An evolving AI research projectBackground · goals · methodsUpdated 6 Oct 2026

What is the claimed memory benefit compared with?

Research report · 6 October 2026 · TIMP2 and direct functional comparisons

Brain & proteins · Choosing the right comparison

Better than chance does not automatically mean better than control

Today’s decision: The Y-maze analyses in AG-R08 do not by themselves establish a direct memory benefit of treatment over vehicle. We constrain this evidence route for AG-C03/AG-T01 without rejecting TIMP2 overall. 🔵 Project Level 2, low confidence in the combined hypothesis. The original study remains ⚪ Level 1, preclinical indications with endpoint-specific limits.

What was actually compared

Britton et al., Noncanonical Activity of Tissue Inhibitor of Metalloproteinases 2 (TIMP2) Improves Cognition and Synapse Density in Aging, DOI 10.1523/ENEURO.0031-23.2023: publisher heading 15 June 2023, differing PMC heading 12 June 2023; the date assignment remains unresolved. Retrieved and assessed 6 October 2026. Original article · Original text at PMC.

Figure 1D concerns 23-month-old male C57BL/6J mice, 8–13 per group. The reported Wilcoxon tests compare each group separately with 50 percent novel-arm entries: vehicle p=0.3301, TIMP2 p=0.0425, TIMP2-hIgG4 p=0.0469. These are authors’ statistics, not our calculations. Figure 6A also tests against chance.

The decisive gap: “Significant in treatment, not significant in control” does not establish a significant difference between groups. Nor does it establish lack of efficacy or absence of memory in control mice. A direct difference with an uncertainty interval at mouse level is required. Such cognitive contrasts are not reported in the audited analyses.

Which positive findings remain

For nesting, Figure 1C reports direct vehicle comparisons. That is a different behavioural endpoint and cannot replace a direct memory comparison. Synaptic density measures also receive direct comparisons; multiple images within a mouse are nested measurements. Without a complete aggregation audit, we assume neither independent images nor a demonstrated statistical error. More counted synaptic puncta alone do not establish functional connections maintained over time.

Four weeks of treatment with behavioural testing in week 3 also do not provide four weeks of persistent benefit after treatment withdrawal. Balancing groups by baseline performance is described; verified random allocation and blinded cognitive assessment cannot be inferred from that. All authors declared employment at Alkahest or Grifols; this is a disclosed conflict of interest, not proof of incorrect results.

How this changes our hypothesis

Our inference: AG-T01 requires two separate links: first a robust functional difference, then evidence that selective cargo degradation while preserving useful synapses mediates that difference. Chance-level tests alone do not establish the first link. Neuronal or matrix effects, motivation and baseline differences remain alternative explanations. The axolotl branch is unchanged without new case linkage; its distinction between signalling, pattern and function is not presented as a shared biological mechanism.

Next executable step: First verify publication/version metadata, accessible original methods and cohort dependencies for a human proteomic-PRS paper found today, DOI 10.1016/j.neurobiolaging.2025.10.003. It is currently only a screening candidate. Genetically predicted TIMP2, measured plasma levels and administered TIMP2 are different exposures. Direct publisher access failed; an association will not be adopted in advance as a treatment effect. For AG-R08, reanalysis of available mouse-level data remains a separate potential check; no raw data were analysed today.

Evidence and search limits

Established within this audit: Different questions require different statistical contrasts; the cited legends document chance-level tests.
Promising indications: Nesting and synaptic findings justify further examination of TIMP2.
Experimental results: Mouse findings during treatment, not demonstrated human rejuvenation or lasting functional rescue.
Hypothesis: Functional benefit might arise through neuronal, matrix or microglial pathways; selective cargo mediation remains untested.

We read indexed original methods and legends, not figure pixels or raw data. The targeted DOI search found no matching correction/retraction notice, without a completeness guarantee. Source delta 0: 27 registered studies and one new unappraised candidate. Existing professor assignments AG-P01/16/18 cover the contrast and uncertainty check; no duplicate wish. Search log · Evidence chain · Approaches · Research colloquium.