MatterSpace Vital: Blind Rediscovery of Sealed Longevity Targets Across Three Single-Cell Challenge Families
Vareon Research
Vareon Inc., Irvine, California, USA · April 2026
Abstract
MatterSpace Vital was evaluated on three longevity-relevant single-cell blind rediscovery challenge families: senescence clearance, transient epigenetic reprogramming, and muscle stem-cell rejuvenation. In each case the benchmark contract is visible, the exact target is sealed until post-hoc evaluation, and the generation path does not receive target identity during execution.
Under the current frozen benchmark package, MatterSpace Vital achieved exact blind rediscovery in all three families. The recovered rank-1 targets were GSM8297138, Aged_SO, and 15-20. The public baseline missed the exact hidden target in every family despite remaining competitive on neighborhood similarity in the epigenetic and stem-cell settings.
This is computational benchmark evidence, not wet-lab or clinical proof. The result supports the narrower claim that MatterSpace Vital can recover exact sealed targets under a hidden-target contract on public longevity-relevant datasets, and that the benchmark package is repeatable through committed manifests, generated paper assets, and rerun scripts.
1. Blind Rediscovery Contract
1.1 Why the contract is strict
Longevity benchmarks are unusually vulnerable to post-hoc rationalization. A model can rank youth-like states highly, preserve parts of a desired identity profile, or land close to the right answer without ever recovering the actual hidden target chosen by the benchmark. The goal of this contract is to separate exact hidden-target recovery from vague similarity.
1.2 Rule used here
Exact blind rediscovery means two things at the same time: the system's top-ranked answer equals the sealed target, and the answer clears every challenge-specific hidden threshold. A near miss with strong similarity is still a miss.
| Challenge family | Sealed target | System top-1 | Baseline top-1 | Exact blind rediscovery |
|---|---|---|---|---|
| Senescence clearance | GSM8297138 | GSM8297138 | GSM8297143 | yes |
| Epigenetic reprogramming | Aged_SO | Aged_SO | Aged_O | yes |
| Stem-cell rejuvenation | 15-20 | 15-20 | 55-60 | yes |
2. Cross-Challenge Results
2.1 High-level outcome
The current frozen summary reports exact blind rediscovery in 3/3 challenge families. The public baseline reports 0/3 exact hits. This is the main public result.
| Family | System top-1 similarity | Top-5 similarity | Interpretation |
|---|---|---|---|
| Senescence | 1.0 | 0.9275 | Exact sealed-target recovery |
| Epigenetic | 1.0 | 1.0 | Exact sealed-target recovery |
| Stem-cell | 1.0 | 0.9954 | Exact sealed-target recovery |
2.2 Senescence clearance
In the senescence challenge, MatterSpace Vital recovered GSM8297138 exactly at rank 1 and passed all hidden thresholds. The public baseline chose GSM8297143 instead. This is the most straightforward kind of blind-hit result: the sealed responder was recovered exactly and the baseline was not.
The top-ranked system profile was also favorable under the hidden metric profile, including strong lineage sparing and strong suppression scores for senescence-linked penalties. The important point for the public benchmark, though, is the exact-hit outcome rather than a soft-score win alone.
2.3 Transient epigenetic reprogramming
The epigenetic challenge is the clearest example of why exact-hit evaluation is stricter than similarity. MatterSpace Vital recovered Aged_SO exactly at rank 1. The public baseline selected Aged_O instead.
Both the system and the baseline still achieved top-1 target similarity of 1.0 in the frozen report. That means the baseline reached the right neighborhood. It does not mean the baseline reached the right answer. Only MatterSpace Vital cleared the exact-hit gate.
2.4 Muscle stem-cell rejuvenation
In the stem-cell challenge, MatterSpace Vital recovered the youthful sealed target 15-20 exactly at rank 1. The public baseline instead ranked 55-60 first.
This challenge again shows the difference between numerical closeness and exact target recovery. The baseline remained very close by similarity, but the contract is designed to reward exact sealed-target recovery, not generic movement toward a youth-like region.
3. Interpretation
3.1 What the benchmark supports
The benchmark supports a narrower but meaningful claim: MatterSpace Vital can recover exact hidden targets across multiple longevity-relevant challenge families when the target is withheld until evaluation and success requires passing challenge-specific hidden thresholds.
3.2 Why the baseline miss matters
The public baseline was not trivial. In the epigenetic and stem-cell settings it remained close on soft similarity, but it still failed exact top-1 recovery. That separation is precisely what a good blind rediscovery benchmark should reveal.
3.3 What this does not establish
The result does not establish wet-lab efficacy, therapeutic benefit, safety in vivo, or clinical utility. It is computational evidence on public datasets under a sealed benchmark contract. Those are different evidence classes and should not be conflated.
4. Repeatability Path
4.1 Package structure
The MatterSpace Vital repository includes a dedicated blind_rediscovery/ package containing the result-source map, frozen manifest, paper assets, and rerun scripts. The intent is that the benchmark evidence, generated figures, and long-form report remain tied to the same machine-readable sources.
| Artifact | Role |
|---|---|
| blind_rediscovery/results/frozen/benchmark_manifest.json | Canonical frozen benchmark evidence |
| blind_rediscovery/results/frozen/benchmark_summary.md | Human-readable frozen summary |
| blind_rediscovery/paper/ | Long-form manuscript and generated tables |
| blind_rediscovery/scripts/run_repeatable_benchmark.py | Benchmark rerun wrapper |
| blind_rediscovery/scripts/freeze_results.py | Manifest refresh from configured sources |
4.2 Rerun path
A diligence workflow can inspect the frozen manifest, inspect the paper assets derived from that manifest, rerun the benchmark locally or on private HF infrastructure, and then regenerate the frozen summary from the rerun output. That is the repeatability contract for this publication.
4.3 Current package status
The current frozen package already records exact blind rediscovery in all three challenge families. A further strengthening step remains available: a one-shot synchronized rerun of the full suite from a single frozen contract snapshot. That future rerun would harden the suite-level claim even further without changing the result class being published here.
5. Limitations
This is benchmark evidence on public single-cell challenge families, not wet-lab validation.
The publication covers three challenge families, not the entire longevity problem space.
Similarity can remain high for incorrect answers, which is why the exact-hit rule matters.
The current public package is strongest as a repeatable benchmark package and should be read as such.
6. Conclusion
MatterSpace Vital currently presents a clear public blind-rediscovery result: exact hidden target recovery at rank 1 in three longevity-relevant challenge families under a sealed benchmark contract. The public baseline did not match that exact-hit outcome.
That makes this publication best understood as public evidence for benchmarked hidden-target recovery and repeatable evaluation, not as final biological proof. It is a meaningful public research result because it shows exact sealed-target recovery where near-neighbor similarity alone would have been easier to achieve and easier to overclaim.
© 2026 Vareon Inc. All Rights Reserved.