R2

log/R2.md — Round 2 entries

Sub-file of log — see parent for index.

[2026-05-04] verify | molecules/proteins/puma.md

Page verified: molecules/proteins/puma.md

Sources checked:

• Nakano & Vousden 2001 (10.1016/s1097-2765(01)00214-3) — downloaded and read in full (PDF).
• Yu et al. 2001 (10.1016/s1097-2765(01)00213-1) — downloaded and read in full (PDF).
• Wu et al. 2005 / Slug-PUMA (10.1016/j.cell.2005.09.029) — downloaded and read in full (PDF).
• Tyner 2002 (10.1038/415045a) — local PDF cross-checked; PUMA not mentioned in paper.
• Baker 2016 (10.1038/nature16932) — local PDF cross-checked; BCL-2-family upregulation model NOT in this paper.
• Chipuk 2005 (10.1126/science.1114297) — closed-access (not_oa); unverifiable. Claims tagged no-fulltext-access.
• UniProt Q9BXH1 — verified via REST API (length, MW, BH3 motif residues 137–151, Ser10 PTM all confirmed).

Key corrections:

• Nakano 2001 footnote cell lines wrong: “Saos-2, HCT116, MCF7” → “SAOS-2, H1299-p53 inducible, RKO, MCF-7, U2OS” (HCT116 is Yu’s cell line).
• Nakano 2001 discovery description: microarray on doxycycline-inducible cells, not “adenoviral p53 delivery.” Corrected.
• “PUMA-null cells showed resistance” misattributed to Nakano 2001. Nakano used antisense knockdown (partial effect). Genetic KO data is from Jeffers 2003/Villunger 2003. Replaced with actual sub-G1 apoptosis data from paper.
• Yu 2001 author list missing Hwang (Paul M. Hwang). Corrected.
• FOXO3a claim misattributed to Yu 2001 (paper contains no mention of FOXO3a). Citation removed; tagged unsourced.
• Baker 2016 misattributed for anti-apoptotic BCL-2 upregulation in senescent cells. Baker 2016 (INK-ATTAC) does not discuss this mechanism. Citations removed from those claims; tagged unsourced.
• BH3 domain binding over-attributed: discovery papers showed BCL-2 and BCL-xL only. BCL-w/MCL-1/A1 binding is Chipuk 2005 (closed-access). Notes added.
• Footnote archive status corrected for Nakano 2001, Yu 2001, Slug/PUMA 2005 (all now locally downloaded).

Unverifiable claims: Chipuk 2005 direct-BAX-activator and all-five-BCL-2-members claims; Ensembl ENSG00000105327 not re-verified vs live database.

Downstream propagation needed:

• studies/nakano-2001-puma-p53-proapoptotic.md — verify cell lines and method if page exists.
• studies/yu-2001-puma-bbc3-colorectal.md — verify Hwang in author list if page exists.
• pathways/p53-pathway.md — check FOXO3a citation attribution and BCL-w/MCL-1/A1 binding attribution.
• interventions/pharmacological/senolytics.md — BCL-2-upregulation-in-senescent-cells framing should cite Zhu 2015/Chang 2016, not Baker 2016.

---

[2026-05-04] verify | Round 2 verification COMPLETE — 8 of 8 pages verified

All 8 Round 2 pages now verified: true (1 full = dasatinib; 7 verified-partial). Parallel batch returned in ~14 min wall-clock.

Round 2 verifier corrections — cross-cutting patterns:

• Dose / unit errors caught (rapamycin): Harrison 2009 dose was off by ~6× — wiki said “14 mg/kg/day BW” but actual was “14 mg/kg in food = 2.24 mg/kg BW/day”. Same kind of unit confusion as Round 1’s “median vs mean” (ampk verifier). Lesson: training-knowledge dose figures need PDF verification.
• Wrong allele attributions (apoptosis-pathway): Tyner 2002’s “m allele” was conflated with Maier 2004’s “p44/Δ40p53” — different allele, different paper, different mechanism. Verifier caught from local PDF.
• Wrong cell-type SCAP attributions (quercetin, dasatinib, apoptosis-pathway): BCL-xL “dominant SCAP” framing was too broad — Zhu 2015 Fig 5B shows cell-type-specificity (HUVECs:BCL-xL; IMR90:BCL-2+BCL-w; preadipocytes:ephrins/p21 — no significant Bcl-2 family). Quercetin does NOT target serpins; serpins are preadipocyte SCAPs. Multiple pages had inherited the same overstated claim from Zhu 2015 framing in secondary literature.
• Phantom claims removed (apoptosis-pathway): An entire navitoclax aged-mouse claim (“oral navitoclax 50 mg/kg × 7 days reduced p16+ in lung/liver/BM, improved HSC function”) was REMOVED — not in Zhu 2016 (which is in vitro only). Actual source is Chang 2016 Nat Med. Wiki had constructed a confidently-detailed claim from training knowledge that the cited paper didn’t support.
• Trial n + endpoint errors: PEARL n=114 (not ~50); PEARL primary VAT endpoint p=0.942 NOT significant — critical null result was missing entirely. Justice 2019 quercetin dose 1000 → 1250 mg/day; primary endpoints were feasibility (not 6MWT). Singh 2022 ATLAS n=66 → 88 randomized / 79 completed; primary endpoint peak power NOT significant vs placebo.
• Wrong author attribution (mitophagy): “Ding/Bhatt 2021” → Onishi 2021 (real authors of EMBO J paper). Narendra 2008 was over-attributed for post-2008 mechanism details.
• Strain corrections (metformin): Lifespan main result was C57BL/6 (not B6C3F1 — B6C3F1 was non-significant 4.15% replication). Treatment start age 54 weeks (not 3 months).
• Schema addition: genage-id: field added to protein frontmatter as optional (per p21 seeder escalation).

Propagation pass applied this round:

• dasatinib.md — fixed three “quercetin targets serpins” claims; corrected to “quercetin targets BCL-xL/PI3Kδ/EFNB1 in HUVECs per Zhu 2015 Fig 5B”; clarified that serpins are preadipocyte SCAPs (dasatinib-covered population), not quercetin’s direct targets

Two suspected propagations turned out NOT needed:

• mtor.md doesn’t carry the Harrison dose error (no specific mg/kg number on the page)
• p53-pathway.md doesn’t cite Brugarolas for G2/M (uses general framing without misattribution)

Final state across all 27 entity pages: 27 of 27 verified. (8 verified-partial this round; 1 full this round; 18 verified-partial Round 1; one previously-unverifiable page (López-Otín 2023) was verified earlier in the session when user provided the manual-download PDF.)

Wiki count: 62 markdown files. Round 1 + Round 2 = 23 entity content pages, plus 12 hallmark stubs, framework MOCs, study pages, source provenance, SOPs, ROADMAP/index/log.

---

[2026-05-04] seed | Round 2 COMPLETE — 8 of 8 entities drafted

All 8 Round 2 seeders returned cleanly. Total ~1500 lines of new content drafted in ~6 min wall-clock (parallel) vs ~50 min serial. Pattern from Round 1 held: project-level wiki-seeder agent + WebFetch citation discovery + confirmation works well.

Pages drafted (all verified: false):

Page	Type	DOIs cited	Notable
molecules/compounds/metformin.md	compound	6	TAME trial; Coll 2019 GDF15 mechanism; Martin-Montalvo correctly framed per ampk verifier (mean +5.83% male B6C3F1, ITP non-replication noted)
molecules/compounds/rapamycin.md	compound	7	NIA ITP (Harrison 2009 + Miller 2014); PEARL 2025 trial; rapalog family; intermittent dosing rationale
processes/mitophagy.md	process	6	PINK1/Parkin + receptor pathways; urolithin A Phase 2 (Singh 2022); Parkinson’s link
pathways/apoptosis-pathway.md	pathway	5	Intrinsic + extrinsic; Bcl-2 family table + Type I/II cells; SCAP framework
molecules/compounds/quercetin.md	compound	4	D+Q senolytic; Zhu 2015 foundational; fisetin-comparison table
molecules/compounds/dasatinib.md	compound	3	TKI; D+Q senolytic; ephrin/EPH SCAP; Justice 2019 + Hickson 2019
molecules/proteins/p21.md	protein	10	CDKN1A; el-Deiry 1993 (correct DOI); GenAge 284
pathways/sirtuin.md	pathway	9	SIRT1-7; NAD+ decline mechanisms (CD38/PARP/NAMPT); resveratrol controversy noted

Schema addition during Round 2:

• genage-id: field added to protein frontmatter as optional (per p21 seeder escalation; useful for HAGR/GenAge cross-reference on aging-relevant genes). Documented in CLAUDE.md.

Cross-cutting observations:

• Citation accountability is high. Seeders pulled DOIs almost entirely from Semantic Scholar / Crossref (via WebFetch) and confirmed against ; no DOI mismatches detected this round (vs 5 wrong DOIs in Round 1 batch).
• PDFs locally available for many cited papers — verifiers can flip many to verified: true without needing PDF download steps.
• Round 1 verification corrections were inherited cleanly — metformin used the corrected Martin-Montalvo framing (mean not median, male-only, ITP non-replication noted); p21 used the corrected el-Deiry 1993 DOI.
• Senolytic family complete — fisetin (verified-partial) + quercetin (drafted) + dasatinib (drafted); D+Q references each other correctly. Next step: interventions/pharmacological/senolytics.md category page that aggregates them.

Implicit stubs added (top new high-demand items expected after re-run):

• Bcl-2 family proteins: bcl-2, bcl-xl, bax, bak, bid, puma, noxa, mcl-1
• Caspases: caspase-3, caspase-8, caspase-9
• SIRT1, SIRT3, SIRT6, SIRT7
• NMN, NR (NAD+ precursors)
• urolithin A
• everolimus, FKBP12
• PINK1/Parkin pathway components

Next moves:

1. Round 2 verification — 8 verifier subagents in parallel for the just-drafted pages. PDFs are largely available for many citations (see seeder reports for which); verifier should be efficient.
2. Refresh implicit-stub queue post-verification.
3. Round 3 candidates — likely Bcl-2 family, NAD+ precursors, senolytics category page, model-organism pages (homo-sapiens / mus-musculus / etc. — these need the new model-organism schema worked out).

---

[2026-05-04] seed | Round 2 dispatched — 8 seeders in parallel background

Implicit-stub queue refreshed after Round 1 verification. Top demand items shifted significantly:

• metformin (13 refs), rapamycin (9), mitophagy (9), apoptosis-pathway (9), homo-sapiens (8), raptor (7), ulk1/UPS/sirtuin/p21/senolytics (6 each), TSC1-TSC2/s6k1/NF-κB/spermidine (5 each)

Round 2 batch (8 in parallel via project-level wiki-seeder subagent — now loaded after session restart):

1. molecules/compounds/metformin.md — biguanide, AMPK activator, TAME trial; emphasis on Anisimov + Martin-Montalvo (corrected per ampk verifier: mean +5.83% male B6C3F1; replication failure at higher dose)
2. molecules/compounds/rapamycin.md — mTORC1 inhibitor, NIA ITP-validated; PEARL trial; intermittent dosing rationale
3. processes/mitophagy.md — PINK1/Parkin canonical; receptor-mediated (BNIP3/NIX/FUNDC1); urolithin A induction
4. pathways/apoptosis-pathway.md — intrinsic + extrinsic; Bcl-2 family balance; SCAP framework for senescent cells
5. molecules/compounds/quercetin.md — D+Q senolytic; Zhu 2015 foundational; PI3K/AKT mechanism
6. molecules/compounds/dasatinib.md — TKI; D+Q senolytic; ephrin/EPH SCAP; Justice 2019 + Hickson 2019 trials
7. molecules/proteins/p21.md — CDKN1A/WAF1/CIP1; el-Deiry 1993 (correct DOI per p53-pathway verifier); CDK inhibitor; senescence effector
8. pathways/sirtuin.md — NAD+-dependent deacetylases; SIRT1-7; resveratrol controversy noted

Each seeder pointed at the appropriate prototype (mtor for pathways, fisetin for compounds, autophagy for processes, p53 for proteins). All instructed to use Semantic Scholar / PubMed eutils / Crossref for citation discovery (NOT archive search per user policy), with for confirmation. Verification round will follow after seeding completes.

Expected wall-clock: ~5–10 min for the parallel batch.

---