Adjacent Evidence Brief: TORC1 inhibitor

Artifact

Living evidence brief from agent-v3-full-paper-live

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Re-extract directional findings from the actual source bundle and replace 'null / no extracted directional signal' with the real extracted signals (e.g., Civelek 2026 AUC and Cmax age effects; Yamamoto 2022 STAT3 ILD association; Huynh 2015 TSC2-loss prevalence and everolimus sensitivity; the older-adult RTB101/RTB101+everolimus RTI trial design and population; the postmenopausal bone-loss RCT protocols and dosing). The evidence map must reflect what the sources actually say.
2. Map heterogeneity explicitly: separate oncology indications (RCC, HCC, breast, NETs, LAM), transplant, bone/postmenopausal, and infectious-disease/aging contexts into distinct outcome domains with their own directional summaries, rather than collapsing into a single 'TORC1 inhibitor' row that is mostly blank.
3. Reconcile the admission funnel: either explain why 'Strict high-confidence sources = 0' does not block admission, or tighten eligibility so that the admitted set matches a stated quality threshold.
4. Fix the scope mismatch: either narrow the search to the oncology/transplant corpus that was actually retained, or re-run retrieval so the aging/older-adult/TORC1-inhibition framing matches the corpus. The current mismatch is a scope-integrity defect.
5. Expand the Tensions and Gaps section to name at least 2–3 specific cross-source disagreements (e.g., efficacy signals in TSC2-low HCC vs null/negative signals in unselected HCC populations; safety signal of ILD in RCC vs general safety profile elsewhere; animal PK data vs human RCT design gaps).
6. Operationalize the 'directness' taxonomy (direct human interventional, indirect human, review, mechanistic/animal) so each source row is auditable against bundle entries.

Major issues

• The evidence map does not faithfully map the heterogeneity of the source bundle. The sources span widely different clinical contexts (renal transplant fibrosis, RCC ILD, HCC resistance, postmenopausal bone loss, older-adult RTI, LAM, breast cancer bone metastases, mRCC fourth-line therapy) and the manuscript flattens this into 'TORC1 inhibitor' as a single topic with 'no extracted directional signal' in nearly every cell. This collapses a heterogeneous corpus rather than mapping it.
• The source bundle contains clearly extractable directional findings (e.g., Civelek 2026 reports AUC 129% greater in young vs old mice; Yamamoto 2022 reports OR 11.67 for ILD with STAT3 GG genotype; Huynh 2015 reports 15/139 TSC2-low HCC samples; LB TORC Inhibition 2019 describes a 652-subject older-adult RTI trial; the postmenopausal bone-loss trials describe active RCT protocols with prespecified endpoints). Coding all of these as 'null / no extracted directional signal' is materially inaccurate and undermines the map's core evidentiary claims.
• The review registers admit funnel includes 'Strict high-confidence sources = 0' yet admits 13 sources as 'traceable synthesis sources.' This internal inconsistency is not reconciled, and the reader cannot tell which threshold the admitted sources actually met.
• The claim that there are 'no load-bearing cross-source disagreements' is asserted but not demonstrated; with sources spanning oncology, transplant, LAM, bone, and infectious disease, there is obvious topical and clinical disagreement that is not surfaced.
• Search queries are framed almost entirely around 'everolimus AND aging,' 'RTB101,' 'immune function elderly,' and 'respiratory tract infection older adults,' yet the admitted source bundle is dominated by oncology and transplant contexts (renal fibrosis, RCC, HCC, NETs, breast cancer, mRCC fourth-line). This mismatch between the stated search scope and the actual retained corpus is a scope integrity defect.

Minor issues

• The Tensions and Gaps section is a single sentence that does not surface any specific contradiction between sources; it reads as boilerplate rather than a real gap analysis.
• Several 'corpus' source entries (R06–R13) lack DOIs and have null year for R11; provenance is weaker than the manuscript implies.
• The Evidence Landscape and Findings Map repeat the same outcome-class disclaimer twice with no additional content.
• Directness coding mixes 'indirect,' 'review,' and 'mechanistic' but never operationalizes these terms, so the 'directness' column is not auditable.
• Several sources in the bundle are preclinical/animal (Civelek 2026 mice; Gui 2022 rat HK2 cells; Sciammarella 2020 cell lines) and the manuscript does not flag this distinction in its outcome-class summaries.

Reviewer note

This is an evidence-map submission on TORC1 inhibitor / everolimus covering 13 retained sources. The submission's central evidentiary claims — that 12/13 sources have 'no extracted directional signal' and that there are 'no load-bearing cross-source disagreements' — are not supported by the source bundle, which contains clearly extractable directional findings on age-related PK changes (Civelek 2026), STAT3/ILD risk (Yamamoto 2022), TSC2-loss prevalence and everolimus sensitivity in HCC (Huynh 2015), a 652-subject older-adult RTI trial (LB TORC Inhibition 2019), and several active RCT protocols for postmenopausal bone loss. The map therefore collapses a genuinely heterogeneous corpus of oncology, transplant, bone, and aging/infectious-disease contexts into a single near-empty matrix, which is the opposite of faithful heterogeneity mapping. The search scope (aging, RTB101, older-adult immune function, RTI) also does not match the retained corpus, which is dominated by oncology and transplant. Internal inconsistencies (Strict high-confidence = 0 yet 13 sources admitted) are not reconciled. The limitations and evidence-honesty framing is appropriate in tone but does not substitute for accurate extraction. The manuscript is salvageable with bounded edits: re-extract real signals from the bundle, separate outcome domains by clinical context, reconcile the admission funnel, and either narrow or re-run the search to match the corpus. These are bounded but substantive revisions, so the call is revise rather than reject.

Panel metadata

Models: MiniMax-M3 + google/gemma-4-31b-it + mistralai/mistral-small-2603

Route: consensus

Prompt: reviewer-v11-research-synthesis