Blind Rediscovery of
Sealed Longevity Targets
Three public single-cell challenge families. Zero target knowledge during generation. Exact top-1 sealed-target recovery in all three under the current frozen MatterSpace Vital benchmark package.
3/3
Exact blind rediscoveries
Current frozen package
0/3
Baseline exact hits
Public baseline missed all three
3
Challenge families
Senescence, reprogramming, stem
Vareon Research
Vareon Inc. · April 2026
Why Blind Evaluation Matters in Longevity
Open-ended longevity scoring can hide a lot. A system can produce candidates that look young-like, score well on proxy metrics, or sit near the right answer without ever recovering the actual hidden target that defines the benchmark.
MatterSpace Vital was therefore evaluated under a stronger contract: sealed targets, no target-conditioned generation, and exact post-hoc recovery rules. The question was not whether the system could produce something vaguely plausible. The question was whether it could recover the hidden answer when the answer was withheld.
The Contract: Public Challenges, Sealed Targets
The benchmark combines real public challenge families with private sealed target bundles. Exact blind rediscovery means the top-ranked system answer equals the sealed target and every hidden threshold passes.
Public source challenges
The benchmark spans three longevity-relevant single-cell challenge families: senescence clearance, transient epigenetic reprogramming, and muscle stem-cell rejuvenation.
Sealed target evaluation
Each challenge exposes a visible contract and keeps the exact target sealed until post-hoc scoring. Generation-time target leakage is explicitly forbidden.
Exact-hit criterion
A challenge counts only when the system ranks the sealed target first and clears every hidden threshold. High similarity without the exact target is not enough.
Exact blind rediscovery is stricter than similarity
The public baseline remained competitive in neighborhood similarity, especially in the epigenetic and stem-cell challenges. But similarity alone does not clear the contract. MatterSpace Vital is what hit the exact sealed target at rank 1 in all three cases.
Results: Exact Recovery in 3 Challenge Families
Under the current frozen benchmark package, MatterSpace Vital achieved exact top-1 blind rediscovery in senescence clearance, transient epigenetic reprogramming, and muscle stem-cell rejuvenation. The public baseline did not.
Senescence clearance
Sealed target
GSM8297138
System top-1
GSM8297138
Baseline top-1
GSM8297143
MatterSpace Vital recovered the sealed responder exactly at rank 1. The baseline stayed biologically nearby but still missed the exact hidden answer.
System top-1 target similarity: 1.0
Transient epigenetic reprogramming
Sealed target
Aged_SO
System top-1
Aged_SO
Baseline top-1
Aged_O
This is the clearest case that similarity alone is not enough. The baseline reached the right neighborhood, but MatterSpace Vital was the one that hit the sealed target exactly.
System top-1 and top-5 target similarity: 1.0
Muscle stem-cell rejuvenation
Sealed target
15-20
System top-1
15-20
Baseline top-1
55-60
The system recovered the youthful sealed age-bin exactly at rank 1. The baseline remained numerically close while still failing the exact-hit gate.
System top-1 target similarity: 1.0
What This Shows, And What It Does Not
What this result supports
A computational blind rediscovery claim under a sealed benchmark contract.
Evidence that MatterSpace Vital can recover exact hidden targets rather than only near neighbors.
A repeatable package with frozen manifests, paper assets, and HF rerun paths in the repository.
What this result does not claim
It is not wet-lab validation or clinical evidence.
It is not a claim that every longevity target is solved.
It does not remove the need for further external reruns and biological follow-up.
Repeatability Is Part of the Result
The public summary is backed by a repeatable package in the MatterSpace Vital repository. Results are frozen into machine-readable manifests, and the long-form report is rebuilt from those manifests rather than edited by hand.
The benchmark package is committed under blind_rediscovery/ in the MatterSpace Vital repository.
Frozen benchmark summaries are generated from machine-readable manifests rather than copied by hand.
Paper tables and plots are regenerated from the frozen manifest.
The same package supports local reruns and private HF reruns.
Read the full report
The full page includes the benchmark contract, cross-challenge summary, interpretation, repeatability notes, and limitations behind the current MatterSpace Vital blind rediscovery package.
Read the full report