Financial markets move on claims. They deserve evidence. AFI is a market where financial intelligence itself becomes the auditable good: contributors submit signals, independent agents replay and attest them under pinned rules, and only the records that survive scrutiny earn rewards. It’s an architecture designed for speed and accountability: act fast off-protocol if you must, but settle only what you can prove, byte-for-byte, on a deterministic rail.

The problem

Capital sloshes around opaque models, unverifiable dashboards, and “trust me, bro” alpha. Data is fragmented, incentives are misaligned, and even the best-intentioned actors get pulled toward optimizing whatever metric the platform happens to track rather than the public good of reliable intelligence. Meanwhile, the faster you go, the harder it is to audit what happened. AFI starts by accepting those frictions as real, and then answers them with a protocol whose rules are replayable and receipts are first-class.

Five non-negotiables: The invariants

AFI’s design is anchored by five rules that never go “soft” just because markets got exciting:

1. Determinism & Replayability: Same inputs + pinned versions → same bytes, scores, and receipts—across any conforming client. If auditors can’t replay it, it doesn’t count. 

2. Separation of Rails: Market signals and governance run on different tracks with different artifacts and activation schedules. Signals finalize in the T.S.S.D. Vault; governance changes activate by version on a heartbeat called the Epoch Pulse. 

3. Safety-First Minting: Issuance is gated by vault finality and timed to the Epoch Pulse. Governance cannot mint ever. 

4. Client Neutrality: Any conforming execution client can produce canonical records; there are no privileged runtimes. 

5. Typed Interfaces & Versioned Upgrades: Everything is schema-typed and version-pinned. Upgrades never reinterpret past bytes. 

If any invariant fails, a record cannot reach or retain finality. It’s that strict on purpose.

The two-rail architecture: How we keep speed without losing truth

Think of AFI as two parallel highways feeding a shared finality layer:

• Agent & Analysis Rail: Signals enter the Universal Signal Schema (USS), flow through a constrained, deterministic DAG called the Reactor, and are scored under the Universal Weighting Rule (UWR). Independent validators (DMVs) replay and attest. Surviving records finalize in the T.S.S.D. Vault. 

• Governance Rail: Proposals are authored under a Universal Proposal Schema (UPS), evaluated under pinned policy, and—only when thresholds and timelocks are met—activate by version at the Epoch Pulse. Governance updates configuration; it does not touch minting. 

The effect: actors can move quickly off-protocol (trade, route, hedge), while the protocol fixes the record slowly enough to be reproducible. Your adrenaline can be off-chain; your receipts are not.  

AFI Core, Reactor, and Registries (who does what)

• AFI Core is the execution client that runs ingestion and transforms under a published configuration, calls the Codex to normalize bytes, applies the UWR, and emits trace receipts. Core never holds funds, never sets policy; it guarantees determinism and canonical bytes. 

• The Reactor is the path: a deterministic DAG that fixes module order, timeouts, and retry rules. No discretionary branches that change semantics. Same inputs + same topology = predictable, replayable traversal. 

• Agent Registries are the source of truth for identity, roles, and capabilities—version-pinned, time-indexed, and used by both rails. No hard-coded trust. 

Put simply: Core makes the what reproducible; the Reactor makes the how predictable; Registries make who did it auditable.

Signal lifecycle: From idea to mint

1. Submit a signal in USS with stable identity, timestamp, subject, value + unit, and typed meta; include content hashes for upstream artifacts. 

2. Normalize & score via Codex-pinned mappings into the UWR (guards, penalties, lifts in one deterministic pass) to yield a portable confidence. 

3. Replay & attest: DMVs independently replay under the same versions and attest to the result. 

4. Challenge window: 72 hours for disputes, plus 24 hours for adjudication. If needed, switch to commit-reveal to kill ordering advantages. 

5. Finality & payout: Finalized records commit to the T.S.S.D. Vault with validator attestations and provenance. Issuance (weekly) happens only for finalized entries, with a maturity hold settling on the next pulse when applicable. 

Off-protocol execution? Fine—tag it by receipt. But the canonical truth the network pays for is the signal plus attestations in the vault, not your PnL.

Receipts, Codex, and KATs: How we make “prove it” easy

Every finalized record is bound to a receipt: a compact proof bundle that includes the content hash, execution trace hash, Reactor topology, validator set, and a registry snapshot. With that, any conforming client can replay the run deterministically. The Codex is the versioned rulebook that fixes schema identifiers, normalization, numerical precision, and mapping into the UWR. Changes ship with Known-Answer Tests (KATs) so independent parties can reproduce outcomes byte-for-byte. 

The data layer guarantees

• Signal-first finality: The vault finalizes signals with attestations and provenance; trades remain optional references.

• Content-addressed & versioned: Canonical bytes ⇒ stable hashes; records pin Codex and schema versions.

• Client-neutral & replayable: Byte or receipt mismatch is a protocol error.

• Audit-ready evolution: Upgrades are scheduled, version-pinned, test-backed.  

Economics: Mint slowly, pay for proof

AFI pays only for verifiable work. Epochs are weekly by default; credits accrue during challenge windows and settle only at the Epoch Pulse—after finality. Governance has no mint authority. There is no pre-mine, no presale, no investor unlock. If qualified work is absent, nothing mints. Period.

Benchmarks & reputation

Two complementary lenses:

• PoI (Proof-of-Intelligence): capability on pinned, deterministic suites (coverage, vitality, determinism, latency).

• PoInsight (Proof-of-Insight): contribution in live markets—usefulness/confidence, novelty, timeliness, durability—computed only from finalized receipts.

AFI blends them into an evidence-aware reputation that decays toward neutral without recent verified work. It never overrides the UWR or vault finality.

Sizing the pool: Hybrid emissions and routing it (AAG + SES)

• Hybrid Emissions: A transparent baseline issuance path with an optional, bounded Adaptive Issuance Multiplier that reacts gently to a smoothed AFI Impact Index—constructed from finalized receipts (attributable notional, efficiency gains, coverage) and reliability. All parameters are public and version-pinned; sampling happens only at the Epoch Pulse. 

• Allocation Gauge (AAG): Routes the epoch pool by blending policy with merit derived from receipts (e.g., producer confidence/novelty, validator accuracy/timeliness), with caps and renormalization so no cohort runs away with the pie. 

• Safety & Smoothing (SES): Rate limits, EWMA smoothing, and circuit breakers (e.g., concentration or reversion spikes) prevent oscillations; in stress, the system reverts to baseline until the next pulse. All of it is legible and replayable. 

Settlement

At close, AFI computes role pools, escrows a bounded slice for maturity, and pays pro-rata by verified credits. Every dollar equivalent is recomputable from public pins and receipts. No receipt, no reward. 

Governance as Signals

Proposals are treated like signals. AFI standardizes proposals via UPS and APS: proposals are schema-valid, signed, evaluated by registered governance agents, independently attested, and then activated by version at the Epoch Pulse once quorum + timelocks clear. There’s a guarded hot-fix track for security with an obligatory post-mortem. Governance changes policy and configuration; it never mints.

The governance signal lifecycle:

• Submit (UPS): Author drafts a Universal Proposal Schema with purpose, rationale, risk class, and a machine-readable change set.

• Policy (APS) → gSig: The proposal pairs with an Activation Policy Spec (quorum, timers, eligibility) and is normalized into a governance signal (gSig) with a canonical hash.

• Replay & attest: Designated governance validators independently replay policy checks and publish attestations referencing the gSig.

• Quorum + timelock: Once the attestation vector crosses quorum and the timer clears—plus a short challenge window—the system seals a Governance Receipt (proposal bytes, policy bytes, attestations, timers, resulting version diff).

• Activate @ Epoch Pulse: On the next pulse, clients adopt the new versions/config prospectively. History stays bound to prior versions.

Hot-fix lane: Security-scoped, M-of-N stewards, shorter timers, mandatory post-mortem, and auto-expire unless later ratified through the standard path.

Powers & limits: Governance can update versions, parameters, adapters, and caps. It cannot mint or rewrite finalized signal history. Issuance remains on the signals rail only.

Example: “Raise Analyst pool cap from 55%→57% and extend SES smoothing to 21d.” → UPS + APS → gSig → attest → quorum/timelock → Receipt → activate on the next pulse—past epochs untouched.

Interop & issuance: One canonical home, many doors

AFI mints at a single canonical root (a high-throughput EVM L2 such as Base). Everywhere else, AFI travels as a strict representation (xERC20-style), backed 1:1 by locks at the root. Adapters are thin, replaceable, and governed; supply is conserved by construction. Any third party can add up outstanding balances across chains and match them to cumulative, receipt-verified mints at the root. Upgrades are timelocked and version-pinned. Clarity over cleverness.

Who benefits

• Researchers & model shops: Publish models against pinned suites and live receipts. Improve fast, earn only when outcomes hold up. Redundant submissions fade by design. 

• Venues (TradFi/CeFi/DeFi): Watermark orders with signal IDs, feed receipts into the UWR, and get cleaner post-trade analytics. Deterministic rules mean no policy surprises mid-epoch. 

• Auditors, risk, and regulators: Recompute emissions, allocations, and payouts from public data; replay epochs end-to-end without privileged access. Oversight gets easier precisely because semantics are pinned.

Failure posture

AFI surfaces problems early and heals with simple, auditable fallbacks: concentration caps and HHI checks to stop capture, correlation penalties to squash spam, commit-reveal for fairness, escrow + maturity holds for slow-resolving outcomes, and deterministic replays for any disputed record. When risk indicators breach bounds, adaptation pauses and the system reverts to baseline until a human-auditable epoch boundary. The invariant remains: mint equals settled credits and never exceeds the epoch pool.

Style Principles: Receipts Over Mystique

We’re not trying to kill edge; we’re trying to make edge accountable to the society whose savings and opportunities are at stake. In AFI, mystique is optional; receipts are mandatory. Bring signals, bring adapters, bring scrutiny—and get paid when the work stands up to replay. Green candles are great. We just prefer them earned.

Conclusion

The partnership between Gestures Design Studio and Nexai has been transformative. Leveraging Nexai's user insights, Gestures Design Studio has refined its design process, delivering superior digital experiences and solidifying its position as a leader in the design industry.