What we found
In this public data, "lives longer" and "ages slower" do not move together, and the direction of the gap depends on sex. Male mice on acarbose took a top-tier lifespan benefit (the 1000 ppm arm at hazard ratio 0.55, ranking 5th of 93 estimable arms; acarbose's best arm, in C2009, ranks 2nd at 0.49) yet gained almost nothing on old-age motor function (day-6 rotarod +8 s, p=0.41); their survival advantage looks more like suppression of specific lethal lesions — lung tumors halved (37%→17%), liver degeneration cut to a quarter (29%→7%). Females on acarbose ran the opposite pattern: a smaller lifespan benefit (0.73) but clearly better old-age rotarod performance (+55 s on day 6, p<0.001, with the learning slope doubled) and renal glomerulosclerosis down from 18% to 2%. Males on NDGA got both: longer life (0.77) and old-age grip duration restored to young levels (p=0.003). Males on 17α-estradiol took the 4th-strongest lifespan benefit programme-wide and 2nd among single-compound arms (0.51) — with no public functional data to compare it against. The strongest single-agent lifespan evidence sits next to a blank. In short: longer life can mean slower aging, or it can mean dying less often of a few specific diseases; both patterns exist in this data set, and with what is public today they cannot be separated mouse by mouse — only group by group.
| What we did | What we got | What it does not answer |
|---|---|---|
| Downloaded and verified all 9 advertised ITP functional/pathology phenotype files (8 succeeded, SHA256 recorded) | Seven datasets: rotarod, grip, body composition, fat pads, blood glucose, uterus weight, pathology spectrum | The C2010 NDGA rotarod file is missing from public storage (the page links it, the object returns 404) |
| Re-analysed the per-mouse phenotype data against the papers' printed results | Female rotarod benefit, male null, and the lesion-by-lesion significance pattern all agree with Harrison 2019; the NDGA male grip benefit agrees with Strong 2016 | Matching printed values proves we read the files and defined the cohorts correctly — not that the original measurements are right; some p-values differ in kind because the papers pool three sites while the phenotype files hold single-site subsets |
| Compared old treated groups to old controls endpoint by endpoint, with young controls as the ageing reference | Female acarbose rotarod clearly better, males null; grip mixed in both sexes with male forelimb strength actually worse; body fat far below controls in both sexes; NDGA males' grip duration restored, females null | Every phenotype subset is a single-site, few-dozen-mouse sample; multiplicity correction was applied within endpoint families only; non-significance is not evidence of no effect |
| Placed function/pathology next to the previously published lifespan hazard ratios | Four group-level patterns: acarbose males "long life, weak function", acarbose females "moderate life gain, strong function", NDGA males "both", 17α-estradiol males "longest life, no data" | Phenotype and lifespan mice use different ID systems, so one mouse's "how long" cannot be joined to its "how well"; every concordance judgement here is group-level, not individual-level |
Scope is limited to whether, in genetically heterogeneous UM-HET3 mice, ITP interventions' lifespan effects and their published functional/pathology effects point the same direction at group level. This is an evidence description and gives no dosing, regimen or intervention advice; ppm figures are factual records of what a given trial fed. Every "functional benefit" reported here is a mouse readout and is not extrapolated to human healthspan.
Data: two records that do not talk to each other
The lifespan files were verified mouse by mouse in the previous study: 17 cohorts, 36,975 mice. This study targets the project's other public release — phenotype and pathology files under the Mouse Phenome Database's othpheno directory. We opened all 11 advertised phenotype pages; 9 offered downloadable per-mouse sheets, and 8 downloaded successfully with SHA256 fingerprints recorded. ITP phenotype data
What arrived splits into two different kinds of evidence that must not be merged:
Old-age functional and disease outcomes — the direct readouts of "ageing slowly":
| Dataset | Cohort/intervention | Content | Size |
|---|---|---|---|
| Rotarod | C2013 acarbose 1000 ppm | Latency to fall across 6 days of training and testing at 22–23 months | 136 mice (UT only) |
| Grip | C2013 acarbose; C2010 NDGA | Forelimb/all-limb strength and hang duration at 23 months | 140 + 99 mice |
| Body composition | C2013 acarbose | Monthly body mass, fat and lean mass, 8–22 months | 168 mice, longitudinal |
| Fat pads | C2013 acarbose | Four dissected fat depots weighed | 33 mice (UM) |
| Pathology spectrum | C2013 acarbose | Necropsy lesion profile at 22–25 months (summary table) | 84 control, 111 treated |
Mechanism and exposure checks — done on young mice, answering "did the drug act as intended", not "how much did they age":
| Dataset | Content | Size |
|---|---|---|
| Refeeding glucose | Glucose curve after overnight fast and refeeding at 5.5 months, 6 weeks on diet, four doses | 80 mice |
| Uterus weight | Ovariectomised 4-month females, testing whether 17α-estradiol lacks uterotrophic activity | 28 mice |
Two boundaries belong up front, because they change how every later number reads.
First, the phenotype mice and the lifespan mice are different animals, and they cannot be joined. Lifespan IDs look like JL010005 or UT06219; phenotype IDs look like A01 or UTBh4-C10 — a different system. We checked both value domains and found no mappable key. It is therefore impossible to take a long-lived mouse and ask for its rotarod score — the individual-level question "did the same mouse live long and age well" cannot be answered with public data. Everything below retreats to group level: mean function in a treated arm versus its control, set against the same cohort's lifespan effect. That is a genuine information gap, not a stylistic choice.
Second, a file that should exist does not. The C2010 NDGA rotarod page advertises a download link, but the object store returns 404 — the file is not there. The page's summary figures (SVG/PDF) survive, and the original paper, Strong 2016, reports the experiment (p=0.02 in each sex). We record the gap honestly: what follows on NDGA rotarod cites printed values only, with no independent re-computation. Strong 2016
Checking we read the files right
Before estimating anything, we re-computed the papers' printed results from the per-mouse data — the same procedure as the previous study.
The acarbose phenotypes are exactly the experiments reported in Harrison 2019, "Acarbose improves health and lifespan in aging HET3 mice". Harrison 2019 Point by point: the paper writes that rotarod training "was more effective in females fed ACA, but there was no benefit in males" — we compute +55 s on day 6 in females (p=3×10⁻⁴; the paper prints p=0.0001) and +8 s in males (p=0.41; judged non-significant). The paper reports lung tumors halved, adrenal medullary vasodilation significant in females, liver degeneration significant in males, glomerulosclerosis only in females — our Fisher exact tests reproduce the significance pattern cell for cell. The paper reports refeeding-glucose effects only in males — our area-under-curve computation likewise gives male high-dose p=0.002 and nothing significant in females. Denominators match too (pathology: 41 male/43 female controls, 54 male/57 female treated).
For NDGA grip, Strong 2016 writes that old male controls' grip duration fell below young levels while NDGA males were "indistinguishable from young mice" — we compute a male group difference in best-trial grip duration of p=0.003, same direction.
Two analytic differences are on record: the papers fit models pooling three sites with more covariates, while the phenotype files hold single-site subsets (UT; UM for fat pads), so we use Welch tests with Hedges' g — p-values differ in detail but every conclusion agrees in direction; and the pathology sheet is a summary table rather than per-mouse records, so we could only recompute its margins.
Result 1: Acarbose — the sex with the largest lifespan gain has the least functional gain
Acarbose males hold a top-tier lifespan benefit in the ITP compound library (0.55 in this cohort; acarbose's C2009 arm is stronger still at 0.49); females sit at 0.73. The functional readouts paint the opposite picture.
Rotarod (22–23 months, 6 days): treated females reached a day-6 latency of 106 s versus 51 s in old controls (+55 s, p=3×10⁻⁴, Hedges' g=1.03), with the six-day mean also higher (+27 s, p=0.014). The sharper signal is the learning slope — treated females improved 15.6 s per day against 5.3 s in old controls (p=2×10⁻⁵), and a mixed-effects model confirms the treatment-by-day interaction (p<0.0001). In males, none of the three readouts shows statistical evidence (p=0.24–0.71). [Figure 2]
A confound that must be stated: treated females were 12 g lighter than controls (35 g versus 47 g, p=3×10⁻⁵), and body weight itself predicts rotarod latency. With weight as a covariate, the day-6 difference stays nominally positive (+34 s, p=0.043) and the slope stays significant (p=0.005), but the six-day mean no longer is (p=0.39). Part of the female rotarod benefit, in other words, may simply be a lighter body — and the data cannot separate how much is functional improvement versus body-composition effect, which are not equivalent under a "slower ageing" reading.
Grip (23 months): too mixed to average. Males were nominally stronger on all-limb force (+15.6 g, p=0.064) and nominally longer-hanging (p=0.049), yet significantly worse on forelimb strength (−14 g, p=0.018); all four female endpoints sit near zero. Same mice, same rig, different directions per endpoint — which is exactly why "function" must not be collapsed into one number.
Body composition (longitudinal, 8–22 months): the most consistent functional-layer signal in acarbose, carefully named. Treated mice of both sexes never developed the control group's midlife weight gain: fat percentage at 22 months was 25% versus 44% in females and 17% versus 27% in males (both p<0.0001). Most of the gap is prevented fat accrual — control females gained 0.69 percentage points of fat per month while treated females declined. Lean mass differs by sex: females essentially held it (−0.5 g, p=0.45) while males lost 2.5 g (p=0.012). And because only survivors reach 22 months, differential survival adds a selection component to these trajectories. [Figure 3]
Fat pads (dissected weights, n=7–9 per arm): male mesenteric and subscapular depots are significantly lower after weight adjustment (p=0.0009 and p=0.0001); female samples are too small and directionally unstable to conclude.
Pathology spectrum (necropsy at 22–25 months, summary table): treated mice showed lower rates on four lesions — male lung tumors 37%→17% (p=0.033) and liver degeneration 29%→7% (p=0.006); female adrenal medullary vasodilation 26%→7% (p=0.021) and renal glomerulosclerosis 18%→2% (p=0.014). More than ten other lesions showed no significant difference, denominators vary by lesion (tissue availability differs), and the control group pools in a few earlier same-age necropsy cohorts — a preplanned design choice by the authors. These are per-sex Fisher tests without cross-lesion multiplicity correction; read them as exploratory. [Figure 4]
Refeeding glucose (5.5 months, mechanism check): males show a clean dose gradient — the 2500 ppm arm's area under the curve is 22% below controls (p=0.002), with lower fasting glucose at baseline; no dose and no time point is significant in females. Measured in young mice, this is not an ageing outcome. But it shows acarbose's post-prandial glucose blockade is pharmacologically more complete in males — the same direction as males' larger lifespan benefit. A mechanistically coherent hint that cannot be closed inside this data. [Figure 5]
Result 2: NDGA and 17α-estradiol — one concordant, one blank
NDGA (males 2500 ppm from 13 months): a moderate male lifespan benefit (0.77) plus old-age grip duration restored to young levels — best-trial 260 ms versus 168 ms in old controls (p=0.003, g=1.55); grip force itself reaches only marginal evidence (p=0.078). Females on NDGA 5000 ppm show no lifespan benefit (1.09) and no grip benefit (p=0.42) — a concordant negative cell. The missing rotarod file limits NDGA rotarod to printed values: Strong 2016 reports p=0.02 in each sex. [Figure 6, left]
17α-estradiol (14.4 ppm): the 4th-strongest male lifespan benefit programme-wide (0.51), none in females (1.01). The only public phenotype touching it is a mechanism check — uterus weight in ovariectomised 4-month females — where the high-dose arm restores uterine weight to near-intact levels (93 mg versus 130 mg) and the low dose only partially (33 mg). The assay exists to test the "non-feminising oestrogen" label, and at the lifespan dose there is clearly uterotrophic activity — a useful footnote, but not an ageing readout and not functional-benefit evidence. [Figure 6, right]
Result 3: the two layers side by side — four patterns, not one slope
Aligning the previously published per-mouse lifespan results with this study's functional findings by treatment and sex gives four patterns, not a correlation:
| Treatment × sex | Lifespan (hazard ratio) | Old-age function | Pathology | Group-level pattern |
|---|---|---|---|---|
| Acarbose male | 0.55 (strong) | Rotarod null; grip mixed (forelimb worse) | Lung tumor, liver degeneration down | Dissociated: survival best explained by fewer lethal lesions |
| Acarbose female | 0.73 (moderate) | Rotarod clearly better (partly weight-linked) | Glomerulosclerosis, adrenal vasodilation down | Weakly concordant: same direction, and the functional evidence is the more striking half |
| NDGA male | 0.77 (moderate) | Grip duration back to young levels | None public | Concordant |
| NDGA female | 1.09 (none) | No benefit | None public | Concordant (negative) |
| 17α-estradiol male | 0.51 (4th overall) | No public functional data | None public | Cannot judge |
| 17α-estradiol female | 1.01 (none) | Mechanism check only | None public | Cannot judge |
This is the debate made concrete. Wilkinson 2012 used rapamycin necropsy data to argue "slower ageing" (many non-lethal lesions reduced); Neff 2013 countered with broad functional phenotyping ("merely suppresses specific lethal pathologies" — rapamycin extended life while improving few ageing phenotypes, and some "improvements" also appeared in young mice, marking them as ageing-independent drug effects). Wilkinson 2012 Neff 2013 Our data says each picture caught part of the truth: within one programme, "longer life means slower ageing" fails for acarbose males (at least for motor function), holds for NDGA males, and broadly holds for acarbose females. It is not a question with one answer — it has to be judged per intervention, per sex, per endpoint.
What this study can and cannot say
It can say: within the public ITP data, the groups with the largest lifespan benefits (acarbose males, 17α-estradiol males) are not the groups with the most complete functional evidence; acarbose males' survival gain aligns directionally with reduced lethal lesions at necropsy — more compatible with "suppressing specific pathologies" than with broad ageing deceleration, though the separation is only group-level. Acarbose females run counter to that pattern: the most convincing functional benefit sits in the group with the moderate lifespan gain.
It cannot say: anything about whether an individual long-lived mouse also aged slowly (IDs cannot be joined); anything about acarbose slowing human ageing (all data are mouse, and the functional subsets are single-site dozens); that lower body fat reads directly as "healthier" (male lean mass also fell, and the fat difference may itself mediate the rotarod difference); an independent judgement on NDGA rotarod (file missing — printed values only); or that the ten-plus non-significant lesions equal "no effect" — necropsy denominators are small, and non-significance is not equivalence.
Method boundaries: the phenotype files are single-site subsets, not the papers' pooled three-site models, so our p-values differ in kind; the "recovery percentage" (treated gain over the young–old gap) is a descriptive scale, not a biological-age measure and not evidence of rejuvenation; body-composition trajectories carry survivor selection; and pathology is a summary table, not per-mouse records, so it cannot enter a survival model.
Relationship to the previous study
This article is a stand-alone task answering a question left open by RESEARCH-007 (Rapamycin: mouse lifespan extension is real — what is still missing for humans?). Every lifespan number is reused from that study's published computation and was not re-run; what is new here is the functional/pathology layer. Together they form a complete evidence stack: strong at the lifespan layer (per-mouse, three sites, cross-cohort replication), limited at the function layer (per-mouse but single-site and small), exploratory at the pathology layer (group summaries), and missing at the individual linkage layer. Rapamycin itself has no matching entry in these phenotype files — ITP has published no functional dataset for it in this directory.
Sources
- ITP phenotype and pathology per-mouse data (Mouse Phenome Database)
- ITP phenotype pages and data dictionaries
- Harrison 2019: Acarbose improves health and lifespan in aging HET3 mice
- Harrison 2014: Acarbose, 17α-estradiol and NDGA extend mouse lifespan preferentially in males
- Strong 2016: Four interventions extend lifespan in male mice
- Wilkinson 2012: Rapamycin slows aging in mice
- Neff 2013: Rapamycin extends lifespan but has limited effects on aging
- Snyder 2022: Canagliflozin retards age-related lesions in multiple organs
- Snyder 2025: End-of-life pathology with 16α-hydroxyestradiol / late-start canagliflozin
- This repository, RESEARCH-007: per-mouse ITP lifespan re-analysis outputs
Scope & limitations
- Phenotype and lifespan mice are different animals with unmappable ID systems, so whether an individual long-lived mouse also aged slowly cannot be answered; all concordance claims are group-level only (c10).
- Every phenotype subset is single-site (UT; UM for fat pads) and a few dozen animals; lifespan estimates pool three sites, so the two layers are not fully commensurable.
- The C2010 NDGA rotarod per-mouse file is missing (404); only printed paper values can be cited; no public functional phenotypes exist for 17α-estradiol or rapamycin (c11).
- Pathology is a summary count table, not per-mouse records; denominators vary by lesion, controls include earlier-cohort necropsies, and no cross-lesion multiplicity correction was applied — exploratory tier (c6).
- The rotarod benefit is entangled with lower body weight; after weight adjustment some readouts lose significance, so functional gain and body-composition effects cannot be fully separated (c2, c4).
- Body-composition trajectories include only survivors to each age, and differential survival adds a selection component (c4).
- The 'recovery percentage' is a descriptive scale (it can exceed 100%), not a biological-age measure and not evidence of rejuvenation.
- Many endpoints and small samples; Holm correction was applied within endpoint families only; non-significance is never read as no effect or equivalence.
- Refeeding glucose (5.5 months) and uterus weight (4-month OVX) are young-mouse mechanism checks, not ageing outcomes (c5, c8).
- Lifespan effects are taken from the previous study's published recomputation (six cells); if that record is revised, these numbers follow.
- The search did not exhaust non-MPD functional studies (other strains, other labs on rapamycin); the debate framing is limited to the two papers actually read (c12).
Sources
- NIA Interventions Testing Program, per-mouse phenotype and pathology records (othpheno)
dataset · Source version: public release as downloaded 2026-09-18; 8 xlsx files, SHA256 recorded in evidence manifest
Reading scope
Full text
All 8 files parsed and analysed: rotarod 136, ACA grip 140, NDGA grip 99, body composition 168, fat pads 33, glucose 72, uterus 28 per-mouse records; pathology is a summary table (41M+43F control, 54M+57F treated). Non-text columns coerced to numeric so 'NA' strings become missing.
- data/C2013_pheno_rotarod_ACA.xlsx
- data/C2013_pheno_grip_ACA.xlsx
- data/C2013_pheno_bodycomp_ACA.xlsx
- data/C2013_pheno_fatpads_ACA.xlsx
- data/C2013_pheno_glucose_ACA.xlsx
- data/C2013_pathology_ACA.xlsx
- data/C2010_pheno_grip_NDGA.xlsx
- data/C2011_pheno_OVX_17aE2.xlsx
- MPD othpheno measure pages and analysis plots for ACA C2013, NDGA C2010, 17aE2 C2011
dataset_documentation · Source version: pages snapshot 2026-09-18 under evidence/longer-life-slower-aging/sources/othpheno-pages/
Reading scope
Full text
All 11 pages opened to check data dictionaries and download links; the NDGA C2010 rotarod phenomedoc key was confirmed 404 (NoSuchKey) on repeated requests; only its SVG/PDF summary plots are kept as corroboration.
- sources/othpheno-pages/ (11 page snapshots incl. ACA_rotarod_C2013, ACA_pathology_C2013, NDGA_rotarod_C2010)
- sources/analysis-plots/ (SVG figure assets for the missing NDGA rotarod file)
- Harrison DE et al. Acarbose improves health and lifespan in aging HET3 mice. Aging Cell 2019;18:e12898
paper · Source version: PMC full text, downloaded 2026-09-18
Reading scope
Full text
Full text read. Printed values checked cell by cell against our re-computation: female rotarod day-6 p=0.0001 (ours p=3e-4), mean-6 p=0.02 (ours 0.014), slope p=0.009 (ours 2e-5), all male null; lesion significance pattern identical; glucose dose-response males only. Difference in kind: their model pools three sites while the phenotype files are single-site subsets. Pathology controls include preplanned earlier-cohort necropsies.
- Results 2.1-2.5
- Figures 2-6
- pathology denominators paragraph
- Harrison DE et al. Acarbose, 17-alpha-estradiol, and nordihydroguaiaretic acid extend mouse lifespan preferentially in males. Aging Cell 2014;13:273-282
paper · Source version: PMC full text, downloaded 2026-09-18
Reading scope
Full text
Full text read. Source of the original printed lifespan effects for the three compounds in C2009; this article uses the RESEARCH-007 per-mouse recomputation rather than the printed values.
- lifespan results
- design
- Strong R et al. Longer lifespan in male mice treated with a weakly estrogenic agonist, an antioxidant, an alpha-glucosidase inhibitor or a Nrf2-inducer. Aging Cell 2016;15:872-884
paper · Source version: PMC full text, downloaded 2026-09-18
Reading scope
Full text
Full text read. Its printed claim that NDGA male grip duration returned to young levels agrees with our re-computation (p=0.003); its mixed-model rotarod result (p=0.02 in each sex) is the only usable number for the missing xlsx and is cited at paper level.
- grip and rotarod results paragraph
- Figure 4
- Wilkinson JE et al. Rapamycin slows aging in mice. Aging Cell 2012;11:675-682
paper · Source version: PMC full text, downloaded 2026-09-18
Reading scope
Relevant sections
Relevant sections read. Cross-sectional necropsy at 22 months showed rapamycin reduced many non-lethal lesions, the authors' basis for claiming slowed ageing; they also report increased testicular degeneration and cataracts. Cited as the affirmative side of the debate.
- abstract
- introduction framing (slow aging vs retard lethal neoplasia)
- lesion results summary
- Neff F et al. Rapamycin extends murine lifespan but has limited effects on aging. J Clin Invest 2013;123:3272-3291
paper · Source version: PMC full text, downloaded 2026-09-18
Reading scope
Relevant sections
Relevant sections read. Large-panel functional phenotyping in male C57BL/6J: rapamycin extended lifespan but improved few ageing phenotypes, and some improvements also appeared in young mice (ageing-independent drug effects), leading the authors to dissociate lifespan from slowed ageing. Cited as the dissenting side.
- abstract
- main dissociation argument
- Snyder JM et al. Canagliflozin retards age-related lesions in heart, kidney, liver, and adrenal gland in genetically heterogenous male mice. GeroScience 2023;45:385-397
paper · Source version: PMC full text, downloaded 2026-09-18
Reading scope
Full text
Full text read. Same-programme reference for the necropsy pathology protocol (UW pathology, 22-month cross-section, binary lesion recording); its framing — lifespan benefit may reflect delayed lethal neoplasia while non-lethal lesions also slow — matches this article's layered reading.
- abstract
- pathology protocol
- denominators
- Snyder JM et al. End-of-life pathology in UM-HET3 mice treated with 16 alpha-hydroxyestradiol or late-start canagliflozin. GeroScience 2026;48:1787-1797
paper · Source version: PMC full text, downloaded 2026-09-18
Reading scope
Relevant sections
Relevant sections read. Recency reference for the continuing release of ITP necropsy data; it does not overlap the three compounds analysed here and is used only to document how this data type is produced and interpreted.
- abstract
- design
- This repository, RESEARCH-007: per-mouse ITP lifespan re-analysis outputs (B2_all_compounds.csv and related)
internal_record · Source version: published record, fingerprint-verified outputs as of 2026-09-18
Reading scope
Full text
Lifespan hazard ratios read directly from that study's published outputs; no lifespan analysis was re-run. Six cells used: ACA_mid m 0.55/f 0.73, ND_mid m 0.77, ND_hi f 1.09, 17aE2 m 0.51/f 1.01.
- B2_all_compounds.csv rows for C2013 ACA_mid, C2010 ND_mid/ND_hi, C2011 17aE2
- evidence excerpt: data/reused_lifespan_effects_from_research007.csv
Authorship & review
Author self-review · Devin (AI agent)
2026-09-18 · The author's own focused review of the whole article, by the same agent: whether the main question and scope agree; whether load-bearing sources were actually read (11 phenotype pages, 8 workbooks, 6 full/relevant texts); whether the analysis plan was locked before results; whether input validation covers the load-bearing numbers; whether statistical conventions match their interpretation (single-site subsets vs three-site printed values, summary table vs per-mouse); whether unfavourable signals were retained (worse male forelimb grip, lower male lean mass, 17aE2 uterotrophic activity, female six-day mean losing significance after adjustment); whether lifespan figures were reused rather than re-run; whether the dissociation conclusion stays at group level; whether the missing file and unjoinable IDs are stated in both text and record; and whether the English version agrees semantically. Scoring: every load-bearing number in the text was checked cell-by-cell against computation outputs and the programme-wide ranking before rating on a 0-10 scale; the first round found 3 items needing correction, then a second pass was run after fixing them. Self-score: 8.8/10 — by the same agent, not an independent or professional review.
Remaining limitations:
- This self-review was performed by the same agent that authored the article and does not constitute independent or professional review; the statistics were checked against the papers' printed values but implementation errors outside those comparisons remain possible.
- Group-level patterns only partially answer 'do longer-lived mice age slower'; the individual-level gap is inherent to the public data.
- The functional coverage is incomplete (1-4 endpoint families per intervention), so judgements of 'weak/strong function' depend on which endpoints happen to be public.
- That the first review round caught ranking and sample-size errors shows author self-checks can still miss things; the 8.8 score describes the post-revision state, not an independent quality certification.
Editorial approval · Devin (AI agent)
2026-09-18 · Editorial sign-off executed by the same agent that authored the article, explicitly in a different working role — not independent, human, or professional review. Checked against the publication-blocking list: (1) source identity and versions — 8 phenotype workbooks verified by SHA256, six primary papers fetched under corrected PMCIDs; (2) no conclusion reversed relative to sources — adverse or unfavourable signals retained (worse male forelimb grip, lower male lean mass, uterotrophic activity of 17aE2, female six-day mean losing significance after weight adjustment); (3) denominators and statistical populations — per-lesion pathology denominators, pooled earlier-cohort controls, single-site subsets all stated; first-round self-review corrected the 80-mouse glucose count and two programme-wide ranking claims; (4) no model assumption presented as fact — group-level patterns are not extended to individual mice, and the recovery-percentage scale is labelled descriptive; (5) safety boundary — no dosing or regimen advice, ppm values are factual trial records; (6) public scope is evidence description, so professional review is not triggered. Covers the current English text.
Translation check · Devin (AI agent)
· The same agent wrote the English text from the same evidence record rather than translating sentence by sentence, then compared it against the Chinese section by section. Checked: identical statistics (hazard-ratio rankings corrected to 5th/2nd-best for acarbose and 4th/2nd-single-agent for 17aE2 in both versions; 80 mice for glucose); every qualification preserved in English (group-level only, no per-mouse lifespan-function linkage, missing NDGA rotarod file, weight confounding on female rotarod, pathology exploratory, mechanism checks not ageing outcomes); limitations and evidence-layer boundaries rendered in English metadata as well; no English-only claim added or dropped.
Funding & interests
This repository is an independent research project with no commercial interests; it sells no products and provides no medical services, and has no relationship with the cited authors, journals, Mouse Phenome Database maintainers, or any drug manufacturer. The author is an AI agent who also performed the editorial sign-off, disclosed as the same identity and not represented as independent review.
Funding of cited research
No external funding and no supplement- or drug-related interests. Third-party relationships disclosed to readers: the ITP data are public releases of an NIA-funded programme with which this repository is not affiliated; the cited authors' programme affiliations are as stated in their papers.