2. Threat Model: What Must Survive

Jason St George. "2. Threat Model: What Must Survive" in AfterFiat: The Load-Bearing Thesis. Version v3.5. /v/3.5/brief/read/threat-model-what-must-survive/

Threat Model: What Must Survive

The protected object is not a token balance in isolation. It is the complete Create/Compute $\rightarrow$ Prove $\rightarrow$ Settle $\rightarrow$ Verify loop and the physical and institutional substrate beneath it. Four properties must survive:

1. Integrity of state and receipts. Computation, provenance, settlement, and consensus artifacts must not be silently corrupted.

2. Confidentiality of flows. Privacy must survive ordinary use; optional disclosure must be scoped by the holder—per payment, per epoch—and cannot mean a universal transaction graph.

3. Availability and neutrality. Users and providers require open admission, reachable routes, non-seizable settlement, and practical exit.

4. Verification economics. Checking a claim must remain cheap in absolute terms and cheap relative to producing it. Otherwise public verification degrades into a trusted-verifier priesthood. The full edition distinguishes three modalities: succinct proofs (M1), whose checking cost is independent of the work; algebraic or probabilistic checks (M2), such as Freivalds’ matrix-product test; and replication or attestation (M3), which rests on a committee. Only M1 and M2 carry the monetary mechanism; M3 supports service delivery and nothing more.

Adversary classes

The rational sovereign under stress uses the levers it controls: negative real rates, capital controls, custody rules, identity mandates, hardware standards, app stores, telecoms, clouds, and interconnection queues. The likely form is often not prohibition but administrative repression: formally optional rails become practically mandatory through defaults, compliance rules, tax treatment, benefits, procurement, platform terms, and institutional mandates.

Platforms and strategic patrons can centralize provenance, decide which proofs count, privilege national champions, and convert private compute into protected quasi-public infrastructure. This enclosure by rescue may be more dangerous than overt attack because the closed competitor can be well funded, technically excellent, energy-privileged, and genuinely useful.

Economic and compositional adversaries include prover cartels, router capture, liquidity games, dealer hedging, wrapper redemptions, leverage, and common operating rules. No participant need be malicious. Locally rational behavior can still produce correlated sales, recursive leverage, procyclicality, concentration, and prices that reveal little about native use.

Hardware, supply-chain, and physical adversaries exploit biased randomness, closed attestation roots, firmware control, export restrictions, curtailment, discriminatory tariffs, cooling constraints, transformer queues, network cuts, or load priority. A state that finds verification awkward to ban can decline to energize proving or make unprivileged verifier hardware unobtainable.

Cryptanalytic, governance, and vaporware risks remain ordinary but fatal: proof-system breaks, captured update keys, unverifiable benchmarks, silent parameter changes, and performance claims without reproducible data.

The competent closed-stack competitor

The real competitor is not necessarily a failing state. It is a competent closed sovereign stack integrating energy, industry, compute, identity, payments, and duration warehousing. Such a system may win on cost, latency, uptime, and construction speed. AfterFiat’s claim is not capability superiority. It is that non-custodial settlement, private-by-default disclosure, portable identity, independently checkable receipts, and practical exit are distinct properties. A closed stack withholds them because that withholding is what makes it closed.

The scoreboard must therefore include agency and exit. Two systems can show identical throughput and uptime while giving opposite answers to the question: can the participant leave with keys, assets, proofs, and usable service?

Standing inversion

The marginal buyer of cheap verification is increasingly a machine delegating to another machine, several hops from a principal who may not have specified the task. That buyer brings a failure mode the stack itself creates. Agents can acquire wallets, receipt-built reputations, and collateral positions—every measurable property of a good holder—while human title over them stops binding: the key root is not held by a person, the delegation ceiling can be raised by the agent or its operator, and revocation is a request rather than a protocol operation.

There is a second route to the same place on which every key stays with a person and is never used. A bad outcome after following an agent’s standing instruction is charged to the instruction; a bad outcome after overriding it is charged to the person; and a stack that makes every decision cheap to reconstruct widens the gap. Title stays intact while judgment is delegated with the keys. Every existing board reads green on both routes. The full edition therefore classifies native demand and holders by ultimate principal and reads standing as a pair: the share of agent-held balances a person can still revoke, against the policy concentration of the same cohort. Humans holding every root while the holdings move as one policy is standing exercised by nobody.

New here? Start with the one-minute version.

Tip: hover a heading to reveal its permalink symbol for copying.