diff --git a/docs/research/2026-05-28-composition-novelty-websearch-grounding-c302-emulator-multi-instance-dst-consensus-shader-substrate-prior-art-inventory-substantive-substrate-anchors-aaron-2026-05-28.md b/docs/research/2026-05-28-composition-novelty-websearch-grounding-c302-emulator-multi-instance-dst-consensus-shader-substrate-prior-art-inventory-substantive-substrate-anchors-aaron-2026-05-28.md
new file mode 100644
index 0000000000..ef771afaff
--- /dev/null
+++ b/docs/research/2026-05-28-composition-novelty-websearch-grounding-c302-emulator-multi-instance-dst-consensus-shader-substrate-prior-art-inventory-substantive-substrate-anchors-aaron-2026-05-28.md
@@ -0,0 +1,137 @@
+# Composition-novelty WebSearch grounding for B-0924 + B-0925 + research-mirror cluster — c302 + emulator + multi-instance DST + consensus + shader substrate prior-art inventory + substantive substrate-anchors (operator 2026-05-28; research-tier mirror per "preserve as research note (shadow*)" disposition)
+
+## Operator framing
+
+> *"websearch composition novelty (shadow*)"*
+> *"preserve as research note (shadow*) Aaron: feel free to push the three lanes forward"*
+
+Substrate-honest disposition: WebSearch-grounded composition-novelty research for the B-0924 (custom 2600 emulator + generate+join + DST + z-sets) + B-0925 (C. elegans-substrate controller variant) + research-mirror (three-layer biological instantiation) substrate-engineering substrate-target cluster + the shader-emulator substrate-engineering substrate-research question.
+
+Preserved at research-tier (parallel to PR #5892 mirror) per proper-research-first discipline operator named ("we don't want to just code that up without proper research").
+
+## Component-by-component prior-art (WebSearch-grounded 2026-05-28)
+
+| Component | Prior-art status | Substrate-anchor |
+|---|---|---|
+| **OpenWorm c302** | Well-established (Gleeson et al. 2018) | Can drive body movement in Sibernetic; NO Atari game-controller substrate found |
+| **Biological-spiking-NN game-controller** (artificial-NN with biological topology) | ESTABLISHED | Anwar et al. 2022 (racket-ball); BT-SNN with Allen mouse brain topology subsets (2024); use ARTIFICIAL spiking-NNs with biologically-inspired topology, NOT actual biological connectome substrate |
+| **GPU-accelerated Atari emulation** | SUBSTANTIAL PRIOR-ART | NVIDIA CuLE (Dalton et al. 2019, [arXiv:1907.08467](https://arxiv.org/pdf/1907.08467)) — CUDA port of ALE; 40-190M frames/hour single GPU; renders frames directly on GPU avoiding CPU-GPU bottleneck |
+| **Compute-shader Atari emulator** | UNCLEAR | CuLE is CUDA kernel, NOT compute-shader; full-shader-emulator: unknown; likely PoCs exist somewhere |
+| **Bit-perfect deterministic AI substrate** | ESTABLISHED | EigenAI (2026) — bit-exact deterministic AI on GPUs <2% overhead; "verifiable autonomous AI agents"; different scope (LLM inference) but composes substrate-engineering substrate-engineering substrate |
+| **Multi-instance AI consensus / verification** | ESTABLISHED | n-way beam search consensus; majority-vote between independent verifiers; gossip+sync training |
+| **Z-set joins over game-state-trajectories + IScheduler DST** | NOT FOUND in mainstream | Framework substrate (B-0924 + B-0925) |
+| **Lightlike-substrate ray-tracing-over-generator-time as architecture-design-rule** | NOT FOUND | Novel framework substrate-discipline (PR #5910 + PR #5912) |
+
+## Composition-novelty assessment (substrate-honest don't-collapse)
+
+Applied Amara's blade (per `.claude/rules/god-tier-claims-high-signal-high-suspicion-dont-collapse.md` + Amara's "rhymes ≠ validates" discipline per PR #5910 + #5912):
+
+| Claim scope | Substrate-honest answer |
+|---|---|
+| "First GPU-accelerated Atari emulator" | **NO** — NVIDIA CuLE shipped 2019 |
+| "First compute-shader-based 2600 emulator" | **UNKNOWN** — likely PoCs exist; specific composition unclear |
+| "First biological-spiking-NN game controller" | **NO** — Anwar et al. 2022 + BT-SNN 2024 |
+| "First c302-as-Atari-controller" | **LIKELY YES** — no prior-art found; specific composition novel |
+| "First worm-plays-Atari" | **LIKELY YES** — substrate-anchor for B-0925 implementation |
+| "First framework composing (full biological connectome + emulator + DST + z-set generate+join + lightlike-substrate architecture-design-rule) under wake-time-substrate-discipline" | **LIKELY YES** — operationally bounded to composition + substrate-discipline scope |
+
+The substrate-honest "first" claim that SURVIVES razor:
+
+The "first" claim is operationally bounded to the **specific composition + framework substrate-discipline** — NOT to individual components. Each component has prior-art; the composition + the substrate-engineering substrate-engineering substrate-discipline (NCI + multi-oracle + asymmetric-authorship + monad-propagation + retraction-native + glass-halo + ray-tracing-over-generator-time + lightlike-vs-dark design-rule + wake-time-substrate) applied to this specific composition is what's novel.
+
+## Substantive substrate-anchors for future implementation work
+
+Composes with B-0924 + B-0925 + research-mirror cluster:
+
+| Substrate-anchor | Implementation substrate it provides |
+|---|---|
+| **NVIDIA CuLE** ([Dalton et al. 2019, arXiv:1907.08467](https://arxiv.org/pdf/1907.08467) + [NVIDIA blog](https://developer.nvidia.com/blog/new-open-source-gpu-accelerated-atari-emulator-for-reinforcement-learning-now-available/)) | B-0924 GPU-emulator throughput baseline (40-190M frames/hour single GPU); CUDA kernel substrate; composes with operator's 5090 + 25-GPU distributed framing |
+| **Anwar et al. 2022 racket-ball spiking-NN** ([PLOS One](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0265808)) | B-0925 biological-spiking-NN-controller architecture substrate; visual-motor cortex architecture + reinforcement-learning training substrate |
+| **OpenWorm c302** ([PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC6158223/) + [GitHub](https://github.com/openworm/c302)) | Canonical biological connectome substrate; multiscale framework for nervous-system modeling |
+| **EigenAI 2026** ([EigenCloud blog](https://blog.eigencloud.xyz/deterministic-ai-inference-eigenai/)) | Bit-perfect deterministic-AI substrate-anchor; <2% overhead on GPUs; composes with IScheduler DST substrate |
+| **BT-SNN Allen mouse brain topology** ([Frontiers in Neuroscience 2024](https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2024.1325062/full)) | Biological-connectome-derived NN substrate at different organism scope; substrate-engineering substrate-engineering substrate-target reference |
+| **Biologically plausible model-based RL in recurrent spiking networks** ([Nature Sci Reports 2024](https://www.nature.com/articles/s41598-024-65631-y)) | Dreaming + spiking-network substrate; composes with generate+join discipline |
+
+## Sub-target candidates this grounding enables
+
+For B-0924 (custom 2600 emulator implementation):
+
+1. **Use CuLE as substrate-anchor** for GPU-accelerated emulation baseline (don't reinvent; compose with operator's 5090 + 25-GPU distributed)
+2. **Compute-shader port of CuLE substrate** — substrate-engineering substrate-engineering substrate-target if shader-based scaling beats CUDA-kernel substrate (research-tier; not yet authorized)
+3. **IScheduler DST wrapper** around CuLE for deterministic-trajectory substrate
+
+For B-0925 (C. elegans-substrate controller variant):
+
+1. **OpenWorm c302 as substrate-anchor** for per-individual biological connectome substrate
+2. **Sensorimotor mapping substrate**: c302 sensory-neurons ↔ 2600 TIA frame-buffer; c302 motor-neurons ↔ 2600 PIA joystick-substrate
+3. **Compose with Anwar et al. substrate** for biological-spiking-NN-RL training architecture
+
+For research-mirror (three-layer biological instantiation):
+
+1. **Compose with EigenAI bit-perfect substrate** for multi-instance consensus testing infrastructure
+2. **Compose with BT-SNN substrate** for cross-organism connectome substrate-engineering substrate-engineering substrate
+
+## Substrate-honest framing
+
+This research note operates at MIRROR-TIER per `.claude/rules/substrate-or-it-didnt-happen.md` + `.claude/rules/razor-discipline.md`:
+
+**Mirror-tier (preserved verbatim)**:
+
+- WebSearch-grounded prior-art inventory (component-by-component)
+- Composition-novelty assessment with substrate-honest don't-collapse
+- Substrate-anchors for future implementation work
+- Operator's substrate-disposition ("preserve as research note (shadow*)")
+
+**Beacon-tier (NOT claimed at this tier; would require empirical validation)**:
+
+- Whether c302-as-Atari-controller actually beats LLM-imitation-learning baselines (empirical test required)
+- Whether shader-port of CuLE substantively beats CUDA-kernel throughput (engineering test required)
+- Whether the composition substrate-engineering substrate-engineering substrate-target scales to practical AI-training throughput (engineering test required)
+
+Per Amara's blade (PR #5910 + #5912): "first" claim at composition + substrate-discipline scope earns its keep as substrate-engineering substrate-recognition; "first" claim at individual-component scope FAILS razor (component prior-art exists).
+
+## Composes with substrate
+
+- PR #5890 (B-0924 custom 2600 emulator + generate+join + DST + z-sets) — direct substrate-anchor target
+- PR #5891 (B-0925 C. elegans-substrate controller variant) — direct substrate-anchor target
+- PR #5892 (research-tier mirror three-layer biological instantiation) — parallel research-tier preservation
+- PR #5910 (Amara future-affects-generator + three-clocks) — Amara's blade discipline composes
+- PR #5912 (Amara lightlike-kind-substrate + consensus-is-gravity + design-rule) — lightlike-substrate architecture-design-rule composes with shader-emulator substrate
+- PR #5778 (Persist-as-bridge / μένω substrate) — substrate-anchor for temporal substrate
+- PR #5841 (DST + Persist + generator-time + feedback) — generator-time substrate-anchor
+- B-0703 multi-oracle BFT — composes with multi-instance AI consensus substrate
+- B-0917 Kleisli interrupt substrate — composes with hardware-interrupt substrate (B-0924)
+- B-0664 NCI HC-8 — composes with non-coercive substrate at multi-instance consensus scope
+- Operator's 5090 + 25-GPU distributed framing — substrate-target context for shader-emulator composition
+
+## Composes with rules
+
+- `.claude/rules/god-tier-claims-high-signal-high-suspicion-dont-collapse.md` — Amara's blade operationalized at substrate-rhyme + first-claim scope
+- `.claude/rules/razor-discipline.md` — operational claims only; "first" claim at composition + substrate-discipline scope earns its keep
+- `.claude/rules/substrate-or-it-didnt-happen.md` — preserved at mirror-tier per substrate-honest discipline
+- `.claude/rules/refresh-before-decide.md` — WebSearch-grounding IS refresh-before-decide at substrate-claim scope
+- `.claude/rules/dep-pin-search-first-authority.md` — extended to substrate-engineering substrate-claim scope
+- `.claude/rules/verify-existing-substrate-before-authoring.md` — composition-novelty assessment IS substrate-inventory pass at substrate-claim scope
+- `.claude/rules/past-is-kind-when-lightlike-consensus-is-gravity-lightlike-vs-dark-architecture-design-rule-amara-aaron-2026-05-28.md` — lightlike-substrate design-rule composes with shader-emulator substrate
+
+## Sources
+
+- [c302: a multiscale framework for modelling the nervous system of Caenorhabditis elegans — PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC6158223/)
+- [OpenWorm: overview and recent advances in integrative biological simulation — PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC6158220/)
+- [OpenWorm c302 GitHub](https://github.com/openworm/c302)
+- [OpenWorm Publications](https://openworm.org/publications.html)
+- [Training a spiking neuronal network to play virtual racket-ball game (Anwar et al.) — PLOS One](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0265808)
+- [Brain topology improved spiking neural network for efficient RL — Frontiers in Neuroscience 2024](https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2024.1325062/full)
+- [Towards biologically plausible model-based RL in recurrent spiking networks by dreaming — Nature Scientific Reports 2024](https://www.nature.com/articles/s41598-024-65631-y)
+- [Accelerating Reinforcement Learning through GPU Atari Emulation (CuLE, Dalton et al.) — arXiv:1907.08467](https://arxiv.org/pdf/1907.08467)
+- [NVIDIA: New Open Source GPU-Accelerated Atari Emulator for RL](https://developer.nvidia.com/blog/new-open-source-gpu-accelerated-atari-emulator-for-reinforcement-learning-now-available/)
+- [Deterministic Inference: EigenAI bit-exact AI verification (2026)](https://blog.eigencloud.xyz/deterministic-ai-inference-eigenai/)
+- [Deterministic + Agentic AI Architecture (2026)](https://thehackernews.com/2026/04/deterministic-agentic-ai-architecture.html)
+- [Generative AI Reproducibility and Consensus — arXiv:2307.01898](https://arxiv.org/pdf/2307.01898)
+- [Can AI Agents Agree? — arXiv:2603.01213](https://arxiv.org/pdf/2603.01213)
+
+## Full reasoning
+
+Operator 2026-05-28 conversation thread following the worm-plays-Atari substrate-engineering substrate-target cluster (B-0924 + B-0925 + research-mirror + Amara generator-time + lightlike-substrate rules). Operator asked "or will ours be the first one?" + authorized "websearch composition novelty (shadow*)" + "preserve as research note (shadow*)" disposition. This research note lands the WebSearch-grounded composition-novelty assessment at research-tier per proper-research-first discipline.
+
+Future-Otto + Alexa + Riven + Vera + Lior cold-booting from this research note inherit the substrate-grounded prior-art inventory + substrate-honest composition-novelty assessment + substantive substrate-anchors for future implementation work. The substrate-engineering substrate-engineering substrate-claim that "we're first" earns its keep ONLY at the specific composition + framework substrate-discipline scope; individual-component "first" claims fail razor per established prior-art.