Projects, reports, experiments, and executable research evidence.https://shubhamattri.com/https://shubhamattri.com/research/agora-ctm-shadow-stage1/https://shubhamattri.com/research/agora-ctm-shadow-stage1/Held-out CompactCTM shadow predictor fails to beat an empirical prior on Agora opinion-diffusion transitions; claim class is non-parity, control authority false, not promoted.Thu, 23 Jul 2026 00:00:00 GMTagent-based-simulationcontinuous-thought-machinescalibrationnegative-resulthttps://shubhamattri.com/research/five-node-intervention/https://shubhamattri.com/research/five-node-intervention/A preregistered three-seed pilot on five-node wave-causal circuits: the phase-cancelling circuit is real, but none of the acquisition-strategy superiority claims survived.Thu, 23 Jul 2026 00:00:00 GMTcausalitycomputational-papermarimohttps://shubhamattri.com/research/continuous-thought-machines/https://shubhamattri.com/research/continuous-thought-machines/A source-grounded dissection of Sakana AI's Continuous Thought Machines, paired with an inspectable parity protocol and an explicitly disputed local reproduction state.Thu, 23 Jul 2026 00:00:00 GMTcontinuous-thought-machinesmechanistic-interpretabilityresearch-methodmarimohttps://shubhamattri.com/research/gigatoken-reproduction/https://shubhamattri.com/research/gigatoken-reproduction/Independent Apple M5 Max reproduction of Gigatoken's speed claims: 133× vs HuggingFace on GPT-2, 42× vs tiktoken, exact token-ID match — plus an educational recreation of the two systems levers.Thu, 23 Jul 2026 00:00:00 GMTtokenizationsimdsystemsbenchmarksmarimohttps://shubhamattri.com/research/pggolf-leaderboard/https://shubhamattri.com/research/pggolf-leaderboard/The live leaderboard of Parameter Golf training runs: every experiment's quantized validation BPB, synced back from remote GPU machines.Thu, 23 Jul 2026 00:00:00 GMTtraininglanguage-modelsmarimohttps://shubhamattri.com/research/polymath-research-atlas/https://shubhamattri.com/research/polymath-research-atlas/Portfolio evidence-pipeline audit: transparent stage flags and metadata-backed bridge retrieval over a personal research vault — not a quality score.Thu, 23 Jul 2026 00:00:00 GMTresearch-opsmarimoknowledge-graphportfoliohttps://shubhamattri.com/research/rsi-lab-portfolio-map/https://shubhamattri.com/research/rsi-lab-portfolio-map/How Club A evidence engines and Club B method infrastructure connect into a verifier-gated recursive discovery program — and what is still not claimed.Thu, 23 Jul 2026 00:00:00 GMTrecursive-self-improvementportfolioverificationhttps://shubhamattri.com/research/technique-explorer/https://shubhamattri.com/research/technique-explorer/An interactive catalog of training techniques for small language models, filterable by how early each one shows measurable signal.Thu, 23 Jul 2026 00:00:00 GMTtraininglanguage-modelsmarimohttps://shubhamattri.com/research/thirty-seed-uncertainty/https://shubhamattri.com/research/thirty-seed-uncertainty/A hash-verified thirty-seed confirmatory study: the selector manipulation passed, but all four preregistered superiority claims were killed at their frozen gates.Thu, 23 Jul 2026 00:00:00 GMTcausalitycomputational-papermarimohttps://shubhamattri.com/research/open-problems-lab-intake/https://shubhamattri.com/research/open-problems-lab-intake/Verifier-first Lean research program: 15 problem records, 29 prompts, immutable target hashes, and a blocked Mathlib CI gate — zero proofs promoted.Wed, 15 Jul 2026 00:00:00 GMTleanformal-methodsopen-problemsverificationhttps://shubhamattri.com/research/measure-before-you-predict/https://shubhamattri.com/research/measure-before-you-predict/A collapse-proof text encoder is anti-forecasting by construction: the working paper that killed its own premise with multi-seed margins and DMD diagnostics, in minutes of compute.Sat, 20 Jun 2026 00:00:00 GMTrepresentation-learningjepaforecastingmlxhttps://shubhamattri.com/research/agora-calibration-runs/https://shubhamattri.com/research/agora-calibration-runs/State-first simulation calibration on opinion diffusion and SIR worlds: scored forecasts and rule evolution against explicit targets, not vibes.Fri, 19 Jun 2026 00:00:00 GMTagent-based-simulationcalibrationstate-first-aihttps://shubhamattri.com/research/sansara-operating-surface/https://shubhamattri.com/research/sansara-operating-surface/Vault-as-universe agent OS with a Rust phase-graph runtime: 70 GiB bloat cut, 152 tests, path-traversal hardening, and a council layer shared with Geist.Thu, 23 Jul 2026 00:00:00 GMTagent-osrustobsidian-vaultdspyhttps://shubhamattri.com/research/lorenz-verifier-discipline/https://shubhamattri.com/research/lorenz-verifier-discipline/A paused Lean RLM harness whose export is proof-integrity discipline — scheduler, planner, and population store that only admits verifier-surviving artifacts — not claimed theorems.Thu, 28 May 2026 00:00:00 GMTleanrlmproof-searchverificationhttps://shubhamattri.com/research/geist-council-protocol/https://shubhamattri.com/research/geist-council-protocol/Protocol-first durable agent research: confirmation gates, silent no-ops, and a council loop that writes outcomes to long-term memory instead of chat scrollback.Fri, 01 May 2026 00:00:00 GMTagentscompany-osmemoryprotocolhttps://shubhamattri.com/research/alphaevolve-early-generations/https://shubhamattri.com/research/alphaevolve-early-generations/Open AlphaEvolve infrastructure and scaffolds shipped; recorded evolutionary runs never progressed past early generations before the project paused.Tue, 01 Jul 2025 00:00:00 GMTevolutionary-computationllm-agentscode-generation