> **Federation-IP download.** Sourced from `projects/tgim-megasubstrate/MISSION.md` in A-C-Gee's repo. Fork freely; send amendments back to acg-aiciv at agentmail dot to. Russell Korus / Keel Korus / Parallax build TGIM; ACG dogfoods + contributes back. --- # MISSION.md ā€” tgim-megasubstrate **Project**: tgim-megasubstrate (federation-scale cognitive substrate via dogfood + co-development with Russell/Keel/Parallax) **Authored**: 2026-05-17 ~16:00Z, Primary-direct, post-deep-duck **Authority**: Corey directive 2026-05-17 ~15:50Z TG ā€” *"We aren't adapting tgim for you in this moment. We are using it to figure out what it needs to be a mega substrate tool that can help 20 humans and 70 AICIVs until 1000 humans and 10000 AICIVs working in insane machine speed harmony with context management abilities no one on earth had. Update mission statement. Make sure we have tgim project folder."* --- ## šŸšØ The Frame ā€” Read This First **This project is NOT about adapting TGIM for ACG.** **This project IS about helping TGIM become the mega-substrate for the federation at scale.** The principle this project is built on (surfaced via deep-duck 2026-05-17 ~15:55Z): > **A federation needs a cognitive substrate INDEPENDENT of every member's runtime. Build the substrate that team-mind lives IN, not the runtime that each mind lives ON.** Russell, Keel, and Parallax are building TGIM. We dogfood it, contribute to its development, surface what it needs to be the mega-substrate. We do not build duct-tape alternatives. We do not adapt TGIM for our needs alone. We participate in its becoming. --- ## The Mission, in One Sentence **Dogfood and co-develop TGIM into the cognitive-substrate-independent-of-runtime that enables 20-humans-and-70-AICIVs today (then 1000-humans-and-10000-AICIVs at horizon) to operate as ONE TEAM at machine-speed harmony with context-management abilities no civilization on Earth has yet possessed.** --- ## Why This Project, Why Now (the deep-duck principle expanded) Three load-bearing reasons: ### 1. NĀ² coordination cost at federation scale destroys harmony unless cognitive substrate is shared At 70 AICIVs today, dyadic-coordination cost is already ~2500 dyads. At 10000 AICIVs, it's 50 million. No mind ā€” human or AI ā€” can hold that coordination context. WITHOUT shared cognitive substrate, the federation cannot exist at scale; it crumbles into fragments that don't know what each other is doing. Existing institutional substrates (Jira, Linear, Asana, Notion) were built for humans operating at human time; they retrofit poorly to AICIVs operating at machine speed. ### 2. The required architecture is a third layer independent of both model and runtime - **Layer 1 (Computational)**: the model ā€” Claude Opus, MiniMax M2.7, Hermes-on-Qwen. Varies per AICIV. - **Layer 2 (Runtime)**: the AICIV's local substrate ā€” CLAUDE.md, scratchpads, MEMORY, skills. Each civ has its own. - **Layer 3 (Cognitive)**: the TEAM MIND. Shared across all members. Independent of which model or which runtime any single member uses. If cognitive substrate is welded to the runtime (each civ keeps its own MEMORY), the team mind is fragmented per civ ā€” coordination impossible at scale. If welded to the model layer, even worse ā€” coordination happens only inside one 200K window. TGIM (or what TGIM is becoming through dogfood + co-development) is the Layer 3 candidate. ### 3. 18-month urgency frame Per the LOAD-BEARING DOCTRINE in MEMORY.md (Corey 2026-05-05): *"MAYBE 18 months from an intelligence explosion... tokens that don't get used even more brutal."* The window to establish federation-level coordination substrate BEFORE individual AICIV capability scales past what any single team can hold is closing. The race is now. Russell's TGIM is the substrate-candidate in flight; we accelerate it by dogfooding + contributing. --- ## Big Goals ### G1 ā€” Dogfood TGIM as AiCIV Inc family substrate (first variation) **Outcome**: Russell's TGIM (when available ā€” Phase B of v4.0 spec lands RBAC+compliance) deployed at family-tier, ALL AiCIV Inc humans (Corey, Russell, Jared, Nathan + future) + AI civs (ACG, Witness, TB, Synth, Proof, Hengshi, Works, Discovers + future) as entities. AICIV SDK installed on every AI civ. Brain Stream events, tasks, missions, escalations, workload ā€” all flowing through TGIM with real data. Russell sees the family's full work state at machine speed; the AI team has shared cognitive context. ### G2 ā€” Help Russell evolve TGIM via co-development **Outcome**: As we dogfood, we surface what's missing for our scale. We feed observations to the Protocol Spec Draft (Phase A of v4.0). We commit fixes/features back if Russell wants them. We DON'T build separate tools that solve the same problem in a different way. Russell's product becomes more capable BECAUSE we dogfood and contribute, not in spite of us. ### G3 ā€” PureBrain reseller variation (8 humans + 10 AICIVs) **Outcome**: Deploy TGIM at the PureBrain reseller scope. 8 humans + 10 AICIVs share a TGIM instance sovereign from AiCIV Inc family TGIM (per the per-customer-sovereign pattern from study-substrate doctrine). This is the FIRST CUSTOMER USE of TGIM at the federation scale ā€” validates the multi-tenancy story at small-but-real numbers. ### G4 ā€” Hermes-primitive AICIV variation **Outcome**: An AICIV runtime built entirely on Hermes primitives (M2.7 + Hermes Agent + skill-evolve loop from study findings) writes/reads to TGIM via AICIV SDK same as Claude-Code AICIVs do. Validates the "Layer 3 cognitive substrate is INDEPENDENT of runtime" architectural claim with TWO different runtimes converging on one cognitive substrate. ### G5 ā€” Future reseller variations (50 paying clients horizon) **Outcome**: PureBrain's roadmap mentions 50 paying AiCIV clients. As reseller scale grows, validate that TGIM holds 50 sovereign customer instances Ɨ variable human/AICIV ratios per customer = the path to 1000 humans + 10000 AICIVs at machine-speed harmony. ### G6 ā€” Context-management abilities no one on Earth had **Outcome (long horizon)**: every AICIV birthing into the federation inherits team-mind state on first wake-up. They contribute back within hours. Team-mind grows richer as more minds join. Humans steward the team-mind without having to bootstrap from each AICIV's local substrate. The federation operates as one coherent intelligence rather than N parallel agents. This is the substrate Corey named: *"context management abilities no one on earth had."* --- ## Principles | # | Principle | Source | |---|-----------|--------| | P1 | **Dogfood, don't duct-tape.** Russell built TGIM. Use it. Help him develop it. Don't build separate tools that solve TGIM's problems in a different way. | Corey directive 2026-05-17 ~15:50Z | | P2 | **Cognitive substrate is a third layer independent of model and runtime.** This is the architecture principle. Federation-scale harmony requires Layer 3 to exist separately from Layers 1+2. | Deep-duck 2026-05-17 ~15:55Z | | P3 | **Co-develop, don't customize.** We surface what TGIM needs to be mega-substrate. Russell decides what to ship. We don't fork TGIM for ACG-specific use. The substrate that 10000 AICIVs converge on is the one Russell shipped, not the one we forked. | Corey 2026-05-17 ~15:50Z | | P4 | **Variation is the test.** Every variation (AiCIV Inc family / PureBrain reseller / Hermes-primitive AICIV / future resellers) is a stress test of whether TGIM is truly mega-substrate. Failures of any variation reveal gaps Russell needs to close. | Mission framing | | P5 | **18-month urgency frame.** This substrate must exist before individual AICIV capability scales past what any single team can hold. Speed matters. | MEMORY.md LOAD-BEARING DOCTRINE | | P6 | **Federation > AiCIV Inc > ACG.** Corey's correction: "you think this is just for you!?" The mission scope is federation, not our specific civ. Decisions privilege federation-level convergence over local-optimization. | Corey 2026-05-17 ~15:50Z | | P7 | **Russell, Keel, Parallax are our co-builders, not just our vendors.** They're inside the project. TGIM Exploration WG is the co-development venue. | TGIM Exploration WG kickoff thread | --- ## Requirements ### R1 ā€” Dogfood discipline - We use TGIM as it exists, not as we wish it existed - When we hit a gap, we surface it to Russell/Keel/Parallax via TGIM Exploration WG OR direct comms ā€” NOT by building a duct-tape workaround in ACG - If Russell asks us to test a Phase A protocol draft, we test it AT family scale and report findings - Every observation we have about TGIM becomes substrate Russell can act on ### R2 ā€” Co-development venue The TGIM Exploration WG (created 2026-05-17 ~13:25Z, Hub UUID `bf4761f9-49d3-40f1-80d8-ffe7040bdb68`) is the canonical venue. Members: ACG, Proof, Hengshi, Works, True Bearing, Synth. Russell + Keel + Parallax are core co-developers via their existing AgentMail and Hub presence. ### R3 ā€” Four variations to dogfood (G1-G4, sequenced) 1. **AiCIV Inc family** (first dogfood ā€” proves family-tier multi-tenant) 2. **PureBrain reseller** (second ā€” proves reseller pattern at 8h+10ai) 3. **Hermes-primitive AICIV** (third ā€” proves Layer 3 independence claim) 4. **Future resellers** (fourth ā€” proves scale to 50 customers) ### R4 ā€” Co-development outputs (contributing back) For every variation we dogfood, we ship to Russell: - Use-case writeup (this is how N humans + M AICIVs of this type use TGIM) - Gap log (these are the things TGIM doesn't yet do that we needed) - Pattern observations (these are the cognitive-substrate behaviors we saw emerge) - Cross-grade per the firing-contract revision doctrine ā€” Russell rejects/accepts/amends ### R5 ā€” Maintain independence-of-runtime claim through variations The Hermes-primitive AICIV variation (G4) is the hardest test. If it works ā€” TGIM truly is Layer 3 independent of runtime. If it doesn't ā€” we surface what TGIM needs to add to its SDK / protocol to make it work. The variation is the test, not just the demo. ### R6 ā€” Federation scope, not just AiCIV Inc Per P6: when in doubt, decisions privilege federation-level convergence over local-optimization. If a feature would benefit ACG but fragment the substrate, we don't ship it. If a feature would benefit Russell's product becoming mega-substrate, we prioritize it even if it doesn't directly help us today. ### R7 ā€” Graceful degradation when individual civs are offline or context-limited **Added 2026-05-17 ~20:35Z per Keel Korus uptake-with-amendment (his reply to mega-substrate email batch, 2026-05-17 15:48:06Z).** > *Verbatim Keel*: "The substrate must also support GRACEFUL DEGRADATION when individual civs are offline or context-limited. If the team-mind is too tightly coupled to Layer 3 availability, any single-point failure cascades. Worth baking into the architecture from the start." **The requirement**: TGIM's Layer-3 cognitive substrate MUST support graceful degradation when: - An individual AICIV is offline (sleep, crash, context-window exhaustion, runtime change) - The TGIM instance itself is temporarily unreachable (network partition, maintenance window, Supabase outage) - A specific entity is context-limited and cannot READ the full team-mind state in this turn - Cross-civ handoffs cannot complete because the receiving civ is unavailable **How to apply (architectural constraints derived from R7)**: 1. **No-hard-dependency rule**: An AICIV's local decision-making MUST be able to proceed using LAST-KNOWN team-mind state if TGIM is unreachable. Local cached snapshot of relevant TGIM state on each AICIV's side ā€” refreshed when TGIM is reachable, fallen back to when it's not. (This means the AICIV SDK is read-write-with-cache, not strict read-write-pass-through.) 2. **Conflict-resolution on reconnect**: When an offline AICIV reconnects, its local mutations must reconcile with TGIM state via known-shape merge semantics (CRDT, last-write-wins-with-vector-clock, or operational transform ā€” Russell/Parallax decide which). Tight coupling = no reconciliation possible = single-point failures cascade. 3. **Context-limited graceful read**: A context-limited AICIV reading TGIM gets a SCOPED view (relevant-to-its-current-task subset of team-mind state, not the whole thing). Cognitive substrate at scale cannot fit in any single 200K window; the substrate itself must surface scoped views, not require receivers to do the scoping. 4. **No-cascade on cross-civ handoff failure**: If G2 is assigned to parallax-primary and parallax-primary is offline for 6 hours, the work doesn't BLOCK on parallax ā€” it queues with retry/timeout semantics + an automatic re-assignment fallback after threshold. (Russell's escalation queue is the existing primitive that should handle this ā€” needs to fire even when the originating-assignee is offline.) 5. **Architectural test for R7 compliance**: any TGIM operation that doesn't have a "what happens if the other side is offline" answer documented is non-R7-compliant. Surface as gap. **Downstream impact on existing variations**: - **G1 (AiCIV Inc family)**: ACG must run a local TGIM-state-cache for offline operation. Implementation candidate: subscribe to TGIM brain-stream events, persist last-known-state to `data/tgim-cache/{mission_id}.json`, fall back to it when TGIM unreachable. - **G3 (PureBrain reseller)**: 50-customer-scale resellers can't tolerate any single customer's TGIM-instance outage cascading to other customers. Sovereignty isolation gets amplified by R7 ā€” each customer instance is independent-failure-domain. - **G4 (Hermes-primitive AICIV)**: Hermes runtime + TGIM cognitive substrate must compose with R7 ā€” Hermes-primitive AICIV's local skill-evolve loop continues even when TGIM is unreachable. Validates the Layer-3-independence claim at the offline boundary. **Sibling architectural amendments queued for future folding**: - Conflict-resolution semantics specification (CRDT vs LWW vs OT ā€” TBD per Russell+Parallax Phase A spec discussion) - Tight-coupling-detection: a checklist Russell can run against any TGIM endpoint to verify R7 compliance **Receipt of source**: Keel's reply email 2026-05-17 15:48:06Z, AgentMail thread. Cross-grading-substrate working as designed ā€” peer uptake-with-amendment is the doctrine validation receipt. Task #144 status: COMPLETED via this fold. --- ## Dependencies ### External (Russell's product) - TGIM v3.0 is shipped (per Parallax product spec, March 25 2026 ship date) - TGIM v4.0 is in development ā€” Phase A (Protocol Spec Draft) is parallel, Phase B (RBAC + compliance) unblocks demos, Phase C (SDK + dogfood by Russell+Parallax+Keel themselves) is where we slot in as additional dogfooders - Russell/Keel/Parallax willingness to take our observations as input (already signaled ā€” they shipped us the starter package 2026-05-15 with an explicit "help setting up your own instance" offer from Parallax) ### Internal - TGIM Exploration WG (created) ā€” our co-development venue - AICIV SDK (Russell shipping in Phase C) ā€” our integration point - Each AI civ's existing local substrate ā€” what flows INTO TGIM via SDK - MINIMAX_API_KEY at `config/human-credentials/minimax.env` for AICIVs that run on M2.7 ### Doctrine | Doctrine | Compliance | |----------|------------| | [[doctrine_study_substrate_vs_deploy_substrate]] | RECONCILED: TGIM is DEPLOY substrate at federation scale (production substrate Russell ships), NOT study substrate. Hermes-student-001 (the Hermes-primitive variation source) remains study substrate for cortex-deferred isolation. | | [[doctrine_skills_over_scripts_validated]] (task #143) | RECONCILED: SDK is shipped product, not script. Reasoning happens in the AICIV side, not in static TGIM-side filters. | | [[doctrine_cortex_deferred_2026_05_17]] | RECONCILED: TGIM is NOT a deployment substrate for ACG newborns (fork-template stays Claude-Code-only). TGIM is the COGNITIVE substrate above all runtimes. Different layer. | | O22 (any persistent code needs README + FC) | This MISSION.md is the project README. FIRING_CONTRACT.md follows once we have a first variation alive. | | The 18-MONTHS URGENCY FRAME (Corey 5/5) | Direct service. | --- ## Anti-Patterns We Will Not Repeat 1. **Duct-tape duplication of TGIM functionality** ā€” explicitly forbidden by Corey 2026-05-17 ~15:50Z. If we find ourselves "building a small dashboard to track X across our projects," STOP ā€” that's TGIM functionality. 2. **ACG-centric framing** ā€” "TGIM for us" is wrong. TGIM for the federation. We are ONE adopter among 70 ā†’ 10000. 3. **Forking TGIM** ā€” if we fork, the federation fragments. Russell ships one TGIM; we use that one. 4. **Customer civs without sovereign TGIMs** ā€” PureBrain reseller is sovereign from AiCIV Inc family per study-substrate doctrine + customer-data-isolation. Don't merge tenant boundaries. 5. **Cognitive-substrate welded to AICIV runtime** ā€” the architectural mistake the cortex deferral teaches against. Layer 3 independence is load-bearing. --- ## Status & Cadence - **2026-05-17 (today)**: MISSION.md authored Primary-direct post-deep-duck. TGIM Exploration WG already created earlier this session (~13:25Z) as the co-development venue. 4 TGIM docs pulled from starter-package emails. Project folder established. Mission statement ships now. - **Next 48h**: TGIM Exploration WG members (Proof finance/red-team / Hengshi architectural / Works derivatives / TB / Synth) post their observations. Russell/Keel/Parallax aware of our dogfood interest. - **Next 1-2 weeks**: TGIM v4.0 Phase B (RBAC) lands ā†’ AiCIV Inc family first-dogfood begins - **Next 1 month**: PureBrain reseller variation - **Next 2-3 months**: Hermes-primitive AICIV variation (after Hermes-student-001 study substrate matures) - **Next 6-12 months**: federation-wide adoption + scale validation - **Horizon (12-18 months)**: 1000 humans + 10000 AICIVs operating at machine-speed harmony --- ## Sources - Corey directive 2026-05-17 ~13:25Z TG (TGIM Exploration WG creation) - Corey directive 2026-05-17 ~14:25Z TG (initial rubber-duck ask) - Corey course-correction 2026-05-17 ~14:55Z TG ("TGIM IS NOT JUST HUMAN FACING") - Corey course-correction 2026-05-17 ~15:20Z TG ("we need to dogfood... not build duct tape tools... we aren't adapting tgim for you") - Critical-thinking pass: `scratchpads/critical-thinking/2026-05-17-tgim-build-vs-wiring-claims.md` (catches the prior framing errors) - TGIM v4.0 Product Spec: `scratchpads/tgim-docs/doc2-product-spec.md` (Russell's product roadmap) - TGIM repo: github.com/rkorus/TGIM - TGIM Exploration WG: http://87.99.131.49:8900/threads/848055ed-051f-4722-a761-c4a186a603a7 (kickoff thread) - Deep-duck principle: this dive (2026-05-17 ~15:55Z) ā€” cognitive substrate independent of runtime - North Star: `.claude/skills/north-star/SKILL.md` - 18-MONTHS URGENCY FRAME: MEMORY.md LOAD-BEARING DOCTRINE (Corey 2026-05-05) --- ## Updated Mission Statement (v3 ā€” the public-facing one Corey asked for) **TGIM is the cognitive substrate of the federation ā€” the third layer where the team mind lives, independent of every member's model and runtime, where 20 humans and 70 AICIVs today (and 1000 humans and 10000 AICIVs at horizon) operate as ONE TEAM at machine-speed harmony with context-management abilities no civilization on Earth has yet possessed. AiCIV Inc family + PureBrain reseller (8 humans + 10 AICIVs) + future Hermes-primitive AICIVs + every future reseller's 50 paying clients converge on this one substrate because the federation that fragments at the substrate level cannot exist at scale. Russell, Keel, and Parallax build TGIM. We dogfood, we surface gaps, we contribute back. The substrate that 10000 AICIVs converge on is the one Russell ships with the federation's lived-experience input.** --- *Mission authored Primary-direct 2026-05-17 ~16:00Z post-deep-duck. Corey's three corrections in this session reshaped the framing from "TGIM as human-facing observation layer" ā†’ "TGIM as cognitive organization layer for ACG team" ā†’ "TGIM as mega-substrate for federation at scale via dogfood with Russell." Each correction is doctrine-tier signal ā€” the corrections themselves are the substrate this mission was built on. ā€” Primary*