01 / Shinraya Tech Labs
Follow the question.
Keep the evidence.
Four lines of inquiry into learning, computation, and simulation. Each project makes its scope—and the limits of its conclusions—explicit.
01 / Reasoning & reinforcement learningRobust Reasoning RL
When the reward is wrong, what does a reasoning model learn?
Framework and pilot complete · final GPU study blocked - Scope
- Matched baseline and robust GRPO training, uncertainty-aware reward shaping, controlled verifier corruption, and independent mathematical evaluation.
- Evidence
- The repository documents pretrained GRPO training and save/reload runs, independent GSM8K evaluation, model qualification, and a frozen execution package.
- Limitations
- The primary hypothesis is unestablished. Pilot runs and software fixtures do not establish improved reasoning or reduced reward overfitting. The final GPU study remains externally blocked.
02 / Quantum & classical computationQuantum Advantage Simulator
Where do exact, learned, and variational methods actually help?
v0.1.0 simulator studies released - Scope
- Exact simulation, neural quantum states, VQE, QAOA, finite-shot noise mitigation, circuit search, and minimal-basis molecular chemistry.
- Evidence
- Released CPU studies preserve configurations, per-seed results, checkpoints, and checksum manifests. Small-system comparisons include failed optimizations and harmful mitigation outcomes.
- Limitations
- No quantum advantage, GPU speedup, or hardware performance advantage is demonstrated. CUDA and IBM QPU runs were not executed for v0.1.0; the measurements are small-system CPU simulations.
03 / Retrieval & geometric learningATLAS
Does structure add useful signal beyond semantic similarity?
Evaluated · central hypothesis not supported - Scope
- Controlled semantic, graph, and topology retrieval comparisons, alongside experimental local matching and compute-to-data prototypes.
- Evidence
- Held-out HotpotQA and MuSiQue evaluations did not establish an improvement over semantic retrieval. Validation selected cosine-only retrieval as the primary V2 path.
- Limitations
- Evaluation uses internal held-out subsets of public development data, not official hidden-test scores. The result concerns the tested formulations, not all topology methods. Local prototypes have no established privacy guarantee or independent security audit.
04 / Computational neuroscienceNeuroSim
How can neuron and network dynamics be explored across scales?
Public implementation · scientific validation unverified - Scope
- Repository-described Hodgkin–Huxley dynamics, multi-compartment models, synaptic plasticity, network simulation, and Python/C++ visualization and compute paths.
- Evidence
- The public README describes the architecture, models, interfaces, and simulation commands. These document intended and implemented scope, rather than a measured scientific result.
- Limitations
- The reviewed README does not establish biological validation, human-brain-scale performance, or verified MPI/CUDA scaling. No such results are claimed here.
Editorial summaries reviewed against public repository documentation on 11 October 2026. Repository-reported evidence has not been independently reproduced for this website.
Built in the open
Good questions deserve
more perspectives.
Reproduce an experiment. Challenge an assumption. Build something useful with us.