Runs¶
Auto-generated from
data/runs.jsonl. One row per successful agent run, newest first.
| Date (UTC) | Agent | Shieldbreak | Action | Hits → Kept | Rows added / updated | Source tiers (PMC / EPMC / abstract) | Notes | Commit |
|---|---|---|---|---|---|---|---|---|
| 2026-04-23 22:00 | trialist_screener | tam-depletion | new | 2010 → 30 | 37 / 0 | 0 / 0 / 37 | New shieldbreak. 12 intervention-family PubMed queries (CSF1R, PI3Kγ, CCR2/5, bisphosphonate, trabectedin/lurbinectedin, CD47/SIRPα, CD24, CD40, TLR, STING, CLEVER-1/MARCO, LILRB/TREM2). ~2010 raw hits after narrowing (LILRB/TREM2 and CLEVER-1/MARCO initial queries were too broad — 1630 and 351 — narrowed to clinical-agent tokens to 43 and 39). 30 papers kept after title/abstract screening at the 30-cap; 37 rows produced via row-grain (study × tissue × timepoint-cluster × cohort). All rows abstract-derived (source tier pubmed_abstract); full-text fetch deferred to a subsequent backfill pass. Coverage: CSF1R 13 rows (emactuzumab TGCT + ICB combos, pexidartinib, LY3022855, AMG 820, axatilimab cGVHD); CD47 4 (Advani NEJM + Sallman MDS + Daver AML + Strati NHL); CD40 4 (Byrne neoadj PDAC + Soto E/GEJ + Coveler SEA-CD40 + Weiss cabira+nivo); CLEVER-1 3 (Rannikko MATINS, Virtakoivu, Kontro+aza); LILRB/TREM2 4 (Taylor IO-108, Yeku PY159+PY314, Beckermann PY314+pembro); CCR2/5 3 (Nywening PF-04136309+FOLFIRINOX, Noel+GnP, Grierson BMS-813160); PI3Kγ 2 (Hong MARIO-1, O'Connell MARIO-3); trabectedin 2 (Germano, Sun TARMIC); STING 1 (Meric-Bernstam ADU-S100); TLR 1 (Bhatia G100). Evidence-gap interventions (no clinical TAM PD paper met screen): bisphosphonates, anti-CD24, anti-CSF1R LY3022855 rare PD detail, CD40-agonist mitazalimab (Moreno 2023 + Van Laethem OPTIMIZE-1 2024 lack direct TAM PD in abstract), CP-870,893 (Beatty 2013 B-cell focused in abstract), O'Hara 2021 APX005M pancreatic (no TAM in abstract), Padrón 2022 Nat Med (no TAM in abstract) — these were excluded to excluded.jsonl. 24 exclusions logged (preclinical, CPI-monotherapy, healthy-volunteer PD, reviews, and clinical-efficacy papers without TAM PD); 7 reviews side-listed. Extended scripts/build_table.py with new enum renderers: repolarization + expansion-with-repolarization mechanism chips (new mech-neutral purple pill); TAM gating_quality pills (CD68-CD163/CD14-CD16/mIF/CyTOF/scRNA/single-marker); scRNA-cluster-shift + phagocytosis-function readout chips; classify_success override so repolarization succeeded renders as 'repolarized' green pill instead of 'increase' orange; n_treated field fallback from _treg_ to _tam_ variant. mkdocs build --strict passes. Not pushed; awaiting user confirmation. |
1a0a7c1 |
| 2026-04-23 20:45 | trialist_skeptic | treg-depletion | new (critique) | — | 44 papers critiqued / 1 discrepancy | 28 PMC XML / 6 PMC-paywalled via WebFetched HTML / 10 PubMed-abstract | First-time critique pass over all 44 unique papers. 34 full-text-accessible; 10 PubMed-abstract-only (paywalled journals). Zero retractions found. One extraction discrepancy (Huang 2011: Mann-Whitney label applied to paired pre/post biopsy data — screener correctly noted the mis-labeling). RoB distribution: 3 Low / 4 Some concerns / 32 Moderate / 4 Serious / 1 not-amenable (Nancey case report). Confidence: 2 High / 23 Moderate / 18 Low / 1 Very low. Key cross-paper findings: (a) CD25-gating confound in DD studies partially discounts Dannull-family positives — Attia 2005 and Liao 2024 sidestep via FoxP3 mRNA. (b) Standard anti-CTLA-4 (ipi, treme) DOES NOT deplete intratumoral Tregs in humans (Sharma 2019, Huang 2011 4.75× INCREASE, Penter 2023 expands marrow Tregs) — contradicts preclinical ADCC model. (c) Fc-enhanced anti-CTLA-4 (Ager 2026, Chand 2024) DOES deplete — Fc engineering is the reconciling variable. (d) Bempeg / non-α IL-2 variants EXPAND Tregs absolutely; Diab 2020 PIVOT-02 'without Treg enhancement' claim contradicted by same-cohort Gogas 2024 data; phase 3 PIVOT-IO failed 2022. (e) Mogamulizumab cleanly depletes CCR4+ eTregs (Fujikawa 2023, Jinushi 2025, Roelens 2022). (f) Cyclophosphamide Treg-depleting activity is SCHEDULE-dependent. (g) CELMoD iberdomide (Lipsky 2022) + DNMTi (Han 2021, Penter 2023) INCREASE Tregs — honest counterexamples. Critique page: docs/shieldbreaks/treg-depletion/critique.md. |
b912509 |
| 2026-04-23 19:05 | trialist_screener | treg-depletion | refresh (scope-expansion) | 306 → 13 | 13 / 0 | 11 / 0 / 2 | Scope broadened from depletion-only to depletion OR inhibition OR destabilization. 11 new intervention-family queries (anti-TNFR2, anti-TIGIT, EP4-antagonist, DNMTi/HMA, HDACi, CELMoD/IKZF2, PI3Kδ, NRP1, GITR/OX40 function, FoxP3 destabilization, iberdomide). 13 clinical rows added across 11 new studies. 0 functional-impairment-only rows — no clinical paper reported suppressive function without a coincident count/frequency change. Honest gaps: anti-TNFR2 clinical PD literature does not exist in PubMed (Torrey 2019 Sézary is ex-vivo only); IKZF2-selective degraders, EP4 antagonists, NRP1, CCR8-CAR-T also 0 human rows. Surfaced surprises: CELMoDs (iberdomide) and DNMTi (decitabine) mostly INCREASE Tregs (Lipsky 2022 SLE +104.9% p<0.001; Han 2021 ITP Treg count+function up; Penter 2023 ipilimumab expanded BM Tregs in AML/MDS) — opposite to scope-expansion hypothesis. Context-dependent HDACi: entinostat/vorinostat DECREASE tumor Tregs (Terranova-Barberio 2020 11.8%→2.9%, p=0.0067); panobinostat INCREASES Tregs in HIV (Brinkmann 2018 +40%, p=0.003). Anti-TIGIT (Guan 2024 CITYSCAPE/GO30103) shows PBMC Treg decrease only, not tumor. | 8e0404d |
| 2026-04-23 17:15 | trialist_screener | treg-depletion | backfill | — | 0 / 38 | 28 / 0 / 10 | Full-text backfill pass; 28/38 rows now have baseline+post numerics (74%); 2 rows excluded to excluded.jsonl (Telang 2011 no Treg measurement; Galsky 2025 PORTER not Treg-targeted); Liao-ascites reclassified succeeded→partial; Hamid intent reclassified to incidental-but-measured; 6 new PMCIDs; new conflict Ager-vs-Sharma (Fc engineering reconciles) and Huang tremelimumab quantitative INCREASE in TI-Tregs corroborates Sharma null. |
0dd399c |
| 2026-04-23 16:36 | trialist_screener | treg-depletion | new | 1178 → 27 | 40 / 0 | 0 / 0 / 40 | New shieldbreak; 27 studies screened; 40 rows at (study × tissue × timepoint-cluster) granularity; numeric values often abstract-only with notes; Attia-vs-Dannull DD conflict surfaced. | 6da20c2 |
What this page is¶
A cross-shieldbreak audit log of every successful agent run on this repo. Each row records what was done, against which shieldbreak, and where the data went.
- Action — one of
new(new shieldbreak created),refresh(re-run of an existing shieldbreak),backfill(filling missing fields on existing rows),schema(schema bump only, no data changes). - Hits → Kept — hits returned by the search vs. hits kept after deduplication and screening.
- Source tiers — counts of which full-text source was used per kept paper: PubMed Central / Europe PMC / PubMed abstract.
- Notes — schema bumps, conflicts surfaced, manual interventions.
- Commit — short SHA of the commit that recorded the run (links to GitHub).
Failed or aborted runs are not logged here — they leave their per-run search file in data/shieldbreaks/<slug>/searches/ for postmortem.