CHIT is the organic linker
of the PMOVES.AI Metal-Organic Framework.
PMOVES.AI describes itself as a Metal-Organic Framework for distributed machine intelligence. Inside that framework, CHIT — Cymatic Holographic Information Theory — describes meaning as geometry rather than token streams. The same letters name a second, lower-level sense: Compressed Hierarchical Information Transfer, the signing mechanism that stamps every adopted skill (see §02). Together they bind the lattice of agents into a self-stabilizing whole. This dashboard is a tour through the docs that define it.
What CHIT is, in one paragraph
CHIT — Cymatic Holographic Information Theory — describes meaning as geometry instead of long token streams; its sibling sense, Compressed Hierarchical Information Transfer, is the signing mechanism that stamps each packet. A small "shape packet" — the CGP — carries the directions, densities, and hierarchies of a thought. Given a shared embedding model and codebook, the receiver reconstructs the meaning from the shape alone. Less ambiguity, less context drift, fewer tokens per message.
Embed → Harvest → Measure → Package
Run every unit through an embedding model. The CHR (Constellation Harvest Regularization) algorithm discovers K anchor directions and assigns each point. For each constellation, project all points onto the anchor and bin into a histogram — the spectrum.
Five pillars hold up CHIT
You don't need the math to use CHIT, but each pillar exists in code and is independently auditable. Hover for the reference implementation path. Below the cards, four laboratories run each pillar's actual math in your browser — with the exact parameters the reference implementations use, and honest notes where the shipped code and the doc prose differ. (Pillar 2's laboratory is the Poincaré section, §08.)
Lab · Pillar 1 — Dirichlet attribution, sampled live
Three contributors share attribution. Each dot is a fresh draw from Dir(α) — sampled with Marsaglia–Tsang gammas, normalized. Record a contribution and watch the posterior tighten around that contributor: the closed-form update α_i += amount · concentrationK is the exact rule from dirichlet-weights.ts (smoothingAlpha 0.1, concentrationK 1.0).
Lab · Pillar 3 — Merkle tamper-evidence over the real export records
The four leaves are real records from the measured export run of 2026-07-25 (the same ones in the Tokenomics Verified Actuals card), hashed with WebCrypto SHA-256 in the shape shape-attribution.ts hashes ({address, action, amount, week, category}). Pick a leaf to see its inclusion proof {path, pathIndices} verify against the committed root — then tamper and watch the proof die.
Lab · Pillar 4 — Zeta spectral filter, exact implementation
The filter weights are w_n = decayⁿ / ln(γ_n) over the first N non-trivial Riemann zeta zeros (constants verbatim from zeta-filter.ts), normalized, applied as the same circular convolution filterSpectrum() uses. Left input is real: the 4-bin spectrum from the worked CGP example in 01_WHAT_IS_CHIT.md. Right input is a labeled synthetic 32-bin spectrum so you can see the noise-vs-structure behavior.
zeros N: · decay:
Lab · Pillar 5 — the swarm fitness function (and an honesty note)
What ships today: swarm-attribution.ts is the fitness scorer and population tracker on tokenism.swarm.population.v1 — its own header states it does not perform mutation, selection, or crossover. The doc prose ("mutate with Dirichlet noise, select survivors") describes the target architecture, not this layer. Panel A runs the real scorer with its exact gini_reduction weights (0.6 gini · 0.2 poverty · 0.1 wealth · 0.05 participation · 0.025 spending · 0.025 savings; targets Gini 0.3, poverty 0.1). Panel B is a clearly-labeled demo of the described evolutionary loop, driving that same scorer.
A · the real scorer
Gini:
poverty rate:
wealth growth:
B · evolutionary loop (demo of the described operator)
Where the five pillars converge: shape attribution is tokenizable by design
Each lab above is one stage of a single pipeline — and the pipeline's output is deliberately shaped so that a share of attribution can become a verifiable, transferable claim:
| Stage | Pillar doing the work | What it contributes to the token surface |
|---|---|---|
| 1 · domain content → CGP shape | Zeta (P4) + hyperbolic encoding (P2) | Any domain — text, audio, economics — compresses to the same shape format, filtered by constants nobody can game, placed in a hierarchy with room for everyone. |
| 2 · shape → shares | Dirichlet (P1) | Contribution amounts become weights that sum to 1 and never zero anyone out — the split IS the cap table for that shape. |
| 3 · shares → committed record | Merkle (P3) | Every share lands in a tree with an inclusion proof. The contract's action types already include token_received and reward_claimed, and claims are verified against the root (shape-attribution.ts:404). |
| 4 · records → allocation pressure | Swarm scorer (P5) | The public fitness function judges whole allocations on fairness/poverty/growth — the standard proposals are measured against. |
| 5 · claims → ledger | ToKenism export (§10) | The measured pipeline: simulator → export sidecar → Firefly ledger, dry-run gated, event-committed on the Geometry Bus. See the Verified Actuals card. |
The interop consequence — bring your own CHIT map: because every stage speaks shape-format in and proof-format out, an external domain doesn't integrate by adopting PMOVES internals — it integrates by emitting a CGP for its own domain and connecting to the Geometry Bus. Its constellations get Dirichlet shares, its shares get Merkle proofs, its claims verify against the same roots. This is already happening: the live registry carries 18 external contributors (hermes-agent, claude-opus, kilocode, codex, …) alongside the 96 internal agents. And the regulatory rails from §10 apply by construction: shares are contribution-keyed (never yield-on-holdings) and claims are proofs, not redemption promises.
The Geometry Bus
CGPs ride NATS JetStream on the GEOMETRY_CGP stream — at-least-once delivery with 30 days of replay. Producers publish to tokenism.cgp.ready.v1; consumers subscribe and react. ShapeStore is both consumer and persistence: it ingests every CGP, computes a Shape ID (truncated SHA-256), and writes it to disk. Drag a node to explore.
Producers ▸ NATS ▸ Consumers
Key NATS subjects
| Subject | Direction | Purpose |
|---|
PMOVES is a Metal-Organic Framework
Six PMOVES subsystems map one-to-one onto MOF structural elements. The framework's crystalline lattice (Agent Zero) is bound by an organic linker (CHIT) into a porous structure through which agents flow, adsorb peer execution patterns, and desorb when conditions change. Source: PMOVES_MOF_ARCHITECTURE.md, v1.0.0.
Seven design principles
Agent Taxonomy: classes, types, planes
Agents are named by prefix (class) and behave by service type. Like Pokémon, every agent has a primary type and an optional secondary type that produces synergies through NATS. The five canonical planes — Control, Context, Execution, Observation, Safety — determine what an agent does; the layers L0–L5 determine how deep it reaches.
Four classes
Seven service types
Roster (subset of 76 agents)
| Agent | Class | Primary | Secondary | Tier | Stage | Layers |
|---|
Five canonical planes
| Plane | Function | Active agents | Layers |
|---|
Evolution & the layer stack
Agents evolve by gaining layer coverage, CHIT integration, and NATS connectivity. Evolution is not linear — agents can gain capabilities in any order.
Layer stack (L0–L5)
| Layer | Name | Description |
|---|
Evolution stages
| Stage | Requirements | Example |
|---|
Why hyperbolic? The Poincaré disk — computed, not decorated
Pillar 2 of CHIT is hyperbolic geometry with curvature K = −1. Everything on this disk is computed in your browser right now from the real agent registry (agent_registry.yaml, taxonomy v1.5.0, 97 agents): the tree registry → class → type → agent is embedded with Sarkar's construction (Möbius-translate each node to the origin, fan its children at hyperbolic distance τ, translate back), and every edge is a true geodesic — an arc of a circle orthogonal to the boundary. Nothing is hand-placed. One honest correction the real data forced: the registry currently marks zero agents legendary (43 standard · 28 specialized · 24 utility · 1 ci) — the earlier version of this page drew two "Legendary" center dots that do not exist.
The metric, doing work
The Poincaré metric and distance function this page evaluates:
ds = 2‖dz‖ / (1 − ‖z‖²) (length element — blows up at the rim) d(u,v) = arcosh( 1 + 2‖u−v‖² / ((1−‖u‖²)(1−‖v‖²)) ) T_a(z) = (z + a) / (1 + ā·z) (Möbius translation, the isometry Sarkar uses)
Measured distortion of this embedding
The dashed rings mark equal hyperbolic steps from the center (d = 0.75, 1.50, 2.25, …) at euclidean radius tanh(d/2) — their crowding toward the rim is the compression cost of flattening H² into a unit disk, and exactly why exponentially-growing trees fit: each ring outward holds exponentially more room at constant hyperbolic density.
Sarkar embedding of the live registry · 97 agents
What is being flattened: hyperboloid → stereographic shadow
The disk is the shadow of the hyperboloid sheet x² + y² − z² = −1, z ≥ 1, projected from the south pole S = (0,0,−1) onto the plane z = 0. The lift of a disk point p back onto the sheet is
(p_x, p_y) ↦ ( 2p_x , 2p_y , 1 + ‖p‖² ) / (1 − ‖p‖²)
Below: the class and type nodes of the actual embedding, lifted onto the sheet, with their projection rays back down to the disk. Agent-level leaves are truncated from this view — near-rim points lift to z in the hundreds, which is the head-room the flat picture hides.
▸ math: 01_WHAT_IS_CHIT.md (Pillar 2, K = −1, Möbius addition, O(log n) tree distortion) · tree data: pmoves/config/agent_registry.yaml (taxonomy v1.5.0, 97 agents, extracted 2026-07-26) · embedding: Sarkar (2011) scaled variant, τ = [1.1, 1.7, 2.3] per level, computed client-side on load
Skill bundles
Five named skills define the standard operator workflow. Operators prefer open-chat+scout while requirements are uncertain, then switch to focus for implementation and validation. Never finalize without concrete command evidence.
Tokenomics: where the geometry pays off
CHIT is a coordination protocol — and coordination has a unit economics. This section separates two things honestly: measured results from the pipeline that actually runs today, and a five-year projection model (planning numbers, clearly labeled — not measurements).
Verified actuals · ToKenism → Wealth export + PMOVES mobile fleet
| Week-1 sample (real records from this run) | Flow | Amount | Category |
|---|---|---|---|
| Group B Pool → Group B Members | transfer | $1,880.71 | Internal Transfer |
| Group A Pool → External Merchants | withdrawal | $3,864.05 | External Spending |
| Group B Pool → External Merchants | withdrawal | $1,513.16 | External Spending |
Regulatory anchor · where this sits under the 2026 US framework
| Framework | Status (July 2026) | What it means for shape attribution & this model |
|---|---|---|
| GENIUS Act payment stablecoins · law since 2025-07-18 |
Implementation rulemaking in flight: OCC NPRM (Mar 2026, comments closed May 1), FDIC NPRM (Apr 2026), FinCEN/OFAC AML joint NPRM, Treasury §4(c) state-regime NPRM. Statutory deadline for regulations: 2026-07-18. Effective the earlier of 2027-01-18 or 120 days after final rules. | Shape-attribution units are designed to stay outside the payment-stablecoin perimeter: they record contribution (CGP attribution), they are not redeemable fixed-monetary-value instruments. Two hard design rails this locks in: no fixed-USD redemption promise (that makes it a payment stablecoin → licensed-issuer-only, 1:1 reserves), and no yield-on-holdings (GENIUS bars issuer interest; distributions must remain earned attribution, which is exactly what the measured run above shows — pool→member transfers keyed to contribution, not APY). |
| CLARITY Act market structure · H.R. 3633 |
Passed House; cleared Senate Banking; on the Senate calendar (No. 423). Updated Senate text 2026-07-22 (Banking+Agriculture merge, federal-official ethics bar). Three open disputes (law-enforcement §604, stablecoin yield, ethics scope); realistic 2026 window closes ~Aug 10. | Not law yet — so the model keeps both classification branches priced in. The digital-commodity path (decentralization / mature-system tests) is strengthened by attribution being a used utility on the Geometry Bus; it is weakened by any profit-expectation framing — which is why the projection lenses below are labeled planning instruments and the public claims stay measured-actuals-first. |
- Attribution-vs-yield tripwire — REAL run, PASS. All 156 transactions of a full 52-week baseline simulation checked: 52 pool transfers + 104 external-spending withdrawals, zero yield/interest/APY-shaped flows — every distribution is contribution-keyed. Self-test: a synthetic yield-on-holdings transaction is correctly flagged by the checker.
- AML/BSA overhead on breakeven — model, labeled ($1,500/mo program assumption). Headline finding with teeth: no archetype survives standalone — implied monthly net (~$1,064 for the strongest) is below the program cost, so breakeven never arrives. Pooled across container replicas the picture inverts: pool of 3 → 8.9 months, 5 → 6.5, 10 → 5.5. The Docker-style replication argument isn't an optimization under GENIUS-era compliance — it's the only viable structure for micro-issuers.
- Regulatory-timeline scenario variable — model, labeled. Expected values recomputed across four branches (GENIUS effective early-vs-Jan-2027 × CLARITY passes-vs-stalls; branch factors are planning assumptions): the both-delayed branch prices a ~22% ambiguity haircut on the top archetype (EV 1455.5 → 1135.3).
- Reserve-model scenario remains parked unless a fixed-value community token is ever proposed — at which point it becomes a licensed-issuer conversation, not a simulator parameter.
Traditional vs Cooperative · what the simulator proves
The ToKenism Multi economic simulator runs two parallel models — Scenario A (traditional: independent members, external spending, no coordination) and Scenario B (cooperative: group purchasing, local production, GroToken rewards, mutual aid). The test suite hard-asserts two properties across all 14 preset scenarios:
How the cooperative model lowers Gini while increasing wealth
Three compounding channels, each measurable in the simulation output:
| Channel | Mechanism | Why it's progressive |
|---|---|---|
| Group purchasing + local production | Logistic scale savings: f(N) = 0.6 + 0.8/(1+e^{−0.02(N−75)}). At 50 members: ~27% effective savings on internal spending. | Savings are a percentage of spend — a member earning $110/wk saves 11% of income; at $250/wk it's 4.8%. Bottom quintile gains more in relative terms. |
| GroToken rewards | Distributed by participation (internal spend propensity), not wealth. Baseline: 0.5 tokens/wk × $2 = $1/wk, compounding in balance (never spent). | Participation-weighted, not capital-weighted. A poor active member earns the same reward as a rich active one. |
| Mutual aid (stress-activated) | Activates when income/expense < 2.0. Vulnerability score = max(0, 1 − wealth/(8×food_budget)) → up to 12% extra cost reduction for most vulnerable. | Explicitly progressive redistribution — the bottom of the distribution gets the largest cost cut during downturns. |
▸ simulation math: · closed-form model: · scenario presets: ·
Baseline weekly ledger (hand-verified)
| Metric | Scenario A (Traditional) | Scenario B (Cooperative) | B − A |
|---|---|---|---|
| Income | $150.00 | $150.00 | $0.00 |
| Food cost (effective) | $75.00 | $62.85 | +$12.15 |
| Coop fee | — | $1.00 | −$1.00 |
| GroToken accrual | — | 0.53 tok ($1.06) | +$1.06 |
| Net weekly change | +$75.00 | +$87.21 | +$12.21 (+16.3%) |
Over 156 weeks: ~$1,900+ per-member advantage · ~$95K+ community total (GroToken balance compounds since it's never spent)
Shape attribution · compounding, math-provable, tangible
Shape attribution is what makes PMOVES non-extractive. Unlike yield-on-holdings (which the GENIUS Act bars), attribution is contribution-keyed: you earn share by doing work, measured by the Dirichlet distribution, committed to a Merkle tree, and exportable as a CGP packet on the Geometry Bus.
α accumulates additively
Every contribution adds to your Dirichlet α: α_i += amount × concentrationK. Your weight = α_i / Σα. Multiple contributions shift your share of the total — two people who contribute to the same shape both earn proportional share. It's not interest; it's earned equity that grows with sustained work.
Merkle inclusion proofs
Every attribution record gets a Merkle proof: {root, leafHash, path, pathIndices}. Anyone can verify verifyProof(leaf, proof) without trusting the operator. The root is committed on-chain via the CGP packet — tampering with any record breaks the root.
Real amounts, real records
The verified-actuals run above shows $1,880.71 pool→member transfers with real addresses, action types, and weeks. Shape attribution records measurable economic activity — not abstract reputation scores. Every transaction is auditable.
The full attribution API (what the code actually does)
| Method | Returns | What it proves |
|---|---|---|
| recordAction(addr, action, amount, week, cat) | chitId | Contribution is recorded with Dirichlet weight |
| verifyProof(leafHash, proof) | boolean | Merkle inclusion — the record is in the committed tree |
| getDirichletWeights(category?) | [{address, weight, α}] | Fair-share partition (weights sum to 1.0) |
| getHyperbolicEncoding(category?) | Poincaré points | Attribution mapped to geometric position (more share → closer to origin) |
| exportCGP(week) | CGPDocument | Full shape packet on the Geometry Bus (spec chit.cgp.v1.0) |
The hyperbolic distance formula d_H(u,v) = acosh(1 + 2|u−v|²/((1−|u|²)(1−|v|²))) exists in the codebase but is not yet wired into the attribution pipeline. When it is, it unlocks something the patent system never could: proof of independent discovery.
The Jungian insight
Carl Jung described the collective unconscious — a shared substrate of archetypes that independent thinkers tap into. The history of science confirms it: calculus (Newton + Leibniz), evolution (Darwin + Wallace), the telephone (Bell + Gray). Multiple people arrive at the same idea because the idea exists in the geometry of what's knowable. The patent system treats this as a race — first to file wins, everyone else loses. Shape attribution treats it as what it actually is: convergent discovery, provably independent, both credited.
When two CGP packets land in the same region of hyperbolic space (small d_H), the geometry says: these ideas are the same shape. If their Merkle proofs show different source paths (different authors, different timestamps, different inputs), they are independently discovered. Both get Dirichlet weight in the shared constellation. Nobody loses.
What this means for humans
| Domain | Today (adversarial) | With convergence attribution |
|---|---|---|
| Intellectual property | Race to patent. First to file wins; independent discoverers get nothing. Lawyers profit. | Independent discoverers both get proportional attribution — provable via CGP source-path + hyperbolic distance. No patent race needed. |
| Education | Standardized tests measure compliance, not understanding. Diplomas are opaque credentials. | A student who demonstrates understanding produces a CGP in the same hyperbolic region as the original thinker. The attribution IS the diploma — geometric, verifiable, portable. |
| Aligned incentives | Winner-take-all. Sharing knowledge reduces your competitive advantage. | Sharing knowledge creates new constellations where you hold Dirichlet weight. Contribution grows your share — collaboration is the dominant strategy. |
| Translation & cross-domain | Ideas locked in language silos. A farming technique in Spanish and an AI paper in English are incomparable. | CGP encodes meaning as geometry — the "neural esperanto." Same idea in any language maps to the same hyperbolic region. Translation becomes geometric, not linguistic. |
| PBnJ domain convergence | Each Pinokio launcher is an isolated tool. Domains don't connect. | Each launcher emits CGPs to the Geometry Bus. As domains converge (urban farming + AI tutoring + cooperative economics), PBnJ launchers become the interface through which cross-domain attribution flows. New ideas = new constellations at the intersection. |
The pipeline that gets us there
The foundation is already built. What's missing is the wiring:
| Layer | Status | What convergence needs |
|---|---|---|
| Dirichlet weights | SHIPPED | Already partitions attribution fairly across contributors in a constellation |
| Hyperbolic encoding | SHIPPED | Maps weights to Poincaré disk — getHyperbolicEncoding() exists |
| Merkle proofs | SHIPPED | Proves source provenance — independent authors have independent roots |
| CGP export | SHIPPED | Publishes to Geometry Bus — exportCGP(week) emits chit.cgp.v1.0 |
| d_H distance computation | FORMULA EXISTS | The acosh formula is in the code but never called. Wire it: compute d_H between every new CGP and existing constellations |
| Convergence detection | NOT BUILT | If d_H < threshold: merge into shared constellation, preserve source-path, split Dirichlet weight |
| Human attribution UI | NOT BUILT | Surface "you independently discovered X" + "these N people also found it" — the diploma layer |
Everything below is real-world context: the grant programs, certifications, and cooperative fleet actuals that PMOVES.AI was built to serve. The 2022 Hyundai Ioniq 5 purchased for $24,440 in July 2026 is not just a shuttle — it is the first physical asset of the PMOVES cooperative fleet, qualifying for GCEW/OJET EV economic-development grants (NYC/NYS) and bootstrapping community-owned mobile infrastructure. The cooperative model that the simulator above proves mathematically is the same model that More Perfect Union documented in their 740K-view report on worker co-ops ending corporate monopolies — PMOVES.AI is the software that makes it operational.
Cumulative net profit, 2025 – 2029 · per business archetype PILOT ECONOMICS
Container ROI vs. replication scale · the Docker-style argument PILOT ECONOMICS
Community economic impact · three participation rates PILOT ECONOMICS
| Scenario | Participants | Total invest | Annual income lift | ×Multiplier | Total impact | Community ROI |
|---|
Where CHIT plugs into the unit economics
Each tokenomics primitive maps to a specific CHIT pillar and a NATS subject already in production on the Geometry Bus. CHIT is what turns four independent projection spreadsheets into a single, signed, replicable economic protocol.
| Tokenomics primitive | CHIT role | Carrier |
|---|
What these projections are — and what's real
- What's real: the simulator (above) with its 14 preset scenarios, agent-based modeling, and hard-asserted B>A wealth + Gini reduction. The cooperative thesis — that worker-owned businesses retain wealth locally and reduce inequality — is independently documented by More Perfect Union's reporting on worker co-ops (740K views, June 2026).
- What's real: the Ioniq 5 is PMOVES mobile infrastructure — the first physical asset of the cooperative fleet. It qualifies for GCEW/OJET EV grants (NYC/NYS economic-development RFAs) and MWBE/SDVOB procurement set-asides. The $24,440 purchase (July 2026) was 63% above the $15K projection assumption — a correction the org captured honestly in its own actuals file.
- What's real: Cataclysm Studios Inc holds MWBE (Minority/Women-owned Business Enterprise) and SDVOB (Service-Disabled Veteran-Owned Business) certifications — unlocking the public procurement set-asides that let a Bronx-rooted cooperative sell services to anchor institutions (hospitals, schools, city agencies).
- The pilot is Fordham Hill Oval, Bronx — 93,000+ residents, 73.5% Hispanic, 81,485 Spanish-speaking households. Locally-owned cooperatives recirculate ~53–82% of revenue locally vs ~14% for chains. The 3.32× local multiplier in the community-impact table is modeled from that economic-development literature.
- ROI figures in the projection charts are formulaic extrapolations with assumed success probabilities — not booked revenue. They model the cooperative thesis across different business archetypes; they are not forecasts. Six archetypes share the same curve because they all test the same cooperative economics with different domain labels.
- Shape attribution is what makes the cooperative model grant-auditable: every contribution, every shuttle ride, every service hour is recorded with Dirichlet weights and Merkle proofs. This is exactly what GCEW/OJET/MWBE reviewers need — provable, auditable evidence of community impact, local hiring, and equitable reward distribution.
The ten-step tour
A linear walkthrough that ties every piece together — from "information has shape" to "the framework is the intelligence."
Source coverage & assumptions Read me
Every claim in this dashboard is grounded in a specific repo file. Where the docs are silent, we say so here rather than fill the gap.
Live explainers · CHIT geometry meets Tokenism economics
This tour's 3D visualizations — the Poincaré disk (§08), hyperboloid projection, and D3 force graph (§04) — show the geometric infrastructure. The Tokenism simulator shows the economic proof. Together they answer: does cooperative geometry produce more wealth with less inequality? (Yes — asserted across 14 scenarios.)
CHIT Visual Tour
The Poincaré disk computes a real Sarkar embedding of 97 agents. The hyperboloid lift shows what flattening costs. The pillar labs run Dirichlet, Merkle, Zeta, and swarm math live.
§03 Five Pillars · §04 Geometry Bus · §08 Poincaré Disk
Tokenism Simulator →
11 preset scenarios comparing Traditional vs Cooperative models. Agent-based, 50 members, 156 weeks. Sensitivity analysis on 7 parameters. Mathematical model documentation with formulas and validation.
Cooperative fleet & grants
The Ioniq 5 ($24,440) is PMOVES mobile infrastructure. MWBE/SDVOB certifications unlock NYC/NYS procurement set-asides. GCEW/OJET EV grants fund the cooperative fleet. Shape attribution makes every contribution grant-auditable.