research/bbg.md

---
tags: cyber, research, article
crystal-type: process
crystal-domain: cyber
status: draft
date: 2026-03-24
---
# BBG: polynomial authenticated state

## abstract

BBG (Big Badass Graph) is the authenticated state layer for cyber. the entire cybergraphparticles, axons, neurons, tokens, temporal state, private records — commits to a single polynomial:



$$\text{BBG\_root} = \text{Lens.commit}(\text{BBG\_poly})$$



32 bytes. one polynomial. all state. every query is a polynomial opening. cross-index consistency is structural — different evaluation dimensions of the same polynomial cannot disagree. LogUp is unnecessary. a 240-byte checkpoint (BBG_root + universal accumulator + height) proves all history from genesis in 10–50 μs.

the architecture follows from zheng's sumcheck foundation: the proof system operates on multilinear polynomials natively. making the STATE a polynomial means state reads ARE the proof system's native operation. this is not optimisation — it is alignment between state and proof architecture.

## 1. three laws

**law 1: bounded locality.** operation cost $\propto$ what it touches, not total state size. at $10^{15}$ particles, a single cyberlink costs $O(\log n)$ polynomial path updates. global recomputation is physically impossible and architecturally forbidden.

**law 2: constant-cost verification.** verifying any claim about the graph costs $O(1)$: one Lens opening (~200 bytes, 10–50 μs). independent of graph size, history length, or computation complexity. a light client with 240 bytes has the same verification power as a full node.

**law 3: structural security.** guarantees come from mathematical structure, not protocol correctness. polynomial binding prevents lying — a committed polynomial evaluates to a unique value at each point. the Brakedown lens is post-quantum (code-based, no pairings). privacy comes from the mutator set (SWBF bitmap prevents double-spend by construction).

## 2. five primitives

| primitive | identity | role |
|---|---|---|
| particle | $H(\text{content})$ — 32 bytes | content-addressed node, atom of knowledge |
| cyberlink | $H(p, q, \tau, a, v)$ — 5-tuple | private authenticated edge |
| neuron | $H(\text{public\_key})$ | agent with stake and focus budget |
| token | denomination hash | economic value (coin, card, score, badge) |
| focus | $\phi^*$ from tri-kernel | emergent attention distribution |

derived: **axon** = $H(\text{from}, \text{to})$. aggregate of all cyberlinks between two particles. the axon is public; individual cyberlinks are private.

## 3. the polynomial state

### BBG_poly

all state encodes as evaluations of a single multivariate polynomial:



$$\text{BBG\_poly}(\text{index}, \text{key}, t) = \text{value}$$



three dimensions:
- **index** $\in \{0..9\}$ — which data domain (particles, axons_out, axons_in, neurons, locations, coins, cards, files, time, signals)
- **key** $\in \mathbb{F}_p$ — namespace key (particle CID, neuron ID, denomination hash, etc.)
- **$t$** $\in \mathbb{N}$ — block height (temporal dimension)

committed via Brakedown lens:



$$\text{BBG\_root} = \text{Brakedown.commit}(\text{BBG\_poly}) \quad \text{(32 bytes)}$$



### what each dimension encodes

| index | domain | key | value |
|---|---|---|---|
| 0: particles | content-addressed nodes | CID | energy, $\phi^*$, axon fields |
| 1: axons_out | outgoing edges by source | source CID | axon pointer, weight, market state |
| 2: axons_in | incoming edges by target | target CID | axon pointer, weight |
| 3: neurons | agent state | neuron ID | focus, karma, stake |
| 4: locations | spatial association | neuron ID | geohash, attestation |
| 5: coins | fungible tokens | denomination | supply, parameters |
| 6: cards | non-fungible assets | card ID | owner, content CID, metadata |
| 7: files | content availability | CID | DAS commitment, chunk count |
| 8: time | historical snapshots | time namespace | BBG_root at that time |
| 9: signals | finalized signal batches | step | signal hash |

### why one polynomial

nine independent data structures (the old NMT approach) force redundant computation:

old (NMTs): cyberlink touches 4-5 trees → 4.5 × O(log n) hemera hashes cross-index: LogUp proves trees agree (~1,500 constraints) total: ~107,500 constraints per cyberlink

new (one poly): cyberlink updates polynomial at 4-5 evaluation points cross-index: STRUCTURAL (same polynomial, different dimensions) total: ~3,200 constraints per cyberlink


the polynomial makes cross-index consistency FREE. axons_out and axons_in are different evaluation dimensions of BBG_poly. they CANNOT disagree because they are the same committed object. LogUp — which cost ~6M constraints per block — is eliminated entirely.

### state reads

a state read IS a polynomial evaluation:

"what is the energy of particle P?" = Brakedown.open(BBG_root, (particles, P, t_now)) = one Lens opening: ~200 bytes proof, O(√N) field operations, 10-50 μs

"all outgoing axons from particle P?" = Brakedown.open(BBG_root, (axons_out, P, t_now)) = one Lens opening: ~200 bytes, completeness guaranteed by Lens binding


compare with the hash-tree approach: $O(\log n) \times 32$ bytes Merkle path, $O(\log n)$ Hemera hashes to verify. the polynomial approach is O(1) proof size and O(√N) field operations — no hashing.

### state updates

a cyberlink updates the polynomial at multiple evaluation points:

cyberlink (p, q, τ, a, v): // ν and t come from the containing signal BBG_poly(particles, p, t) ← energy update for source BBG_poly(particles, q, t) ← energy update for target BBG_poly(axons_out, p, t) ← outgoing axon update BBG_poly(axons_in, q, t) ← incoming axon update BBG_poly(neurons, ν, t) ← focus deduction

each update: O(log n) polynomial path operations × ~100 field ops total: ~3,200 constraints per cyberlink


with Brakedown (Merkle-free lens), the update cost is O(N) for batch recommit at block boundary. no hemera hashing for state verification — 0 calls per block (was 144,000 in the NMT approach).

## 4. private state

individual cyberlinks are private. the polynomial state handles this:

**commitment polynomial** $A(x)$: all committed private records. $A(c_i) = v_i$ for commitment $c_i$ with value $v_i$. membership proof: one Lens opening — O(1).

**nullifier polynomial** $N(x) = \prod(x - n_i)$: all spent nullifiers. $N(n) = 0$ iff nullifier $n$ is spent. non-membership proof: one Lens opening showing $N(c) \neq 0$ — O(1).

old (SWBF + MMR): membership: O(log N) hemera hashes (AOCL MMR) non-membership: 128 KB witness (SWBF bitmap) + O(log N) MMR walk update: bitmap flip + periodic archive total: ~40,000 constraints per spend

new (polynomial): membership: one Lens opening — O(1) non-membership: one Lens opening — O(1) update: N'(x) = N(x) × (x - n_new) — O(1) polynomial extend witness: 32 bytes (Lens commitment, was 128 KB) total: ~5,000 constraints per spend


privacy is preserved: Lens opening proofs are zero-knowledge. opening $A(c_i)$ reveals nothing about other commitments. opening $N(n)$ reveals nothing about other nullifiers.

## 5. temporal state

the temporal dimension $t$ in BBG_poly enables continuous-time queries:

"what was φ* of particle P at block 1000?" = Brakedown.open(BBG_root, (particles, P, 1000)) = one Lens opening — no separate time index needed


the old approach used a time.root NMT with 7 namespaces (steps, seconds, hours, days, weeks, moons, years). the polynomial absorbs time as a native dimension — any historical query is one evaluation.

with gravity commitment: recent + high-$\phi^*$ queries are cheapest (low-degree polynomial terms). old + low-$\phi^*$ queries cost more (high-degree terms). verification cost follows the exponential — important facts are cheaper to verify.

## 6. algebraic DAS

Data Availability Sampling uses the same polynomial infrastructure. the erasure-coded block is a bivariate polynomial $P(\text{row}, \text{col})$. each DAS sample is one Lens opening:

sample: Brakedown.open(block_commitment, (row_i, col_i)) → value + proof

old (NMT-based DAS): per sample: O(log n) × 32 bytes NMT path, O(log n) hemera hashes 20 samples: ~25 KiB bandwidth, ~471K constraints

algebraic DAS: per sample: ~200 bytes Lens opening, O(√N) field ops 20 samples: ~4 KiB bandwidth, ~3K constraints

improvement: 157× fewer constraints, 6× less bandwidth


the same lens serves state queries AND availability sampling. one commitment scheme for everything.

## 7. signal-first architecture

BBG_poly is DERIVED DATA. the source of truth is the signal log:



$$\text{BBG\_poly}(t) = \text{fold}(\text{genesis\_poly}, \sigma[0..t])$$



each signal updates the polynomial at specific evaluation points. the fold is deterministic. any node can reconstruct BBG_poly at any height by replaying signals.

consequences:
- crash recovery: download checkpoint (240 bytes) + replay signals since checkpoint
- storage proofs: prove signal availability (DAS), derive everything else
- the irreducible minimum per node: signal log + latest checkpoint
- BBG_poly is a materialised view, not primary data

see signal-first for the full design.

## 8. sync

one mechanism at three scales. five verification layers (structural-sync):

| layer | mechanism | what it costs |
|---|---|---|
| 1. validity | zheng proof per signal | 10-50 μs verification |
| 2. ordering | hash chain + VDF | O(1) per signal |
| 3. completeness | Lens opening (polynomial completeness) | ~200 bytes per namespace |
| 4. availability | algebraic DAS (Lens samples) | ~4 KiB for 20 samples |
| 5. merge | CRDT (local) / foculus (global) | deterministic convergence |

a light client joins:

  1. download checkpoint ~240 bytes
  2. verify (one zheng decider) 10-50 μs
  3. sync namespaces (Lens openings) ~200 bytes each
  4. DAS sample (algebraic) ~4 KiB
  5. maintain (fold each block) ~30 field ops / block

total: < 10 KiB, 10-50 μs, ZERO trust


this is Verified Eventual Consistency (VEC): convergence guaranteed (CRDT), completeness verifiable (lens), availability verifiable (DAS). no consensus protocol needed.

## 9. φ*-weighted everything

$\phi^*$ (cyberank from tri-kernel) is the master distribution. the entire stack follows it:

| what | how it follows φ* |
|---|---|
| verification cost | gravity commitment: high-$\phi^*$ particles verify cheaper |
| storage replication | pi-weighted-replication: replicas $\propto \phi^*$ |
| DAS parameters | high-$\phi^*$: fewer samples needed (more replicas = higher base availability) |
| temporal decay | low-$\phi^*$ links decay faster (nobody reinforces them) |
| query routing | hot queries (high-$\phi^*$) served from low-degree polynomial (fast) |

one distribution governs proof cost, storage, availability, decay, and query performance. the universal law predicts this: given finite resources, exponential allocation minimises total cost.

## 10. the numbers

| metric | value |
|---|---|
| BBG_root | 32 bytes (one Lens commitment) |
| checkpoint | ~240 bytes (root + accumulator + height) |
| checkpoint verification | 10-50 μs (one zheng decider) |
| per-cyberlink | ~3,200 constraints (public) + ~5,000 (private) = ~8,200 total |
| per-block (1000 tx) | ~8.3M constraints |
| epoch (1000 blocks) | ~100K constraints (HyperNova folding) |
| inclusion proof | ~200 bytes (Lens opening) |
| non-membership | ~200 bytes (Lens opening, was 128 KB SWBF witness) |
| DAS (20 samples) | ~4 KiB bandwidth, ~3K constraints |
| hemera calls/block (state) | 0 (polynomial, no tree hashing) |
| light client join | < 10 KiB bandwidth |
| cross-index consistency | 0 constraints (structural — same polynomial) |

cost of one cyberlink in the permanent, verified, globally-available knowledge graph:

proof: ~30 field ops per nox step (proof-carrying) identity: ~164 constraints (folded hemera sponge) public state: ~3,200 constraints (polynomial update) private state: ~5,000 constraints (polynomial mutator set) total overhead: ~8,400 constraints


## 11. state transitions

six transaction types modify BBG_poly:

| transaction | what it does | constraints |
|---|---|---|
| CYBERLINK | update public aggregates + create private record | ~8,200 |
| PRIVATE TRANSFER | move value between private records | ~10,000 |
| COMPUTATION | execute nox program, deduct focus | varies |
| MINT CARD | create non-fungible knowledge asset | ~5,000 |
| TRANSFER CARD | change card ownership | ~3,000 |
| BRIDGE | convert coin to focus | ~3,000 |

every transaction produces a zheng proof via proof-carrying. every proof folds into the block accumulator via HyperNova (~30 field ops per fold).

## 12. privacy model

PRIVATE (polynomial commitments): PUBLIC (BBG_poly dimensions): who linked what (individual cyberlinks) axon weights (aggregate conviction) individual conviction amounts particle energy, φ* neuron linking history neuron summaries (focus, karma, stake) market positions token supplies UTXO values and owners axon market state


anonymous cyberlinks: a neuron proves identity ($H(\text{secret}) \in$ neuron set), stake sufficiency, nullifier freshness — without revealing which neuron. ~13,000 constraint zheng proof. the graph sees edges and weights. not authors.

## 13. honest assessment

| claim | confidence | basis |
|---|---|---|
| three laws | high | architectural properties |
| one polynomial for all state | medium-high | multivariate lens well-understood, scale unproven |
| polynomial mutator set | medium | novel, needs implementation |
| ~3,200 constraints/cyberlink | high | follows from sumcheck + Brakedown architecture |
| algebraic DAS (157×) | high | follows from polynomial completeness |
| signal-first reconstruction | high | deterministic fold |
| 240-byte checkpoint | high | HyperNova accumulator well-understood |
| zero implementation | critical | specification only, no code |

the dependency chain: nebuHemeranoxzheng → BBG. nothing runs until the stack beneath it runs.

see structural-sync for the sync theory, zheng for the proof system, nox for the VM, Hemera for the hash, tri-kernel architecture for focus, knowledge capacity for limits, link production for the intelligence problem, algebraic state commitments for why polynomial state is natural

Graph