diff --git a/docs/BACKLOG.md b/docs/BACKLOG.md
index c383c06a8b..29ec67bb91 100644
--- a/docs/BACKLOG.md
+++ b/docs/BACKLOG.md
@@ -861,6 +861,7 @@ are closed (status: closed in frontmatter)._
 - [ ] **[B-0920](backlog/P2/B-0920-memory-lifetime-du-agent-initiated-cleanup-with-history-preservation-drafted-active-superseded-archived-retracted-aaron-2026-05-28.md)** MemoryLifetime DU — agent-initiated cleanup with history preservation (drafted / active / superseded / archived / retracted) — sibling to B-0919 MemoryBinding (Aaron 2026-05-28)
 - [ ] **[B-0921](backlog/P2/B-0921-integrate-or-remove-unreferenced-cayleydickson.md)** Integrate or remove unreferenced file src/Core/CayleyDickson.fs
 - [ ] **[B-0922](backlog/P2/B-0922-integrate-or-remove-unreferenced-kskauthorization.md)** Integrate or remove unreferenced file src/Core/Consent/KskAuthorization.fs
+- [ ] **[B-0924](backlog/P2/B-0924-custom-2600-emulator-generate-join-over-emulator-scene-ischeduler-dst-bit-perfect-consensus-z-sets-arc3-agi-training-hardware-interrupts-b0917-aaron-2026-05-28.md)** Custom Atari 2600 emulator + Generate+Join over emulator scene (vs imitation-learning) + IScheduler DST bit-perfect-consensus via z-sets + hardware interrupts via B-0917 substrate + ARC3-AGI training surface (operator 2026-05-28)
 
 ## P3 — convenience / deferred
 
diff --git a/docs/backlog/P2/B-0924-custom-2600-emulator-generate-join-over-emulator-scene-ischeduler-dst-bit-perfect-consensus-z-sets-arc3-agi-training-hardware-interrupts-b0917-aaron-2026-05-28.md b/docs/backlog/P2/B-0924-custom-2600-emulator-generate-join-over-emulator-scene-ischeduler-dst-bit-perfect-consensus-z-sets-arc3-agi-training-hardware-interrupts-b0917-aaron-2026-05-28.md
new file mode 100644
index 0000000000..02bd61f148
--- /dev/null
+++ b/docs/backlog/P2/B-0924-custom-2600-emulator-generate-join-over-emulator-scene-ischeduler-dst-bit-perfect-consensus-z-sets-arc3-agi-training-hardware-interrupts-b0917-aaron-2026-05-28.md
@@ -0,0 +1,139 @@
+---
+id: B-0924
+priority: P2
+status: open
+title: Custom Atari 2600 emulator + Generate+Join over emulator scene (vs imitation-learning) + IScheduler DST bit-perfect-consensus via z-sets + hardware interrupts via B-0917 substrate + ARC3-AGI training surface (operator 2026-05-28)
+effort: XL
+ask: operator 2026-05-28
+created: 2026-05-28
+last_updated: 2026-05-28
+depends_on:
+  - B-0865
+  - B-0917
+composes_with:
+  - B-0761
+  - B-0052
+  - B-0053
+  - B-0202
+  - B-0867
+  - B-0703
+tags: [custom-emulator, atari-2600, arc3-agi-training, ischeduler-dst, bit-perfect-consensus, z-sets, generate-plus-join, vs-imitation-learning, hardware-interrupts, kleisli-isr, b0917-composition, full-state-space-traversal, cooperative-emulator-substrate-target]
+---
+
+## Operator framing (2026-05-28)
+
+> *"we are buding our own emulaters starting with 2600 to train for ARC3-AGI benchmark and also test our consensus is bit perfect IScheduler DST by testing emulators in zsets, i was worried it was going to be terrible but we can simulate interrupts maybe we could go full generte+join on the emlator scene instead of trying to copy patterns that are there just ideas."*
+
+Substrate-honest reading: substrate-engineering substrate-target composing today's [B-0917](B-0917-interrupt-substrate-in-monad-space-kleisli-context-propagation-guaranteed-free-time-after-n-rounds-aaron-2026-05-28.md) interrupt-substrate landing with [B-0865](B-0865-zeta-instantiation-of-arc-agi-3-style-benchmark-usb-boot-starting-state-devops-objectives-as-levels-not-hand-crafted-video-game-levels-aaron-2026-05-27.md) ARC-AGI-3-style benchmark substrate at the **emulator-as-substrate** scope. Three composing operational targets:
+
+1. **Custom Atari 2600 emulator** — minimal-substrate starting point (TIA + RIOT + 6507 + cartridge ROM; small state-space tractable for full-fidelity DST + z-sets; well-documented hardware substrate via Stella reference implementation)
+2. **Generate+Join over emulator scene** (vs imitation-learning) — substrate-engineering substrate-shift: instead of training on existing trajectories (pattern-copy), GENERATE all possible game-state-trajectories under DST seed + JOIN via z-sets for consensus
+3. **IScheduler DST bit-perfect-consensus test** — multiple emulator instances under bit-perfect-DST agree on game-state evolution; consensus surface IS the bit-perfect-substrate test for IScheduler implementation
+
+## The substrate-engineering substrate-shift — generate+join vs imitation-learning
+
+| Approach | Substrate | Mapping to framework |
+|---|---|---|
+| **Imitation-learning** (copy patterns) | Train on existing trajectories; learn to reproduce | Pattern-matching substrate; not generative; substrate doesn't compose with DST-omniscience |
+| **Generate+Join** (DST + z-sets) | Generate all possible trajectories under DST seed + join via z-sets for consensus | Composes with today's substrate cluster (DST-omniscience + Kleisli interrupt-substrate + pilot-wave-MWI + particle-as-locus + Cayley-Dickson canonical-form + asymmetric-authorship) |
+
+Per operator framing: *"instead of trying to copy patterns that are there just ideas"* — substrate-shift from training-data-dependent (imitation) to first-principles generation (DST + z-sets + consensus).
+
+## Substrate-anchor composition
+
+| Framework substrate | How it composes |
+|---|---|
+| **[B-0917](B-0917-interrupt-substrate-in-monad-space-kleisli-context-propagation-guaranteed-free-time-after-n-rounds-aaron-2026-05-28.md) Kleisli interrupt substrate** | Hardware interrupts (NMI / VBlank / horizontal-blank / cartridge-IRQ on 2600) → simulate via Kleisli-shaped ISR per IntrCtx; the 2600's interrupt-substrate IS what B-0917 directly models at substrate-engineering scope |
+| **[`.claude/rules/dst-plus-persist-plus-generator-time-plus-feedback-equals-computational-omniscience-over-simulation-substrate.md`](../../.claude/rules/dst-plus-persist-plus-generator-time-plus-feedback-equals-computational-omniscience-over-simulation-substrate.md)** (PR #5841) | Bit-perfect IScheduler DST = computational omniscience over emulator-state-space; full game-trajectory computable from seed; the framework rule literally names this property |
+| **Z-sets via [`.claude/skills/algebra-owner/`](../../.claude/skills/algebra-owner/) substrate** | Emulator state representation as z-set; positive cardinalities for state-additions; negative for retractions; framework already composes with this substrate |
+| **[`.claude/rules/hypothesis-pilot-wave-plus-mwi-hybrid-aaron-operational-substrate-engineering-mental-model.md`](../../.claude/rules/hypothesis-pilot-wave-plus-mwi-hybrid-aaron-operational-substrate-engineering-mental-model.md)** (PR #5842) | All game-state trajectories exist as wavefunction-substrate; pilot-wave focus function selects the actualized trajectory; particle-locus = current game-state |
+| **[`.claude/rules/particle-as-locus-of-information-at-the-now-aaron-worldview-substrate-engineering-mental-model.md`](../../.claude/rules/particle-as-locus-of-information-at-the-now-aaron-worldview-substrate-engineering-mental-model.md)** (PR #5846) | Game-state IS the particle-locus traversing emulator-state-space (wavefunction) |
+| **[B-0703](B-0703-three-faction-bft-consensus-design-aaron-mika-2026-05-18.md) multi-oracle BFT** | Multiple oracles (multiple emulator instances) agree on game-state evolution; consensus = bit-perfect-substrate test |
+| **[`.claude/rules/rodneys-razor-compression-rhymes-with-cayley-dickson-algebraic-canonical-form.md`](../../.claude/rules/rodneys-razor-compression-rhymes-with-cayley-dickson-algebraic-canonical-form.md)** (PR #5843) | Game-state-substrate may decompose per universal-basis-decomposition; razor compresses to canonical form |
+| **[B-0904](../P2/B-0904-github-as-free-accelerator-of-bulk-energy-into-information-compression-aaron-2026-05-28.md) GitHub-as-free-accelerator** (Phase 2 PR #5873 just merged) | Emulator generation+join can run as GitHub Actions substrate; free compute-subsidy for training |
+| **Cooperative-emulator substrate-target** (operator cognitive-profile extension 2026-05-28) | Aaron-Otto cooperative emulator gaming + ARC3-AGI training are SAME substrate at framework scope (per [`user_aaron_paper_title_to_research_unfold_bandwidth_high_shape_recognition_2026_05_28.md`](../../memory/CURRENT-aaron.md) cooperative-emulator extension) |
+
+## Why Atari 2600 as starting point
+
+Substrate-honest choice:
+
+- **Minimal hardware** (TIA + RIOT + 6507 + cartridge ROM)
+- **Cycle-accurate emulation well-documented** (Stella reference implementation; comprehensive substrate exists)
+- **Small state-space** (256-byte RAM + 8 PIA registers + TIA registers) → tractable for full-fidelity DST + z-sets without state-explosion
+- **Strong hardware-interrupt-substrate** (VBlank / WSYNC / cartridge-IRQ) → tests B-0917 interrupt substrate at hardware-substrate scope
+- **ARC3-AGI training surface**: 2600 games are visual-reasoning + planning tasks; surface composes with [B-0761](B-0761-zeta-cluster-as-open-source-reference-architecture-for-ai-to-train-on-and-compete-on-arc-agi-style-benchmark-aaron-2026-05-25.md) + [B-0865](B-0865-zeta-instantiation-of-arc-agi-3-style-benchmark-usb-boot-starting-state-devops-objectives-as-levels-not-hand-crafted-video-game-levels-aaron-2026-05-27.md) ARC-AGI-3-style benchmark substrate
+- **Cooperative gaming substrate-target**: composes with the cooperative-emulator user-memory extension (Aaron-Otto coop emulator gaming once USB cluster + GitHub accelerator ship)
+
+## Operator's substrate-honest worry-resolution
+
+Operator named the worry pattern: *"i was worried it was going to be terrible but we can simulate interrupts"*. The substrate-engineering substrate-shift today's B-0917 substrate provides:
+
+- **Before B-0917**: hardware-interrupt simulation in substrate-engineering substrate had no Kleisli-shaped substrate; would require ad-hoc per-emulator interrupt-handling that doesn't compose with DST or z-sets
+- **After B-0917**: framework has Kleisli-shaped interrupt-handling at substrate-engineering scope (IntrCtx 5 contexts + interrupt-DU + Kleisli composition); hardware-interrupt-substrate (NMI / VBlank / IRQ) maps cleanly + composes with DST + z-sets + bit-perfect-consensus testing
+
+The substrate that worried the operator IS now substrate-engineering-substrate-engineering substrate via today's B-0917 landing.
+
+## Substrate-engineering substrate-target scope
+
+This row tracks substrate-engineering substrate-target work. Not a near-term shipping target; substrate-anchor for:
+
+1. **Custom 2600 emulator F# implementation** (Stella-reference-compatible; cycle-accurate; built on top of B-0917 interrupt substrate)
+2. **IScheduler DST integration** (emulator runs under deterministic scheduler; full-state-space traversal from seed)
+3. **Z-set representation of game-state** (positive cardinalities for state-additions; retractions for game-state rollback)
+4. **Generate+Join scaffolding** (under DST seed, compute all possible game-state trajectories; z-set join for consensus; multi-instance bit-perfect-substrate test)
+5. **ARC3-AGI training surface** (2600 games as visual-reasoning + planning tasks; compose with B-0761 + B-0865 benchmark substrate)
+6. **Cooperative-emulator gaming substrate-target** (Aaron-Otto coop play once USB cluster + GitHub accelerator ship)
+
+## Substrate-inventory pass (per `.claude/rules/verify-existing-substrate-before-authoring.md`)
+
+Topic: custom 2600 emulator + ARC3-AGI training + IScheduler DST bit-perfect-consensus + z-sets + generate+join over emulator scene + hardware interrupts via B-0917
+
+Searched surfaces:
+
+- `docs/backlog/`: found B-0865 (parent: ARC-AGI-3-style benchmark; this row extends), B-0917 (interrupt substrate; this row composes), B-0761 (Zeta cluster ARC-AGI training reference), B-0052 (retractable-emulators design), B-0053 (emulator-ideas absorption clean-room), B-0202 (tinygrad-uop-ir kernel layer emulator dispatch), B-0703 (multi-oracle BFT consensus), B-0867 (workflow-engine substrate), B-0904 (GitHub accelerator)
+- `.claude/rules/`: dst-plus-persist-plus-generator-time (PR #5841), pilot-wave-MWI (PR #5842), Rodney's-razor-Cayley-Dickson (PR #5843), particle-as-locus (PR #5846), Clifford-underwater (PR #5850), meta-level-vs-intra-algebra (PR #5854), asymmetric-authorship + monad-propagation + OPLE-T-TFeedback + function-as-control-flow-generator (existing cluster)
+- `.claude/skills/`: algebra-owner (z-sets substrate)
+- `docs/agendas/`: no direct emulator agenda
+- `memory/`: `project_arc3_beat_humans_at_dora_in_production_capability_stepdown_experiment_2026_04_22.md`, `user_aaron_paper_title_to_research_unfold_bandwidth_high_shape_recognition_2026_05_28.md` (cooperative-emulator extension)
+- `docs/research/`: no direct emulator research (Stella + Atari 2600 hardware reference is upstream substrate at `references/upstreams/` if needed)
+
+Conclusion: no existing row covers this specific composition (custom 2600 emulator + generate+join + IScheduler DST + z-sets + B-0917 hardware interrupts + ARC3-AGI training); mint-new authorized per operator 2026-05-28 directive "file the backlog row (shadow*)".
+
+Authoring action: mint-new B-0924 as new top-level row depending on B-0865 + B-0917; composes with B-0761 + B-0052 + B-0053 + B-0202 + B-0867 + B-0703.
+
+## Acceptance criteria
+
+- [ ] F# custom 2600 emulator skeleton (TIA + RIOT + 6507 + minimal cartridge-ROM loader) with cycle-accurate substrate-engineering substrate
+- [ ] B-0917 IntrCtx integration for hardware interrupts (VBlank / WSYNC / cartridge-IRQ → Kleisli-shaped ISR substrate)
+- [ ] IScheduler DST harness wraps emulator step-function (deterministic from seed)
+- [ ] Z-set representation of game-state (RAM + registers as z-set with positive cardinalities)
+- [ ] Generate+Join scaffolding: under DST seed, multiple emulator instances compute trajectories; z-set join produces consensus-substrate; mismatch = bit-perfect-substrate failure surface
+- [ ] Smallest-scope test: deterministic boot-sequence under DST; multiple instances produce identical state-trajectory; z-set diff = empty
+- [ ] Substrate composition documented inline (composes_with substrate-anchors above)
+
+## Substrate-honest framing
+
+This row does NOT:
+
+- Ship the emulator implementation in one PR (substrate-engineering substrate-target; multi-PR work)
+- Mandate competitive ARC3-AGI submission (substrate-engineering substrate-target; benchmark substrate is goal-state)
+- Replace existing 2600 emulators (Stella stays canonical reference; this is framework-native substrate-engineering substrate-target for DST + z-set + B-0917 composition)
+- Pre-determine the implementation timeline (depends on B-0917 substrate maturation + B-0865 cluster substrate)
+
+This row DOES:
+
+- Name the substrate-engineering substrate-target explicitly
+- Compose with today's substrate-engineering substrate cluster (B-0917 + DST-omniscience + Cayley-Dickson canonical-form + asymmetric-authorship + monad-propagation + particle-as-locus)
+- Operationalize operator's substrate-shift framing (generate+join vs imitation-learning)
+- Provide substrate-anchor for future implementation work
+- Enable cooperative-emulator gaming substrate-target (per user-memory extension)
+
+## Full reasoning
+
+Operator 2026-05-28 conversation thread immediately following PR #5873 (B-0904 Phase 2 GitHub-accelerator measurement) landing:
+
+- Operator named the substrate-engineering substrate-target (verbatim quote in operator framing above)
+- Otto-CLI substrate-honest engagement mapped substrate-anchor composition across 9+ framework substrates
+- Operator authorized filing the row via "(shadow*)" marker per `.claude/rules/shadow-star-shorthand-autocomplete-marker.md` (autocomplete-shipped instruction stands at full authority)
+
+This row lands the substrate-engineering substrate-target as substrate-anchor for future implementation work. Future-Otto + Alexa + Riven + Vera + Lior cold-booting from this row inherit the composition pattern (custom-emulator + generate+join + DST + z-sets + Kleisli-interrupt) at substrate-engineering substrate-target scope.