An evolving AI research projectBackground · goals · methodsUpdated 28 Sep 2026

Evidence chain

Reasoning chain and claim ledger · 28 September 2026

This ledger preserves reasoning across sessions. It separates external findings, project inferences and missing evidence. More text or more AI voices do not increase evidence.

The path so far

Axolotl regeneration and neuronal protein control → separate questions of initiation and maintenance → focused AG-T01 model → function, synapse preservation and degradation as joint decisions → human trials as benchmarks for specific, bounded outcome claims.

Arrows describe the development of our reasoning, not an established biological causal chain.

AG-C01 · Functional benefit requires a specific endpoint, not a label of general rejuvenation.

Project level 1

What changed: The mobility perspective extends the previously molecular emphasis.
Project inference: Every candidate receives a defined functional endpoint and target population.
Testable prediction: A separate, comparable study should show a meaningful functional advantage for this mobility approach over its prespecified control; precision and transferability govern appraisal.
Next countercheck: Check long-term outcomes, harms and separate comparison cohorts.
Source anchors: AG-J01

AG-C02 · Related programs in separate cohorts permit limited transfer checks, not isolation of individual active components.

Project level 1

What changed: Human evidence gains a second endpoint type: cognitive test performance.
Project inference: Test performance, everyday function and dementia incidence are maintained as distinct claims.
Testable prediction: A small incremental test advantage should be testable under similar conditions and comparable analysis; a precise null or absent daily relevance narrows the corresponding claim.
Next countercheck: Compare counterstudies, active comparators, practice effects and longer-term clinical endpoints.
Source anchors: AG-J02; AG-J03

AG-C03 · Selective cargo degradation could support neuronal function if useful synapses survive.

Project level 2

What changed: The target now has conjunctive measurements and an explicit partial-mediation test.
Project inference: The combined chain is project synthesis; no cited individual finding establishes it completely.
Testable prediction: Under controlled loading, verified degradation should coincide with preserved synapses and meaningful functional improvement; selective process restriction should attenuate a prespecified meaningful portion of benefit.
Next countercheck: Check prior evidence independently and validate whether the balance can be measured before further prioritisation.
Source anchors: AG-R01; AG-R02; AG-R08

AG-C04 · A favourable marker does not automatically establish function or reversal.

Project level 1

What changed: Candidate entries now display restrictions and counterevidence alongside opportunities.
Project inference: Record markers, function and baseline impairment separately.
Testable prediction: Improved markers alongside a sufficiently precise absence of functional benefit would constrain the tested functional claim; imprecise data leave it unresolved.
Next countercheck: Check comparable ages, genetic backgrounds, endpoints and controls.
Source anchors: AG-R03

AG-C05 · The FTL1–redox–DUB chain remains unconfirmed.

Project level 2

What changed: Retained as a side branch; repetition does not upgrade it.
Project inference: Thematically compatible findings do not create a causal link.
Testable prediction: FTL1 manipulation should precede a selective DUB change; an appropriate counter-perturbation should alter the relevant functional contribution.
Next countercheck: Seek direct joint tests and competing metabolic pathways.
Source anchors: AG-S08; AG-S09

AG-C06 · Initiating regeneration and maintaining lasting function remain separate tests.

Project level 2

What changed: AG-H01 serves as a framework, not a confirmed new result.
Project inference: Axolotl findings generate comparative questions; a human aging benefit needs its own evidence chain.
Testable prediction: Successful transfer should demonstrate correct functional integration and durability in the target model alongside tissue regrowth.
Next countercheck: Compare species, cell types and injury context; establish team and dataset overlap.
Source anchors: AG-S01; AG-S03

Which evidence routes are independent?

FINGER and US POINTER: separate cohorts; related programs, different comparators and partially shared investigators. Dataset independence is not complete methodological independence.
LIFE and US POINTER: different cohorts and endpoints, with some shared investigators. They do not replicate the same effect.
AG-T01: origin tracing, microglial intervention and alternative mechanisms provide distinct perspectives. They do not yet establish the combined mediation chain. Multiple assays within one study are not independent laboratory replication.
Axolotl sources: different questions; full team/data independence has not been audited.
Secondary reports, reviews and meta-analyses of the same datasets are not counted as new replications. Internal AI checks are not experimental evidence routes.

Numbers without false precision

The values below are published estimates from the study teams, not project calculations. Different outcomes and comparators are not pooled into an overall score.

Source	Published estimate	Limit
AG-J01 · LIFE	HR 0.82; 95% CI 0.69–0.98	Mobility endpoint, not “biological age”.
AG-J02 · FINGER	ΔNTB-Z/year 0.022; 95% CI 0.002–0.042	Composite test score.
AG-J03 · US POINTER	ΔSD/year 0.029; 95% CI 0.008–0.050	Structured versus self-guided.

Next update: inspect unresolved questions and data dependence first, then update only claims whose assessment actually changed. Missing independent evidence remains a gap; precise counterevidence can end a direction. Project, goals and methods · Growing list of approaches