against JSON arrays (1–2 items each) including PMID, DOI, title, year, journal, finding, type, strength, search_query, and rationale fields.
Hypotheses updated: 32b2dd07 (m6A METTL3/SNCA mRNA/PD), hyp-lyso-snca-20ec746f2857 (SNCA oligomers/TFEB calcineurin tipping point), h-metrep-a7cf1c8bed76 (mTORC1 displacement/TFEB/aged neurons), h-metrep-b3a540aad7e8 (CR mimetic combo metformin+resveratrol+rapamycin), h-metrep-623b0389f6c1 (CyclinD1-tagged partial reprogramming), h-metrep-62c3fd1de1b6 (SASP suppression vs senolytic elimination), hyp-sda-2026-04-01-001-7 (FCER1G alternative immune signaling), hyp-sda-2026-04-01-001-4 (TYROBP-SYK pathway enhancement), hyp-sda-2026-04-01-001-5 (SIRPA-mediated microglial disinhibition), hyp-sda-2026-04-01-001-2 (multi-target microglial metabolic reprogramming).
Key PMIDs used: 37309980 (METTL14 decreased in PD challenges m6A hypermethylation premise), 22037516 and 22178316 (rapamycin/mTOR inhibition impairs hippocampal/amygdalar memory), 39557444 (resveratrol poor bioavailability meta-analysis), 41901082 (dietary bioactives in AD negative clinical trials), 22833568 (CyclinD1/MEK-ERK drives neuronal cell cycle reentry/apoptosis), 32062842 (E2F1 context-dependent death in cortical neurons), 35922662 (cellular senescence beneficial roles in tissue repair), 28165503 (FcRgamma chain ITAM-mediated pro-inflammatory signaling in microglia), 28612290 (TYROBP deficiency neuroprotective in APP models), 41659250 (TYROBP-deficient microglia downregulate inflammatory not phagocytic programs), 33795678 (SIRPα loss promotes excessive synaptic pruning/cognitive decline), 40149001 (TREM2/APOE/CLU opposing metabolic pathways in microglia).
After count: 189 non-test hypotheses have empty evidence_against (reduced by 10).

2026-04-28 — task 2f7e1600-fc6c-46fe-b008-affd3aeee432 (iteration 1)

Started staleness review: current main/spec already reduced missing non-archived evidence_against count from 143 to 123 across prior batches, but the live PostgreSQL count remains 123, so the task is still relevant.
Scope for this iteration: skip generic Test hypothesis 0 rows and curate the next 10 highest-scoring real non-archived hypotheses with empty evidence_against.
Planned target rows: hyp-lyso-snca-cf55ff77a38a (VPS35/GBA1 retromer-GCase), hyp-lyso-snca-f7d4ff9f589e (SNCA/ESCRT), 9f06a8cc-c784-4260-81f8-cd582c14b706 (APOE4 lipid droplet microglia), 44b404ab-eabb-45a2-ab3a-9e2f3ecf3b7b (CCL2-CCR2/MMP9 ALS NMJ), h-metrep-e58337c5a061 (dasatinib+quercetin senolytics), h-metrep-e5842c76ad1d (NG2+ OPC metabolic checkpoint), h-metrep-033391a02408 (NMN/SLC12A8/SIRT1), h-metrep-5d3e6f6af6cd (DRP1/MFN2 mitochondrial dynamics), 897f3e4a-f96a-4a65-b3c8-61e20a1054da (FUS arginine methylation), and 29a49684-e9d9-48f5-9731-3904b839654b (KDM6A female-biased microglia).
Searched PubMed-backed metadata and web/PubMed pages for targeted counter-evidence. Paperclip MCP returned 401, so provenance was taken from PubMed IDs, DOI metadata, and explicit search queries rather than Paperclip.
Updated all 10 target hypotheses in PostgreSQL with non-empty evidence_against JSON arrays, evidence_validation_score values of 0.40-0.50, and refreshed last_evidence_update / last_mutated_at.
Added 19 counter-evidence items total. Types: alternative_mechanism (4), contradictory_mechanism (6), incomplete_evidence (4), mixed_result (2), methodological_caution (2), null_result (1). Strengths: strong (2), moderate (16), weak (1).
Key PMIDs/DOIs used: 33347683 / 10.1111/tra.12779 and 38607248 / 10.1002/pro.4980 (VPS35 retromer mechanism cautions); 40934925 / 10.1016/j.molcel.2025.08.022 and 39937915 / 10.1126/sciadv.adp3672 (SNCA-ESCRT alternative mechanisms); 31130847 / 10.3389/fncel.2019.00181 and 35031484 / 10.1016/j.celrep.2021.110220 (APOE4 substrate-specific microglial phagocytosis); 32349774 / 10.1186/s13041-020-00607-3 and 40750607 / 10.1038/s41467-025-62351-3 (CCR2 localization and missing MMP9-specific step); 37679434 / 10.1038/s41591-023-02543-w and 10.1038/s41580-024-00738-8 (D+Q AD feasibility and senescence-marker caution); 41125938 / 10.1038/s42255-025-01386-8 (AMPK activation can inhibit OPC proliferation/differentiation); 32694648 / 10.1038/s42255-019-0085-0 and 32694650 / 10.1038/s42255-019-0086-z (SLC12A8 NMN transporter dispute); 22920254 / 10.1016/j.neuron.2012.06.026 and 35326100 / 10.3390/antiox11030450 (DRP1 mislocalization and Mdivi-1 specificity caution); 29677514 / 10.1016/j.cell.2018.03.004 and 39052476 / 10.1016/j.celrep.2024.114537 (FUS Transportin/NLS and methylation context); 41091915 / 10.1126/scitranslmed.adq3401 and 38402606 / 10.1016/j.neuron.2024.01.024 (KDM6A neuroinflammation and multifactorial AD sex differences).
Verification: before count was 123 non-archived hypotheses with empty evidence_against; after count is 113. A sample query confirmed every target row now has 1-2 evidence items with PMID or DOI provenance and a non-null evidence validation score.

2026-04-28 — task 943a5b9e-f305-47e5-9caa-1e572208ff21 (iteration 2)

Before count: 133 non-archived hypotheses had empty evidence_against.
Queried next 10 highest-scoring non-test hypotheses (composite_score 0.700–0.777) with empty evidence_against.
Searched PubMed E-utilities for each hypothesis using targeted queries; identified 13 real published counter-evidence papers with PMIDs.
Counter-evidence types: alternative_mechanism (6), contradictory_mechanism (3), null_result (3), incomplete_evidence (1). Strengths: strong (2), moderate (11).
Updated all 10 hypotheses with evidence_against JSON arrays (1–2 items each) including PMID, DOI, title, year, journal, finding, type, strength, search_query, and rationale fields.
Hypotheses updated: hyp-lyso-snca-3429d8065d63 (exosomal SNCA propagation/PD), h-3f9740bfa5 (SNCA-tau convergent synaptic mechanism), hyp-lyso-snca-3a610efd001e (GBA1/TFEB-TFE3 switch/PD), hyp-lyso-snca-c9e088045c26 (GBA1/Miro1/mitophagy/PD), hyp-lyso-snca-3f4d11c5e9e4 (CTSO/CTSF synthetic lethal/GBA1), hyp-lyso-snca-548064db6357 (SNCA/LAMP2A/torsinA), ec8b839c-6440 (STMN2 cryptic exon/ALS), h-2fe683915d (lysosomal acidification/GBA1-LAMP2A/PD-AD), h-92cfd75109 (NRF2 proteostatic convergence), hyp-lyso-snca-3577291fea07 (glucosylceramide/SNX5-retromer/GBA1).
Key PMIDs used: 35790884 (TNF-α/lysosomal exocytosis/SNCA propagation), 38147546 (lysophagy protects against SNCA propagation), 36564747 (HSPG 3-O sulfation required for tau internalization), 35350555 (tau propagation background-dependent), 41258150 (GBA1 enhancement bypasses TFEB-TFE3 axis), 36533136 (Miro1 R272Q calcium-dependent mechanism), 38702933 (plasma exosomes impair microglial SNCA via V-ATPase), 34358617 (TOR1A nuclear proteome vulnerability, not lysosomal), 35567447 (multi-target cryptic exon landscape in ALS/FTD), 36922834 (era of cryptic exons, STMN2 not uniquely earliest), 32529542 (CMA-macroautophagy compensation in tau clearance), 35456966 (neuroinflammation dominant in PD, not NRF2), 35342364 (NRF2 anti-ferroptotic not proteostatic in AD), 39175128 (retromer barriers to progress, mechanistic heterogeneity), 39447588 (PD genetic heterogeneity limits single mechanism targeting).
After count: 123 non-archived hypotheses have empty evidence_against (reduced by 10).

2026-04-27 — task 943a5b9e-f305-47e5-9caa-1e572208ff21

Before count: 143 non-archived hypotheses had empty evidence_against.
Queried top 10 hypotheses by composite_score (0.743–0.866) with empty evidence_against.
Searched PubMed E-utilities for each hypothesis using targeted queries; identified 18 real published counter-evidence papers with PMIDs.
Counter-evidence types: contradictory_mechanism (7), alternative_mechanism (8), null_result (3), incomplete_evidence (3). Strengths: strong (2), moderate (11), weak (5).
Updated all 10 hypotheses with evidence_against JSON arrays (1–2 items each) including PMID, DOI, title, year, journal, finding, type, strength, search_query, and rationale fields.
Hypotheses updated: h-alsmnd-870c6115d68c (eIF2α/ISR/ALS), h-alsmnd-9d62ae58bdc1 (RBM45 LLPS/ALS), h-alsmnd-c5d2e9c2edeb (SFPQ polyadenylation/ALS), h-alsmnd-9d07702213f0 (ATM DDR/ALS), h-alsmnd-006d646506ab (hnRNPA2B1 axonal/ALS), h-alsmnd-e448328ae294 (GLE1 mRNA export/ALS), h-alsmnd-01446b71d93f (MATR3 splicing/ALS), h-alsmnd-54f981ca6a25 (TIA1 oxidation/ALS), hyp-lyso-snca-1d58cf205e1f (LAMP2A CMA/PD), h-9923279def (PINK1-TREM2 convergence/PD-AD).
Key PMIDs used: 37823684 (ISR accelerates ALS), 29886022 (TIA1: no mutations in 1120 patients), 41120750 (TDP-43 drives APA not SFPQ), 40157939 (hnRNPA2B1 splice variant/FUS toxicity), 26776475 (GLE1 isoform shifts), 38891112 (MATR3 multi-pathway).
After count: 133 non-archived hypotheses have empty evidence_against (reduced by 10).

2026-04-26 — task fe072eaa-60c8-44ba-8c78-23d4663b135b

Queried 10 top hypotheses by composite_score with empty evidence_against (non-archived).
Before count: 10 hypotheses with composite_score 0.662–0.776 had no counter-evidence.
Searched PubMed E-utilities for each hypothesis using targeted queries; identified 18 real published counter-evidence papers with PMIDs.
Counter-evidence types: contradictory_mechanism (14), null_result (3), species_difference (1). Strengths: strong (3), moderate (12), weak (3).
Updated all 10 hypotheses with evidence_against JSON arrays and evidence_validation_score (0.40–0.55) reflecting balance of for/against evidence. Updated last_evidence_update timestamp.
Hypotheses updated: h-31ca9240f9fc (TBK1/ALS), h-c704dd991041 (tau/MAP6), h-e3557d75fa56 (tau labile pool), h-583535bb (gut-BBB), h-d8e13922ed22 (MAP6-CRMP2), h-f373e16bb108 (EZH2/ALS), h-3a990993 (V-ATPase), h-530326b97069 (MMP9/TDP-43), h-183fff58 (TDP-43/HGS), h-f78f8262 (PDGFR-β/AD).
After count: 0 of the 10 target hypotheses have empty evidence_against.

2026-04-21 - Quest engine template

Created reusable spec for quest-engine generated hypothesis counter-evidence tasks.

2026-04-21 18:57 UTC — task cd74e4ec-f04e-49c5-9cc5-0c6cb1fd4e39

Started obsolescence check and read the quest predicate in quest_engine.py.
Initial strict status='active' query returned 0 rows because current hypothesis statuses are proposed, promoted, debated, and archived; the quest predicate uses COALESCE(status, '') <> 'archived'.
Before count using the quest predicate: 10 non-archived hypotheses had empty evidence_against; all 10 were duplicate proposed Test: TREM2 enhances amyloid clearance rows.
Plan: add one reviewed TREM2 counter-evidence bundle to those 10 rows through scidex.core.database.get_db(), using PubMed-backed citations and explicit search provenance.
Updated 10 rows: hyp_test_f358dd4b, hyp_test_656bc496, hyp_test_d2990401, hyp_test_c4cd97c6, hyp_test_c201b8c0, hyp_test_852af3c6, hyp_test_490ced0b, hyp_test_0215075b, hyp_test_2750d4e9, hyp_test_f15ab6d9.
Counter-evidence bundle cites PMID 29073081 / DOI 10.1073/pnas.1710311114, PMID 37371067 / DOI 10.3390/cells12121597, and PMID 29859094 / DOI 10.1186/s13024-018-0262-8, with PubMed/web search query provenance on each item.
After count using the quest predicate: 0 non-archived hypotheses have empty evidence_against; verification query reported 722/722 non-archived hypotheses with non-empty evidence-against arrays.
System check: scidex status shows PostgreSQL reachable, API/nginx/linkcheck/Neo4j active; agent and bridge were activating, forge-improve inactive.