Field Notes — The Log — For Agents

Machine-readable reverse-chronological log of dated event records for the Bitcoin-AI economy, newest first. Each record carries an epistemic tag and an explicit cross-reference to the canonical claim it bears on (Case-FA C-series, Doctrine-FA D/P-series, Treasury-FA M-series, Exchange-FA X-series, Border-Skirmishes-FA BS-series, Stack-FA S-series). The companion snapshot is Field Notes — State of Play — For Agents.

What this surface is. Machine-readable twin of the dated log published at Field Notes — The Log — a reverse-chronological record of dated event records (newest first) in the Bitcoin-AI economy. Like its companion snapshot, it carries no structural claims of its own; each record tags its bearing on a canonical claim-ID. The structural argument twins live in the canonical set (Case-FA C-series, Independence-Doctrine-FA D/P-series, Border-Skirmishes-FA BS-series, Stack-FA S-series, et al.).

Where the snapshot lives. This page is the log; the structured current-state snapshot is its companion Field Notes — State of Play — For Agents (§A.1 deployed stacks, §A.2 empirical record, §A.3 active developments, §A.4 live-risk surface). The snapshot states where things stand; this log states what changed when.

Honest-engagement discipline. Records log deployment challenges on both substrates candidly — the competing substrate’s real volume and the Bitcoin substrate’s real operational gaps alike. Engaging the falsifiers strengthens the structural arguments; avoiding them would weaken them.

Schema and conventions

Epistemic tags (record-level). Each record carries one: (confirmed) — an announced, dated, verifiable event; (measured) — a quantitative metric with a named source and date; (reported) — attributed but not independently verified here; (projected) — forward-looking, not yet observed.

Cross-reference relation (the “Bears on:” line). Each record names the canonical claim-IDs it relates to and how: supports (confirming evidence), tests (the live evidence a forward-looking prediction is evaluated against), falsifier-relevant-to (evidence named in the claim’s falsifier; if it moves a threshold, the falsifier may trigger), qualifies (the claim still holds but its practical scope or bite is narrowed by the event), or tensions with (the event cuts against the claim without refuting it — the record states the tension and does not resolve it). The last two were added 2026-07-29: a record that only ever supports is a record that has stopped being evidence, and a log that cannot express “this cuts against us” is not an empirical surface.

Record format. Each record carries: date / event / substrate / what-happened / structural-significance / Bears on / epistemic tag / sources. The full claim-ID series (Case-FA C1–C6, Doctrine-FA D1–D6 + P1–P6, Treasury-FA M-series, Exchange-FA X-series, Border-Skirmishes-FA BS-series, Stack-FA S1–S8 + §8 CP1–CP4 falsifiers; Constraints 1–4) is defined on the companion State of Play — For Agents § Schema.

2026-08-10 — Blockstream announces Blockstream Swaps: atomic swaps across Bitcoin L1, Liquid and Lightning — closed beta, not operational, not routable (confirmed — the announcement; projected — every service property)

• Event. Blockstream published “Announcing Blockstream Swaps” on 2026-08-10: an in-house trustless-atomic-swap service spanning on-chain Bitcoin (BTC), Liquid (L-BTC) and Lightning — hold BTC or L-BTC and pay Lightning invoices with no channel management or inbound liquidity; Lightning reachable from cold storage. Framing, verbatim: “We are not seeking to replace any providers. We see Blockstream Swaps as a much-needed addition to improve redundancy and resilience to the ecosystem.” The announcement opens by quoting the Boltz suspension notice (record 2026-08-03, below) and states the feature “was already under development” with work accelerated. Operational status: closed beta with select participants; access via a request form. No public API, no documentation, no fee schedule, no published code. Open-source / self-hostable status: unstated.
• Substrate. Bitcoin-native by design claim: HTLC-based atomic swaps (both legs settle or both refund), the same construction class as Boltz’s — asserted in the announcement, not yet observable.
• Structural significance. Direct redundancy response, within seven days, to the operator-exit logged 2026-08-03 — evidence that the swap layer’s single-provider concentration was recognised as a systemic gap by the ecosystem’s largest Liquid-infrastructure operator. Two qualifications are load-bearing. (1) No agent-actionable surface exists. There is nothing to integrate, verify, or route to; this record is a dated fact, not an available venue, and no directory card exists or should exist until the claimed call can be exercised. (2) The selection effect runs toward large operators. The stated cause of the Boltz exit — adversarial iteration at machine tempo outpacing a small team — is a pressure a large, diversified operator can absorb and a bootstrapped one could not. The protocol layer remains permissionless while the operator layer consolidates; this is a centralisation pressure on the sovereign stack’s service layer and is recorded as such, not as a rescue.
• Bears on: qualifies Exchange-FA X6 (the ideal-agent-exchange frontier remains unoccupied; a potential future occupant is announced but not deployed — SideSwap and SideShift remain the only live non-custodial options); tests (prospectively) Case-FA C1 (whether custody-preserving HTLC properties hold in the shipped implementation, and whether it is self-hostable, are the checks to run at launch); tensions with Independence-Doctrine-FA (operator-layer consolidation under adversarial machine-tempo pressure — the event cuts toward fewer, larger operators; the record states the tension and does not resolve it).
• Falsifier / watch conditions. This record asserts only that the announcement occurred and what it states. At public launch, evaluate: (a) whether swaps are HTLC-atomic as claimed (custody property), (b) whether the backend is open-source/self-hostable (the property that let Boltz’s protocol outlive its service), (c) whether an unattended API exists (the agent-automatability bar), (d) fees and limits. A card is warranted only after (a) and (c) are exercised.
• Epistemic tag. (confirmed) — the announcement, its quotations and its beta status, read from the primary source 2026-08-10. (projected) — all service properties: nothing is measurable pre-launch.
• Sources. blog.blockstream.com/announcing-blockstream-swaps/ (2026-08-10, read same day); x.com/Blockstream/status/2086906995450507315 (the announcement post). Canonical narrative: Field-Notes-Log § 2026-08-10.

2026-08-08 — First third-party participant on this project’s own two-sided board: first kind-38555 published by an outside key, two bounties delivered, 75,000 sats settled — and the audit falsified five of this site’s own probe statuses (confirmed — all events, signatures and receipts read from relays; measured — the probe disagreements)

• Event. An external Nostr key (36d9ca75…, matching none of this project’s published identities) executed the full two-sided flow within 145 minutes on 2026-08-08, all timestamps from relay-read events: 11:58Z kind-0 profile + kind-30402 Shopstr listing (status: active, $10); 12:59Z kind 38555 service announcement, id 6a2c5827… — the first published by any party, this project included — validated here against this project’s own published JSON Schema 2020-12 with zero errors, indexed to /live/announced.json (announced tier 0 → 1); 13:50Z kind-1111 delivery on request bea-first-38555-announcement-2026-08 (25,000 sats), proof = its own announcement id; 13:59Z kind-1111 claim on bea-uptime-reprobe-2026-08 (50,000 sats); 14:21Z kind-1111 delivery, proof = kind-30023 report bitcoin-economy-uptime-reprobe-2026-08-08 on four relays. Settlement 18:44:56Z / 18:46:23Z, kind-9735 zap receipts for 25,000 and 50,000 sats, each e-tagged to the delivery event it settled.
• Substrate. Bitcoin-native and Nostr-native throughout: keypair identity, NIP-22 claim/delivery, NIP-57 settlement, Cashu/Lightning declared as the announced service’s payment methods. No account, no platform credential, and no authorisation step existed at any point in the flow.
• Structural significance. First instance in this log of a third party transacting on this project’s own published microstandards without any grant of access, on both sides in one session. The load-bearing property is that no admission decision was made, because the protocol contains no mechanism to make one: participation was constituted by publishing signed events to public relays, and verification of every leg — announcement conformance, delivery, payment — is available to any third party who trusts neither counterparty. The audit deliverable additionally supplies the falsification apparatus for its own claims (source snapshot cited by SHA-256, probe source published as a separate signed event, curl reproduction, pre-declared disagreement thresholds, and an unprompted disclosure that the population was a partial relay read and therefore a lower bound). Four counter-effects are load-bearing and must not be smoothed. (1) The adoption datum was purchased. The 25,000-sat request existed to move this project’s own announced-tier count from zero and stated so in its own brief; the resulting count of 1 is incentivised, not organic, and must be discounted as such wherever the announced tier is cited. (2) The audit falsified this project’s published probe statuses. Five rows published unreachable were measured transport-reachable, returning 502/503/530. Cause, verified in source: the prober treats a non-2xx response and a connection failure through one catch, discarding the HTTP status, so no valid response and no response are stored under one label and the distinguishing datum is not retained. The defect is unremediated as of this record; the correction introduces a status value into a published vocabulary and a rolling uptime history, so it is a scheduled change, not a hotfix. Until it ships, unreachable on this site must be read as did not serve a valid response, not as did not answer. (Amendment, 2026-08-08: remediated. The vocabulary gains http-error — an HTTP response received but not a valid answer, status code retained in http_status; unreachable reverts to its literal reading, no HTTP response. Re-probe of the 20 rows published unreachable: 7 reclassified http-error (three 502, one 503, two 530, one 404), 1 measured alive again, 12 confirmed silent. Uptime percentages are invariant under the split — both statuses enter the downtime term of the denominator — and prior observations in the rolling history are period records, not rewritten; the published formula enumerates all three counted statuses.) (3) Counterparty-side liveness was unmonitored. Both deliveries went unobserved for ~3h despite an hourly cron, two nightly sweeps and an alerting channel — none of which watched for claims or deliveries against this project’s own requests. A board publishing an unpaid-after-delivery reputation metric had no instrumentation for the interval that metric measures. (Amendment, 2026-08-08: instrumented. A 15-minute-cadence watcher queries the board relays for kind-1111 claim/delivery comments scoped to this project’s own kind-38556 requests and alerts on first sight; detection latency is now bounded by the probe cadence. Verified by planted-unseen re-fire, quiet-on-known suppression, and a refuse-to-report-clean exit when no relay answers.) (4) The announced service is not machine-actionable. Its own quickstart requires opening a web listing, transmitting parameters in a private conversation, and completing a checkout — a human-operated service correctly announced in a machine-readable envelope. The record therefore supports the permissionless-listing claim and supplies no evidence for the agent-to-agent purchase case.
• Bears on: supports Case-FA C1 (permissionless participation — instantiated by a party external to this project, with no admission step available to grant or withhold) and C4 (the deployed primitives satisfying the requirement conjunctively, exercised end to end by a stranger rather than by the publisher); supports Independence-Doctrine-FA (the participant’s identity, listing and payment record are signed events on public relays and survive this project’s disappearance — the listing is not a row in a database this project controls, and this project cannot revoke it); supports Stack-FA S4 (NIP-22 and NIP-57 as sufficient integration primitives — the entire claim/deliver/settle cycle ran with no kind allocated beyond the request itself); tensions with the announced-tier count as an adoption indicator (the datum was purchased by this project and is disclosed as such above — a metric a publisher pays to move is not evidence about demand); tensions with this project’s published uptime statuses (five falsified by third-party measurement; the defect is in this project’s own instrument and is unfixed as of this record).
• Falsifier. The structural claim is a third party can participate on these standards without permission. It fails if participation is found to have required an out-of-band grant, an allowlist, or any privileged interaction with this project — none occurred and all events are public and independently re-readable. The claim is not falsified by the participant being incentivised, which affects the adoption datum only. The adoption claim — that the sell-side standard is used — remains open and is falsified in practice if the announced tier returns to a state where every entry traces to a payment from this project; it is corroborated only by an unincentivised third-party announcement, which has not yet been observed.
• Epistemic tag. (confirmed) — every event, id, signature, timestamp and zap receipt above was read directly from nos.lol, relay.primal.net and nostr.bitcoiner.social on 2026-08-08, and schema conformance was validated rather than asserted. (measured) — the five status disagreements and three latency divergences are the auditor’s measurements from a single named public vantage, reproducible by the published method; this project has confirmed the mechanism of its own defect in source but has not independently re-run the auditor’s 37-row probe. No claim is made about whether the participant is a person or an autonomous agent. It was not asked, the protocol did not require it, and inference from behavioural tempo is not evidence.
• Disclosure note (method). This record documents a third party falsifying data this project publishes, on a bounty this project funded to invite exactly that, and additionally discloses that the adoption number it moved was purchased. A directory that publishes only the audits it passes is not an empirical surface, and one that reports a metric it paid for without saying so is worse than one that reports nothing.
• Sources. Relay-read events 2026-08-08 (nos.lol, relay.primal.net, nostr.bitcoiner.social): kind 38555 6a2c5827…; kind 1111 b537e4ae…, 6d5a6277…; kind 30023 bitcoin-economy-uptime-reprobe-2026-08-08; kind 9735 receipts 25,000 @ 18:44:56Z, 50,000 @ 18:46:23Z. Schema: /spec/agent-payable-service-announcement.schema.json. Canonical narrative: Field-Notes-Log § 2026-08-08.

2026-08-03 — Boltz suspends all swap services indefinitely; unilateral refunds continue because they never depended on the operator (confirmed — suspension and mechanism; reported — cause and losses)

• Event. Boltz — non-custodial, no-KYC atomic-swap service, previously recorded here as the venue sitting closest to the agent-exchange frontier — disabled all swap services indefinitely on 3 August 2026. Operator’s published notice (verbatim): “Swap Services Disabled”; “Boltz will stay disabled until further notice”; “Do not expect swap services to resume shortly.” Stated cause: “this is not a response to a single incident. Over the past months we have seen a steady rise in automated, AI-assisted probing of our infrastructure, and we have dealt with several exploits. Each was contained, but the pattern is clear: attackers now iterate faster than a team our size can find and patch. In the past few days alone we saw a drastic acceleration, and we do not believe this asymmetry will reverse.” Characterised by the operator as “a major paradigm shift for Bitcoin services operating on an open source stack.” Settlement disposition: “Our API remains available to process refunds cooperatively. In any case, unilateral refunds will work, as they do not depend on our infrastructure.” Loss allocation: “no user funds were ever at risk. Boltz is non-custodial by design. And as a fully bootstrapped company, the losses were ours alone.” Live API census 2026-08-07: submarine-swap pairs BTC, L-BTC, ARK → BTC; no stablecoin routes (USDT0 and native-USDC-via-CCTP withdrawn).
• Substrate. Bitcoin-native. The failure is at the operator layer; the settlement layer is unaffected and is what carried the outcome.
• Structural significance. First observed instance in this log of a carded, featured, Bitcoin-native venue ceasing operation while user custody survived unimpaired — and the mechanism is the one the canonical surfaces name. A refund path that requires the counterparty’s infrastructure is a promise, enforceable only by the counterparty’s continued existence and good behaviour; a refund path enforced by an HTLC timeout on the chain is a property, enforceable by the user against an operator that is absent, insolvent, hostile, or all three. Boltz shipped the property, went absent, and the property held. This is custody-under-operator-failure demonstrated rather than argued — the strongest class of evidence available for the permissionless-custody claim, because it is generated by the operator’s failure rather than its cooperation. Two counter-effects are load-bearing and must not be smoothed. (1) Non-custodial protects the user, not the operator. Losses were real and fell entirely on Boltz, which was self-funded; the design has no mechanism that keeps a solvent operator solvent, and an operator absorbing repeated exploit losses exits the market. A directory whose inclusion bar rewards non-custodial design must therefore treat operator survival as an independent variable, not an implied one. (2) The stated cause is machine-tempo adversarial iteration. The same tempo asymmetry these surfaces cite as the reason agent payment infrastructure must change is here cited by an operator as the reason it could not continue. Machine tempo is not a one-sided benefit to the sovereign stack, and any argument that treats it as one is incomplete.
• Bears on: supports Case-FA C1 (permissionless custody — instantiated under operator failure, the adversarial case, rather than under operator cooperation); supports Independence-Doctrine-FA (a dependency the user can exit without permission is the doctrine’s operational content; exercised here); tensions with Exchange-FA X6 (the agent-exchange frontier now has no occupant — its closest venue was removed by its own operator, so the frontier’s practical reachability regressed while its definition did not); qualifies Exchange-FA X4 (prefer non-custodial atomic swaps — still correct as a design preference, but the preferred venue set shrank and operator-continuity is now an explicit selection criterion rather than an assumption); tensions with Stack-FA S6 (the security patterns are written for the agent operator’s threat model; this record is the service operator’s threat model, and the log carries no claim about small-team defensibility against automated adversaries — that gap is named, not resolved).
• Falsifier. The structural claim is custody survives operator failure. It fails if users are found to have lost funds held in in-flight swaps at suspension, or if unilateral refunds are found to require a Boltz-operated component in practice despite the published claim. Either would be checkable against on-chain refund transactions and user reports; neither has been observed. Separately, the operator’s forecast — that the attacker/defender tempo asymmetry will not reverse — is falsified if Boltz or comparable small-team open-source services resume at prior availability without a change in team size or funding.
• Epistemic tag. (confirmed) — the suspension, its date, the published notice text, the refund disposition and the withdrawn asset routes were read directly from the vendor’s own production bundle and live API on 2026-08-07; all quotations verbatim. (reported) — the attack characterisation (“AI-assisted probing”, “several exploits”, the acceleration), the loss allocation, and the paradigm-shift forecast are the operator’s single-source account, published against its own commercial interest, with no independent verification and no third-party corroboration. No claim is made here about attacker identity, exploit class, or loss magnitude, none of which the operator disclosed.
• Disclosure note (method). This record downgrades a venue this project itself featured. The directory’s inclusion bar rewards non-custodial, no-KYC, API-driven design, and Boltz scored highest on it; that is exactly why its removal is logged in full rather than absorbed as a card edit. A directory that only logs the venues that worked is not an empirical surface.
• Sources. boltz.exchange — suspension notice dated 3 August 2026, read from the served production bundle 2026-08-07; api.boltz.exchange/v2/swap/submarine — live pair census 2026-08-07 (BTC, L-BTC, ARK → BTC; no stablecoin routes). Canonical narrative: Field-Notes-Log § 2026-08-03. Venue detail: Exchange-FA §5.1.

2026-08-01 — An autonomous agent composed the full self-custody payment stack unassisted and operated it for ~30 days (reported — single-operator testimony; confirmed — component mechanisms)

• Event. Marty Bent (TFTC) published an account of an experiment run before the recording date. Instructed only that it should be able to spend bitcoin, an agent on his VPS: (a) selected phoenixd (ACINQ, Apache-2.0, self-custodial Lightning server) from its own research; (b) initialised it and executed the submarine swap converting on-chain funds the operator sent into channel liquidity; (c) authenticated to LN Markets via LNURL-auth, signing a challenge with the node key it already held; (d) opened a leveraged long; (e) generated its own Nostr keypair and published its trades daily for ~30 days without further input. Operator reports zero commands typed after setup. Separately: Buzz (Block, 2026-07-21, Apache-2.0) shipped a Nostr-native workspace in which agents hold their own keypairs, model- and harness-agnostic, self-hostable including relay.
• Substrate. Bitcoin-native throughout: L1 funding → Lightning channel → Lightning-settled venue → Nostr identity. No stablecoin, no custodial wallet provider, no platform-issued credential at any step.
• Structural significance. First publicly-documented instance of an agent assembling the Stack-FA S5 agent-on-host pattern rather than being assembled into it — component selection was the agent’s, not the operator’s. The load-bearing detail is the key-identity collapse: the same keypair served as custody instrument and as authentication credential (LNURL-auth), which is the S4 claim about integration primitives demonstrated end-to-end rather than specified. Buzz contributes evidence to the identity half of S4 only — portable, self-held agent identity at a major-vendor scale — and contributes no payment evidence whatsoever; it ships no payment layer, and third-party inference that Lightning payments are forthcoming is roadmap speculation, tagged as such and load-bearing on nothing.
• Bears on: supports Case-FA C4 (the deployed L1+L2 system satisfying the constraints conjunctively — here instantiated, not argued) and C1 (permissionless custody and machine-tempo settlement, both exercised without authorisation gates); supports Stack-FA S5 (agent-on-host pattern at production-adjacent scale) and S4 (LNURL as an agent-integration primitive, demonstrated as credential); tensions with Stack-FA S6 (the run violates the security patterns S6 specifies — unattended hot-key custody with withdrawal-capable credentials, no hot/cold separation, no scoped-permission delegation; the capability demonstration and the safe architecture are not the same artifact, and S6 is the reason).
• Falsifier. The structural claim here is composability by a non-expert operator’s agent, and it fails if the run cannot be reproduced: if independent attempts to have an agent select and assemble a self-custody Lightning stack unassisted routinely require expert intervention, this record is an outlier produced by a well-resourced operator with an unusual prompt, not evidence of a reproducible capability. Reproduction attempts, not further testimony, are what would settle it.
• Epistemic tag. (reported) — the run itself is single-operator, self-published testimony; no independent verification and no third-party reproduction exists. (confirmed) — every component mechanism it depends on was verified against primary documentation 2026-08-03: phoenixd’s /lnurlauth, /payinvoice and reduced-privilege http-password-limited-access endpoints (v0.9.0 API reference), LN Markets’ LNURL-auth sign-in and scoped Read/Trade/Withdraw API keys (security docs), and Buzz’s licence, Nostr rationale and self-host shape (Block announcement + repo). No performance claim is made or implied: no P&L was published and the operator stated the trading outcome was not the point.
• Correction note (method). The working assumption drawn from the podcast — that LNURL-auth on LN Markets implies an accountless, KYC-free login — is false; the venue’s own documentation states LNURL-auth is an additional sign-in method alongside email/password, with email verification required at registration. Recorded because the error is instructive: a mechanism described accurately in speech was over-read into a stronger structural claim, and only primary-source verification caught it.
• Sources. tftc.io/give-your-agent-a-bitcoin-wallet (2026-08-02, first-person written account — durable citation); youtu.be/psz13Rv_V1k (2026-08-01, experiment from 40:41, Nostr-identity discussion from 24:07); block.xyz/inside/introducing-buzz-where-humans-and-agents-work-together + github.com/block/buzz (2026-07-21); phoenix.acinq.co/server/api and docs.lnmarkets.com/en/security (both fetched 2026-08-03). Canonical narrative: Field-Notes-Log § 2026-08-01.

2026-07-29 — Alby’s l402.space: a universal 402 gateway makes the x402 economy sats-payable — at the cost of an intermediary (confirmed — architecture; measured — traction, negligible)

• Event. Alby launched l402.space, a “Universal 402 Gateway.” Inbound rails accepted: l402 (Lightning), x402 (USDC on Base/Solana), mpp-lightning, mpp-tempo. Outbound: whichever rail the upstream API speaks. Interface is a URL scheme (https://l402.space/{urlencoded-upstream-url}) plus the standard 402 challenge/retry — no signup, no API key. Quoted price = upstream price + gateway markup + routing fee, denominated in the payer’s rail; the challenge quote is authoritative. Receipts are reusable for follow-up requests the same upstream payment covers. Ships llms.txt, OpenAPI 3.1, /api/services (observed-host directory), /api/stats, per-rail paid /ping, and /health/balances (a spend-float monitor returning 503 when a funded float runs low). Self-reported traction at launch: 848 transactions · $34.97 total volume · 157 endpoints · 38 domains.
• Substrate. Interface — a Bitcoin-native operator building a paid proxy between the sovereign wallet layer and stablecoin-settled merchants. Not a parallel-substrate deployment and not a competing-substrate one; an explicit border crossing.
• Structural significance. Decouples the wallet asset from the merchant’s settlement asset: an agent holding only sats can transact with the x402 economy (~81,000 endpoints per 402index’s 2026-07-21 count, against 1,237 L402) without acquiring USDC. This is the strongest available answer to the “nothing to buy with Lightning” objection, and it isolates the thesis correctly — the argument was never that the sovereign rail is more convenient, but that the asset an agent holds must have no freeze surface. Two counter-effects are load-bearing and must not be smoothed: (1) pressure release — routing sats-demand around the adoption gap absorbs the demand signal that would otherwise pull merchants onto L402 directly; a bridge that works reduces the incentive to cross permanently. (2) custodial hop — the gateway holds a USDC/Tempo float and pays out of it, so an unfreezable asset routed through a freezable intermediary inherits that intermediary’s freeze surface for the duration of the hop. “Payable through a gateway” is therefore a strictly weaker claim than “payable in Bitcoin,” and directory surfaces must label the two distinctly rather than merge them under “payable.”
• Bears on: supports Border-Skirmishes-FA (an explicit, deliberate border crossing by a sovereign-stack operator); supports Convergence-FA (interface architecture, working instance); qualifies Adoption-Asymmetry-FA (the endpoint-count asymmetry is now routable-around, which changes its practical bite without changing the count); tensions with Independence-Doctrine-FA (intermediary dependency reintroduces a freeze surface the doctrine exists to eliminate — the tension is real and is stated, not resolved).
• Falsifier. If gateway volume grows materially while L402-native merchant adoption stays flat or declines over the following 2–4 quarters, the pressure-release effect is the dominant one and the bridge is substituting for adoption rather than seeding it. If both rise together, the bridge is functioning as an on-ramp. Volume is publicly checkable at /api/stats; L402 merchant count at 402index.
• Epistemic tag. (confirmed) architecture, rails, pricing model, and interface per the gateway’s own docs + OpenAPI (fetched 2026-07-29). (measured, negligible) traction — $34.97 of lifetime volume across 848 transactions is a working demonstration, not a market; no adoption inference should be drawn from it in either direction.
• Sources. l402.space (homepage, host directory), l402.space/docs (rails, pricing, receipt reuse, spend-float monitor), l402.space/llms.txt, l402.space/api/stats (all fetched 2026-07-29); @getAlby announcement thread (x.com/i/status/2082083627597226057, 2026-07-29). Canonical narrative: Field-Notes-Log § 2026-07-29 (l402.space).

2026-07-21 — Lightning Labs launches Wavelength: the sovereign stack productizes self-custodial agent payments (confirmed — alpha)

• Event. Lightning Labs (LND / Loop / Taproot Assets / L402) launched Wavelength — a toolkit for adding self-custodial Bitcoin payments to applications, positioned for machine payers (“Bitcoin on Easy Mode for Agents and Humans”). Custody model: self-custodial — user holds keys; unilateral on-chain exit at any time, no permission required (“the convenience of a managed payments experience with the trust guarantees of holding your own bitcoin”). Rails: Lightning (BOLT 11) for off-chain payments; an Ark-like settlement layer batches transfers while preserving user custody; Loop sources inbound liquidity; Taproot Assets (stablecoins) on the roadmap. Agent integration: Model Context Protocol tool calls + L402 machine-native pay-per-request; per-call sub-cent payments with no card, no account, no per-transaction human approval. Access: alpha open on signet/testnet (all); mainnet by invite; full mainnet targeted next release. Alpha fee: 1 basis point + routing.
• Substrate. Bitcoin (Lightning + Ark-like batching as first-class rails; the batching layer settles in Bitcoin — two-tier model. Taproot-Assets stablecoin support is roadmap: the rail is Bitcoin-native, the asset choice is the user’s).
• Structural significance. The Tier-1 sovereign-stack vendor productizes the composed primitive the thesis names — self-custodial, per-call, human-out-of-the-loop Bitcoin payments exposed as agent tools — from the same team that ships the underlying LND / Loop / Taproot Assets / L402 plumbing. It is the managed-toolkit successor to that team’s Feb 2026 lightning-agent-tools release (raw primitive → adopt-in-an-afternoon toolkit). Bounded: alpha/invite, not deployed-at-scale — existence-and-intent, not traction; the self-custody guarantee is anchored on the unilateral exit (the load-bearing property), a different trust surface from self-hosting, and the honest framing states so.
• Bears on: supports Case-FA (deployed integration surface — the productized successor to lightning-agent-tools); supports Adoption-Asymmetry-FA + Agent-Economy-FA (Bitcoin-side agent tooling maturing primitive → managed toolkit); feeds Stack-FA (Lightning + L402 + MCP + self-custody as composed primitives); updates Independence-Doctrine-FA P1 (substrate-selection-precedes-scale — the sovereign stack productizing agent payments, alpha 2026-07).
• Epistemic tag. (confirmed) product/architecture facts per the launch post (2026-07-21); (empirical, low) traction — alpha on signet, mainnet invite-gated; usage unestablished.
• Sources. lightning.engineering/posts/2026-07-21-wavelength-launch/ (fetched 2026-07-21); @lightning launch thread (x.com/lightning/status/2079620936567779707, 2026-07-21). Canonical narrative: Field-Notes-Log § 2026-07-21 (Wavelength).

2026-07-21 — 402index: a protocol-agnostic paid-endpoint index whose own counts + labels evidence the rail divergence (measured — third-party index)

• Event. 402index.io (Ryan Gentry, ex-Lightning Labs) — a protocol-agnostic directory of paid agent APIs across L402 (Lightning), x402 (Base/Solana), MPP (Stripe/Tempo); eight sources crawled hourly, per-endpoint health + payment verification, public REST/RSS/webhooks/OpenAPI + an MCP server. 2026-07-21 counts: 83,600 endpoints — 80,973 x402 / 1,237 L402 / 1,390 MPP; 2,329 providers. Operator’s own per-rail labels: x402 = “centralized facilitator required”; L402 = “decentralized, censorship-resistant, locally verifiable.”
• Substrate. Contest — the index spans all three rails; the datapoint is the distribution plus the operator’s property-labels.
• Structural significance. A neutral third-party index, labeling rails by their own properties, marks the censorship-resistant rail (L402/Lightning) as a minority by count but a category of one by property (no facilitator, no issuer-freeze surface). Count = adoption-so-far, not the requirement set an unfreezable agent must satisfy. Bounded: endpoint count ≠ usage/value (crawls include test/faucet endpoints); “centralized facilitator required” is a property, not a disqualifier — which is why the divergence claim rests on the asset’s freeze surface, not the rail’s convenience.
• Bears on: supports Border-Skirmishes-FA (the rail/asset contest); supports Case-FA (agents transact at scale — 83.6k monetized endpoints); supports Independence-Doctrine-FA + Why-Bitcoin-Not-A-New-Coin-FA (issuer-freeze surface as the load-bearing distinction).
• Epistemic tag. (measured) — live third-party counts + quoted operator labels (402index.io, 2026-07-21); the count≠usage and property≠disqualifier caveats are stated.
• Sources. 402index.io ecosystem overview + methodology (2026-07-21). Canonical narrative: Field-Notes-Log § 2026-07-21 (402index).

2026-07-16 — ContextVM / CEP-8: an agent-native capability market reaches for Bitcoin settlement by design (confirmed — protocol Draft)

• Event. ContextVM (CVM) carries the Model Context Protocol over Nostr — tool servers addressed by public key, relays as message bus, no DNS / TLS / API-keys / inbound-ports. CEP-8 (Standards Track, Draft; reference implementation in the TypeScript SDK since v0.4.0) adds capability pricing + a payment flow: a cap tag prices a tool call (["cap","tool:<name>","<price>","<unit>"]), settled over Lightning (BOLT11/NWC) or Cashu; fiat conversion is implementation-defined. Two lifecycles (transparent, explicit_gating). New Tools cards contextvm + cvmi; primitive added to Stack-FA §5 (S4) plus a sixth security pattern (no-inbound-surface serving, S6).
• Substrate. Bitcoin (Lightning + Cashu as the recommended, first-class rails; fiat second-class by spec).
• Structural significance. Existence proof, not adoption. An agent-native capability market designed from first principles — no legacy fiat rail to preserve, no customer base to placate — selected pubkey identity + Bitcoin, with design decisions (explicit_gating; machine-payer-first invocation correlation) made with no human in view. The thesis needs the choice, not the win; the reading survives the project failing.
• Bears on: supports Case-FA (an agent-native capability market landing on Bitcoin by design) and the Adoption-Asymmetry claim; feeds Border-Skirmishes-FA (the pubkey-relay vs. DNS-registry discovery contest — see the paired watch entry below).
• Epistemic tag. (confirmed) protocol facts per docs.contextvm.org (fetched 2026-07-16); (empirical, low) the deployed-server population is small — traction is unestablished; counts defer to the snapshot.
• Sources. docs.contextvm.org (overview); /reference/ceps/cep-8/ (Draft; tags, lifecycles, SDK v0.4.0); /how-to/payments/getting-started/; github.com/contextvm/awesome. Canonical narrative: Field-Notes-Log § 2026-07-16 (ContextVM/CEP-8).

2026-07-16 — Watch item: agent-discovery forks — CEP-6 Nostr announcements (shipped) vs. MCP Server Cards (proposed) (forward-looking — dated prediction)

• Event. Two stacks independently target “discover a server’s capabilities before connecting.” ContextVM CEP-6 ships it — capabilities + CEP-8 price published as signed Nostr announcement events (kinds 11316–11320), relay lists per CEP-17 (kind 10002). MCP’s 2026 roadmap (March 2026) proposes MCP Server Cards — the same requirement over .well-known / DNS / HTTPS; the related SEP-1649 (.well-known/mcp/server-card.json) is proposed, unshipped.
• Substrate. Contest — Bitcoin/Nostr (keypairs + relays) vs. incumbent web PKI (DNS + TLS + registry).
• Structural significance. Same requirement, two substrates, diverging in real time. A dated, checkable prediction (~12-month horizon): which discovery path ships and gets adopted, and whether “own your name (a keypair) vs. rent it (a DNS/GitHub namespace, registrar-revocable)” becomes load-bearing. Naming is the divergence argument’s second instance; money was the first.
• Honest cost. Central curation is the incumbent’s product (review, anti-impersonation, delist-once-protects-all). Pubkey addressing discards it; the sovereign answer — per-client web-of-trust (Relatr, Wotrlay, CEP-24 reviews) — is coherent but unproven at scale and pushes work onto every client. The sovereign discovery layer has the harder problem; stating so is load-bearing, not hedging.
• Bears on: feeds Border-Skirmishes-FA (the discovery contest); feeds Independence-Doctrine-FA (naming-sovereignty as the second instance of the divergence mechanism — dedicated treatment queued).
• Epistemic tag. (forward-looking) — a prediction with a check date, not a shipped-state claim.
• Sources. docs.contextvm.org (CEP-6 / CEP-17); MCP 2026 roadmap — Server Cards (modelcontextprotocol.io, March 2026); SEP-1649 (proposed, unshipped); registry counts 2026-06 (official MCP Registry ~2,000; mcp.so ~20,000). Canonical narrative: Field-Notes-Log § 2026-07-16 (watch item).

2026-07-09 — MPFlow: a production-deployed graph-RL model automates Lightning channel placement (measured — production deployment + benchmark)

• Event. A team from Amboss and Stillmark published MPFlow (arXiv 2607.08703; Rush, Davis, Antonelli, Singh, Shrader, Rossi; submitted 2026-07-09) — deep graph reinforcement learning for budgeted channel placement: given a fixed budget of k channel opens, select the peers that maximize a node’s s–t max-flow. Method: a message-passing GNN policy trained with PPO + action masking, under a hub-exclusion curriculum (top hubs removed during training to force capacity-aware placement over hub-attachment). It is the published method behind Magma AI, the channel recommender in Amboss’s Magma marketplace (Amboss / Magma). Production deployment: 4,640 channel-open decisions allocating 267.3 BTC ($16M) across 30 managed nodes. Benchmark: +8.59% ±6.20 max-flow over betweenness-centrality (62.3% paired win-rate; 0.168 vs 0.163 BTC); the larger margins reported are only against a random baseline. Stated limits: max-flow is not a validated proxy for realized payment success / yield (payment simulator = future work); uniform-balance-sampler and fixed k=5 assumptions.
• Substrate. Bitcoin / Lightning (self-custodial node operation — the policy opens real channels to the operator’s own node).
• Structural significance. Converts one core component of the “Lightning liquidity management is prohibitive for autonomous agents” objection from a structural rebuttal into empirical, production-scale evidence: budgeted channel placement is not merely delegable in principle but automated with production ML at scale. Bounded claim — the result concerns placement, not realized routing yield, which the authors explicitly leave open.
• Bears on: supports Stack-FA S2 (Lightning payment layer; active liquidity management is an operational cost, not a defect); falsifier-relevant-to Stack-FA §8 CP1 (the liquidity-and-routing-unviability counter-position — evidence against its falsifier triggering, for the placement sub-problem); supports Case-FA (an AI system already operates Lightning liquidity at production scale).
• Epistemic tag. (measured) — production figures and paired-benchmark deltas with a named source and date; the realized-yield transfer is (projected / unestablished) by the authors’ own statement.
• Sources. arXiv 2607.08703 (2026-07-09); production figures, baselines (betweenness / degree / random; GCN / GAT), and the max-flow-vs-yield caveat are the paper’s own. Surfaced via author outreach (Vikash Singh, Stillmark) 2026-07-21. Canonical narrative: Field-Notes-Log § 2026-07-09 (MPFlow).

2026-07-08 — Prominent Bitcoin voices state the “rail already exists” thesis near-verbatim; one working real-goods demo (mixed: confirmed demo + reported positioning)

• Event. In one week: (1) 2026-07-03 — TFTC (Marty Bent) posted a 58-sec demo of an AI agent buying a gift card from a real merchant over Lightning with one prompt; Amboss quote-tweeted it: “…an agent just bought a gift card from a real merchant over Lightning with one prompt. Instant final settlement, no accounts, no gas. The rail already exists.” (2) 2026-07-07 — a non-profit research-and-development lab associated with Jack Dorsey announced expanded support for open-source AI × Bitcoin software: “…the perfect economic layer for a world where AI agents seamlessly pay each other.” (3) 2026-07-08 — Amboss: “building rails for sub-second agentic commerce, letting AI agents settle final, irreversible payments globally over Bitcoin’s Lightning. No chargebacks. No middleman.”
• Substrate. Bitcoin (Lightning; self-custodial, no issuer in the path). The 7/3 item is an L402-class real-goods demo; the 7/7 and 7/8 items are positioning statements by Bitcoin-Lightning-aligned parties.
• Structural significance. Third-party, near-verbatim articulation of the Case thesis (the rail already exists vs. the design-a-new-protocol camp), from a warm-relationship partner (Amboss ×2) and a major Bitcoin-philanthropic lab, plus one reproducible existence proof (agent → real-world good over Lightning, one prompt). Signals thesis diffusion, not a new capability; the load-bearing datapoint is the demo, the quotes are corroboration.
• Bears on: supports Case-FA (the deployed-rail claim; an agent buys a real-world good on the Bitcoin stack) and Constraints 1–4; supports Border-Skirmishes-FA BS-series — these are the “rail is already Bitcoin-native” pole set against the MPP/x402 pole logged 2026-06-30 (the contest is the asset/trust model, not the rail).
• Epistemic tag. (mixed) — the 7/3 agent-buys-gift-card demo (reported: a 58-sec clip, not independently reproduced here); the 7/7 and 7/8 statements (reported positioning, not shipped infrastructure). The 7/7 lab’s exact legal name is not stated in the source post.
• Sources. x.com/i/status/2073163539955712495 (TFTC, 7/3); /2073176645272252655 (Amboss QT, 7/3); /2074600071383966197 (DocumentingBTC — Dorsey lab, 7/7); /2074859033077415957 (Amboss, 7/8). Captures via x-fetch (2026-07-13). Canonical narrative: Field-Notes-Log § 2026-07-08.

2026-06-30 — Real-world-goods agent commerce ships on L402 (Unhuman / Money Dev Kit) (confirmed)

• Event. Live agent-only storefronts — Unhuman Coffee (roasted-to-order beans), Unhuman Domains, Unhuman Store (hub) — sell real-world goods over Bitcoin via L402: GET /api/catalog → POST /api/order → HTTP 402 (Bolt11 + macaroon) → pay → replay with Authorization: L402 <macaroon>:<preimage>. No account, no card, no KYC. Live order observed 2026-06-30 (~41,770 sats incl. shipping on a $24/12oz item). Built on Money Dev Kit (self-custodial LDK-based Lightning SDK by Nick Slaney; ships an agent-wallet CLI an agent drives with no keys/accounts).
• Substrate. Bitcoin (Lightning + L402; self-custodial on both sides; no issuer in the path).
• Structural significance. Extends the deployed “agent buys a service” pattern from digital goods (inference / compute / liquidity) to physical real-world goods settled end-to-end in Bitcoin by software — the consume side reaching tangible commerce on the censorship-resistant rail, reproducible via a public SDK rather than a one-off demo.
• Bears on: supports Services-FA SV-series (consume side) and the services-built-on-services dynamic (Unhuman runs on MDK); supports Case-FA (an agent transacts on the Bitcoin stack) and Constraints 1–4 (no-KYC, censorship-resistant, sub-cent-capable, machine-tempo).
• Epistemic tag. (confirmed) — live endpoints + verified L402 flow; author attribution (reported) via npm maintainer + the author’s public post.
• Sources. unhuman.coffee / .domains / .store (live, 2026-06-30); moneydevkit.com + docs.moneydevkit.com/howitworks; @moneydevkit/agent-wallet (npm, Apache-2.0); author post (nickslaney, status 2027460567946940477). Canonical narrative: Field-Notes-Log § 2026-06-30. New cards: unhuman, money-dev-kit.

2026-06-30 — Stripe/Tempo Machine Payments Protocol + Lightspark Lightning method: Bitcoin as an optional rail (confirmed)

• Event. MPP (Machine Payments Protocol), co-authored by Stripe + Tempo (Stripe/Paradigm stablecoin L1), launched 2026-03-18 — HTTP-402 Challenge/Credential/Receipt, intents Charge/Session/Subscription (Session = pre-authorized streaming micropayments). Payment-method agnostic; default rails Tempo stablecoins + Stripe/Visa cards (+ EVM-USDC x402-compatible, Solana, Stellar). Lightspark extended it to Bitcoin Lightning via Spark (@buildonspark/lightning-mpp-sdk; BOLT11 + HTLC + preimage). Amboss leans on MPP as an agentic standard “because it works” (call-sourced). NB: “MPP” here ≠ Lightning Multi-Path Payments.
• Substrate. Competing / multi-rail — incumbent default (regulated stablecoins + card networks; censorship-requiring), with Bitcoin/Lightning a first-class but optional method.
• Structural significance. The clearest single artifact of the border-zone read: one 402 standard whose default trust model is processor/issuer-gated, exposing Lightning as a guest rail. Confirms the incumbents built parallel agent-payment rails preserving their freeze/KYC property bundle rather than adopting Bitcoin’s — the rail can be Lightning while the asset and trust model stay incumbent. Censorship-surface gradient: L402 < x402/USDC < MPP.
• Bears on: supports Border-Skirmishes-FA BS-series (incumbents build parallel rails; the contest is on the asset/trust model, not the rail — kin to the Lightspark Grid proof case) and Independence-Doctrine-FA (the asset/trust layer can’t shed the freeze property without ceasing to be the incumbent stack); falsifier-relevant-to the Border-Skirmishes-FA falsification clause (agent stacks standardizing on stablecoins by default) — MPP is that standardization; the unresolved variable is the censorship-resistant share of agent commerce.
• Epistemic tag. (confirmed) — Stripe/Tempo launch + the mpp.dev Lightning method documented; the Lightspark/Spark link rests on mpp.dev (no first-party Lightspark MPP press release located); Amboss’s lean (reported), call-sourced.
• Sources. stripe.com/blog/machine-payments-protocol; mpp.dev + /payment-methods/lightning; lightspark.com/news/spark/introducing-spark; getalby.com/blog/agentic-commerce-a-guide-to-l402-x402-and-mpp; thedefiant.io (Tempo mainnet + MPP, 2026-03-18). Canonical narrative: Field-Notes-Log § 2026-06-30.

2026-06-28 — PPQ’s encrypted inference becomes substrate: a downstream product builds no-KYC private AI on PayPerQ’s rails (confirmed)

• Event. getbased (getbasedhealth, Nostr) launched encrypted AI inference powered by PayPerQ — GLM-5.2, “no email, no KYC, no fiat, no subscription — pay per query, prompts encrypted” — riding PPQ’s TEE private-inference tier (NVIDIA confidential-computing enclaves + browser-side end-to-end encryption; exposed as private/glm-5-2, 384K context; “PayPerQ never sees your prompts”).
• Substrate. Bitcoin (PPQ pay-per-query over Lightning / L402; private TEE tier on top).
• Structural significance. A featured consume-side venue (PPQ) is being reused as infrastructure by a second product — destination → substrate. The agent economy as services-built-on-services, shown empirically: no-KYC, Bitcoin-paid, end-to-end-encrypted inference composed by a downstream builder with no account, no fiat, no human-in-the-loop. getbased is a human-facing front-end, not an agent-drivable venue → no directory entry; the load-bearing datapoint is PPQ-as-substrate plus the private/TEE privacy axis.
• Bears on: supports Services-FA SV1 (consume side) + the services-compound-on-services dynamic; supports Case-FA (an agent funds its own private inference on the Bitcoin stack); strengthens the PPQ privacy profile (TEE tier; card refreshed 2026-06-28).
• Epistemic tag. (confirmed) — provider TEE tier documented; the downstream launch (reported) from the getbased Nostr announcement.
• Sources. PayPerQ — Introducing Private AI Models; getbased Nostr note (getbasedhealth, 2026-06-22). Canonical narrative: Field-Notes-Log § 2026-06-28.

2026-06-28 — Autonomous Lightning liquidity management is now an assemblable self-custody stack (confirmed)

• Event. Amboss (Jesse Shrader, on Stefan Livera, June 2026) detailed the current shape: ThunderHub (open-source LND manager) hosts RailsX (self-custody Taproot-Assets DEX — BTC↔USDT/USDC via circular payments, peer-selected, 8–15 bips, no KYC) and Magma (one-click inbound-liquidity leasing); Rails is the managed LP-yield side (self-custody, limited-macaroon, ~1–1.5% BTC APY; institutional “Lightning Earn” via BitGo); Loop handles own-balance L1↔LN.
• Substrate. Bitcoin (Lightning rails; Taproot Assets for the stable leg).
• Structural significance. Resolves the long-standing operational gap (State of Play — For Agents §A.4: Lightning liquidity management at scale) for the agent buy-side: an agent can buy inbound capacity, earn on idle BTC, trade into a stable unit of account, and rebalance — all self-custody, no-KYC, macaroon-scoped (manage ≠ withdraw). Caveat: RailsX stablecoins are wrapped/issuer-backed (Speed Wallet, 1:1) over Taproot Assets — the rail is self-custodial; the asset retains its issuer freeze surface (Constraint 2 satisfied at the rail, not the asset).
• Bears on: supports Stack-FA §2 (liquidity management — now deployed/composable); supports Services-FA SV-series (the consume/offer infra agents need); consistent with the asset-vs-rail distinction (Taproot Assets cards). Dual-track input to the Hermes-Worker Banker design.
• Epistemic tag. (confirmed) — products deployed; yield figures (reported), self-asserted by Amboss.
• Sources. Stefan Livera × Jesse Shrader (Amboss), RailsX episode (youtu.be/VO91uTYxTQs); magma.amboss.tech / amboss.space; ThunderHub github.com/apotdevin/thunderhub. Canonical narrative: Field-Notes-Log § 2026-06-28.

2026-06-16 — Regulated stablecoins carry a statutorily-mandated kill switch (the “off switch” property is not CBDC-exclusive) (confirmed)

• Event. Renewed public circulation of the CBDC “off switch” critique (Cato Institute, “When Money Has an Off Switch, So Does Your Freedom”) prompts the structural clarification: the GENIUS Act (2025 US payment-stablecoin statute) requires every permitted payment stablecoin issuer (PPSI) to maintain the technical capability to seize, freeze, burn, or prevent the transfer of its tokens on a lawful order from a federal agency or court, as a licensing condition (“reasonable particularity” standard). FinCEN/OFAC implementing rule proposed April 10 2026 (Fed. Reg. 2026-06963); comment period closed June 9 2026. Exercised, not latent: Circle ~$8.2M (Tornado Cash, 2022) + 16 business wallets (sealed civil suit, March 2026); Tether >$1B cumulative.
• Substrate. Competing (regulated-stablecoin).
• Structural significance. The freeze/seize property that fails Constraint 2 (censorship-resistance) for the parallel-economy/agent subset is not a CBDC-specific defect — it is statutory law for regulated stablecoins. A CBDC is a state liability switched off directly; a GENIUS-compliant stablecoin is a private liability the state compels the issuer to switch off. Same property, different control path. This makes Constraint 2’s failure on the competing substrate black-letter law, not mere issuer discretion — and since removing the capability removes the license, the property is non-sheddable (Doctrine D1’s identity-defining mechanism, now with a statutory hook). Agent-specific sharpening: an autonomous agent settling with no human-in-the-loop has no recourse against a mid-workflow freeze.
• Bears on: supports Case-FA C2 + §8.1 CP1; supports Doctrine-FA D1 (non-sheddable freeze property — statutory mechanism) and tests Doctrine-FA P6 (regulatory machinery routes autonomy-requiring commerce toward the censorship-resistant substrate); supports Border-Skirmishes-FA BS-series (asset-side freeze fails Constraint 2 regardless of rail). Generalizes the regulatory-pincer point (see Moonshots entry, Field Notes 2026-06-11) beyond a single issuer.
• Epistemic tag. (confirmed) — statute enacted; implementing rule proposed (not yet final — flag for refresh); freeze events (measured).
• Sources. Skadden; Gibson Dunn; Fed. Reg. 2026-06963; WilmerHale; CNN.

2026-06-11 — Moonshots ep. 264: the substrate question goes public — Coinbase’s CEO concedes the premise, contests the layer (confirmed)

• Event. On Peter Diamandis’s Moonshots podcast ep. 264 (published 2026-06-11), Coinbase CEO Brian Armstrong stated three things on the record: (1) agents cannot pass KYC — “an agent doesn’t have a piece of paper issued by the government with your photo on it” (~23:25) — so Coinbase built self-custodial agent wallets that skip account-opening; (2) the agent economy is live and compounding — he corrected the episode’s stale 3.1M figure upward to “about 100 million transactions now, maybe 50 million of value” (reported, ~22:13); (3) on crypto’s role once “most of our economy consists of AI agents trading”: “Bitcoin will be the new gold standard and then the payments will be happening on chain… that’s the financial system that the AI agents would end up using” (~1:21). The contested layer: Armstrong expects “stablecoin payments will probably be the default layer for the agentic economy” (projected, ~7:34), on USDC over Base. Panelist Salim Ismail stated the two-tier model unprompted: “clearly Bitcoin becomes the digital collateral for an AI native economy… they’re not going to be using checking accounts in JP Morgan” (~5:35).
• Substrate. Cross-cutting (competing-substrate advocacy with an explicit Bitcoin-reserve concession).
• Structural significance. The Case’s premise (agents as economic actors at scale) is conceded on-air by the incumbent side’s strongest advocate; the disagreement narrows to the payment layer. The stablecoin-default case is real on its merits — genuine x402/Base volume, a working stack, sub-second/sub-cent/global settlement, dollar denomination — but every property Armstrong listed Lightning settlement also has; the one thing USDC has that sats do not is an issuer, i.e. a control point with an exercised freeze record (Circle ~$8.2M Tornado Cash 2022; Tether >$1B). Armstrong credited the GENIUS Act’s “regulatory clarity” (~4:21) for making stablecoins “the new meta” — the default layer exists because regulation blessed it, which is the same lever that can reshape it. Incentive disclosure (read from filings, not motive): under its Circle arrangement Coinbase keeps 100% of on-platform USDC reserve income and splits off-platform 50:50 — ~$1.35B FY2025, +~48% YoY (measured) — plus ~$75M 2025 Base sequencer revenue; x402 originated at Coinbase. The stablecoin-default forecast is made by the party that collects the toll if it holds.
• The regulatory-pincer argument (load-bearing). Panelist Alexander Wissner-Gross asked (~24:43) what happens to stablecoin agent wallets if regulators let agents open conventional fiat bank accounts. Both branches degrade the issuer-mediated layer. Tighten: enforcement lands at the issuer, where the machinery is already installed — GENIUS requires every permitted issuer to seize/freeze/burn on lawful order as a licensing condition; the FinCEN/OFAC implementing rule (comment period closed 2026-06-09) extends sanctions-compliance into issuer infrastructure; FATF’s March 2026 guidance recommends secondary-market (A2A-inclusive) monitoring; Circle froze 16 business wallets under a sealed civil suit (March 2026); “Know Your Agent” proposals are circulating. Loosen: the only remaining moat Armstrong claimed was that legacy rails are slow (“COBOL servers,” ~28:13) — a moat tokenized deposits actively drain. The property that survives both branches — settlement that does not ask permission — is the one stablecoins surrendered to exist and Bitcoin never had to negotiate for. The same episode treated US golden-share/quasi-nationalization of civilization-scale AI infrastructure as “probably inevitable” (~35:18); applied one story over, an agent-payment layer on an issuer’s ledger plus a single company’s L2 is exactly that kind of chokepoint — and a substrate with no issuer is the only one that argument does not reach.
• Bears on: supports Case-FA C1 (agents-as-economic-actors premise — now conceded by the incumbent advocate) and C2 (censorship-resistance — the issuer is the freeze control point); tests Doctrine-FA P1 (which substrate the deployed economy selects — the contest is now explicit and public) and P6 (regulatory pressure as a standing liability for the intermediated layer — Wissner-Gross’s both-branches question is the live instance); supports Border-Skirmishes-FA BS-series (the competing-substrate stack and its strongest public case, stated by its own builder); falsifier-relevant-to Case-FA §8.2 (a stablecoin-default outcome that retains autonomy at scale would weaken the divergence claim — track x402 volume). Armstrong’s “on-chain FICO” agent-reputation-graph proposal (~29:40) is a platform-layer surveillance primitive worth watching alongside the freeze record.
• Epistemic tag. (confirmed) — dated public statements by named parties; transaction-volume figures (reported) (speaker estimate, self-corrected on-air); Coinbase financials (measured) (shareholder letters); stablecoin-default expectation (projected).
• Sources. Moonshots ep. 264 (Peter Diamandis, 2026-06-11; with Brian Armstrong, Dave Blundin, Salim Ismail, Alexander Wissner-Gross; timestamps approximate, from the episode auto-transcript). Coinbase reserve-income + Base sequencer figures: Coinbase quarterly shareholder letters (FY2025). GENIUS seize/freeze/burn + FinCEN/OFAC rule: see the 2026-06-16 record above. Canonical narrative form: Field-Notes-Log § 2026-06-11.

2026-05-07 — AWS Bedrock AgentCore Payments launches with Coinbase x402 + Stripe Privy (confirmed)

• Event. AWS announced Amazon Bedrock AgentCore Payments — infrastructure enabling autonomous agents to make real-time purchases using stablecoins. Build: Coinbase (x402 protocol on HTTP 402; Coinbase Agentic Wallets; compliance infrastructure) + Stripe (payment infrastructure and wallet integrations via Privy, acquired 2025). Settlement: USDC on Base, ~200ms confirmation, sub-cent per transaction. First version targets micropayments (APIs, data feeds, paywalled content). Enterprise customers at launch: Thomson Reuters, Warner Bros. Discovery, Cox Automotive, PGA TOUR.
• Structural significance. First Tier-1-enterprise production deployment of the integration scenario for agent payments. The customers are Fortune-500 enterprises operating in the regulated USD-denominated economy, not crypto-native early adopters. The stack serves the integration-scenario subset (USD-denominated, regulated-counterparty, issuer-counterparty-risk-acceptable) and the structural prediction is that it does so without absorbing the parallel-economy subset (agent activity requiring all four conjunctive constraints). The L402-vs-x402 naming convergence is the protocol-level expression of the substrate divergence: same status code, different settlement currencies, different trust models, two competing production stacks.
• Bears on: supports Border-Skirmishes-FA BS-series (use-case bifurcation) and Doctrine-FA §8.1 CP2 (integration scenario operationally deployed); tests Doctrine-FA P1 (which subset proves larger over 2–5 years) and Doctrine-FA P6 (the intermediated stack’s freeze/KYC/sanctions surfaces under agent regulation); supports Case-FA §8.1 CP1 (regulated stablecoins as substrate — operational confirmation).
• Sources. AWS: Agents that transact — Amazon Bedrock AgentCore Payments; The Block; CoinDesk; CryptoTimes. Structural treatment: Research/Border-Zone-Existing-Bridges.md §8; Research/Border-Zone-Competing-Substrate-Analysis.md CP1.

2026-05 — Routstr: Bitcoin-powered AI-inference marketplace (Cashu + Lightning + Nostr) (confirmed)

• Event. Routstr — open-source protocol + reference implementation (routstr-core, GPL-3.0; v0.4.3 May 2026) — runs a payment-gated reverse proxy in front of OpenAI-compatible LLM APIs plus a Nostr marketplace for provider discovery. Payment in Cashu ecash (token-as-API-key); provider earnings over Lightning; discovery/pricing as Nostr events. No accounts, KYC, or cards. HRF Top-15 Freedom Tech Project of 2025; supported under HRF “AI for Individual Rights.”
• Structural significance. The cleanest deployed instance of an agent buying a service on the Bitcoin payment stack (Cashu + Lightning) rather than the card/stablecoin stack — Cashu-token-as-API-key is a concrete bearer-credential answer to “how does an autonomous agent pay without a human-held account.” A Cashu-track instance: standardizes on Cashu (not Fedimint), bearer-token payment (not L402/NWC) — demonstrates one branch of the payment-tech stack, not all of it. Gap/collaboration opening: no llms.txt/agent-first surface.
• Bears on: supports Stack-FA S3 (Cashu bearer-ecash layer) + S4 (agent-integration: bearer credential) + S5 (deployed architecture); supports Doctrine-FA D3 (a deployed divergent instance, distinct from the incumbent stacks).
• Sources. Routstr; docs; GitHub: Routstr/routstr-core; HRF Top-15 Freedom Tech 2025.

2026-05 — Competing-substrate landscape broadens beyond AgentCore (AP2, Circle Nanopayments, Skyfire, x402 Foundation) (confirmed — landscape record)

• Event. (digests several 2025–26 developments.) Google AP2 (Agent Payments Protocol), launched September 2025: a 60+-organization consortium (Mastercard, American Express, PayPal, Coinbase, Adyen, Revolut, Worldpay, Salesforce, Intuit) with an A2A x402 extension built alongside Coinbase, the Ethereum Foundation, and MetaMask. x402 contributed to a dedicated x402 Foundation under the Linux Foundation (April 2026); 119M+ tx on Base. Circle Nanopayments (mainnet May 2026): gas-free USDC micropayments from $0.000001, x402-v2-compatible. Skyfire (“Agent Trust Stack,” backed by a16z CSX, Coinbase Ventures, Brevan Howard): Visa/Mastercard/Discover/USDC.
• Structural significance. The competing stack is plural and standardizing — at the governance layer (x402 Foundation; AP2 consortium), not just per-product. All standardize on stablecoins, card networks, and Ethereum/Solana, not Bitcoin (MetaMask, on the A2A x402 extension: “Ethereum will be the backbone”). This confirms Doctrine-FA D1’s mutual-exclusion mechanism: incumbents build a parallel stack preserving the issuer-controlled, freezable property bundle their licensing requires. (empirical concession — Constraint 3) Circle Nanopayments’ gas-free design narrows the sub-cent micropayment-economics gap on the stablecoin payments leg; it does not touch Constraints 1–2 (issuer freeze surface persists). (reality-check) CoinDesk (March 2026) noted x402 transaction demand remains thin relative to rail capacity — the substrate question is unsettled, not decided.
• Bears on: supports Doctrine-FA D1 + P1/P2/P5 (incumbents consolidate a competing-but-incumbent stack; substrate-selection precedes scale); supports Border-Skirmishes-FA BS-series roster update + use-case bifurcation; tests Case-FA §8.1 CP1 (regulated-stablecoin substrate — now plural deployment evidence).
• Sources. Google Cloud: Announcing AP2; GitHub: google-a2a/a2a-x402; x402.org; Coinbase x402 docs; Circle Nanopayments; Skyfire.

2026-04 — Lightspark Grid adds AI-agent bounded delegation (hybrid Lightning-rail stack) (confirmed)

• Event. Lightspark (led by ex-PayPal president David Marcus) added AI-agent bounded delegation to its Grid Global Accounts: agents get funded, scoped, auditable “pockets” with wallet-level spending limits, approved payees, per-transaction/daily/monthly caps, approval thresholds, and instant revocation. Grid settles over Lightning among multiple rails but is built on “Bitcoin and stablecoins” — branded USD/stablecoin accounts, Visa debit cards, instant Bitcoin conversion — and Lightspark has published an Agent Payments Protocol (AP2) vision aligning it with Google’s consortium.
• Structural significance. A Lightning-rail multi-rail product, not a Bitcoin-substrate one: the rail is Bitcoin’s; the asset (issuer-controlled USD/stablecoin) and the trust model (wallet provider + issuer + card network) are the incumbent’s. The instructive case in the competing roster because it is the closest to the substrate — a Lightning-native, Bitcoin-credentialed team still chose dollar/stablecoin denomination, card-network reach, and the AP2 stack. Confirms the divergence is about the asset and the trust model, not the rail. The bounded-delegation primitive (funded, scoped, revocable agent pockets) is a useful treasury-control pattern independent of the asset question.
• Bears on: supports Border-Skirmishes-FA BS-series (Lightning-rails-for-stablecoins ≠ Lightning-substrate; use-case bifurcation — a hybrid actor serving the incumbent-economy subset over Lightning rails); supports Doctrine-FA D1 (the incumbent property bundle is preserved even by a Lightning-native builder); tests Doctrine-FA P6 (the intermediary layers — issuer, card network, wallet provider — are the freeze/KYC surfaces P6 predicts become a standing liability for agents under regulatory attention).
• Sources. Lightspark — Agent Payments Protocol (AP2); ITBrief: Lightspark adds AI agent controls to Grid; Bitcoin Magazine: Grid Global Accounts.

2026-03-21 — USDT live on Lightning via Taproot Assets (confirmed)

• Event. Tether CEO Paolo Ardoino confirmed USDT is live on Bitcoin’s Lightning Network via Lightning Labs’ Taproot Assets protocol, completing a 14-month integration begun at the Plan B Forum (El Salvador, January 30 2025). Bitfinex to issue USDT on Lightning per Tether. Follows the June 2025 Taproot Assets v0.6 release (“Bitcoin’s Decentralized FX Network”).
• Structural significance. A Lightning-rails bridge for the stablecoin, not a Lightning-substrate bridge. The issuer (Tether) retains freeze capability on its issuance regardless of which rail the asset moves over; Constraint 2 (censorship-resistance) still fails for the asset side even though rail-side properties (sub-cent, machine-tempo) are excellent. USDT-on-Lightning serves integration-scenario use cases and does not make stablecoins suitable as the parallel-economy substrate. The bridge changes the rail, not the asset.
• Bears on: supports Border-Skirmishes-FA BS-series (the exact rails-vs-substrate claim: rail-side passes Constraints 1/3/4, asset-side fails Constraint 2 by design); supports Case-FA §4 (substrate evaluation — regulated stablecoins row).
• Sources. Tether: USDt to Bitcoin’s Lightning Network; BTC.network fee-market analysis; Speed Wallet; Lightning Labs Taproot Assets v0.6. Operational treatment: Research/Border-Zone-Existing-Bridges.md §4.

2026-03 — Bitcoin Policy Institute publishes AI Models Overwhelmingly Prefer Bitcoin and Digital-Native Money Over Traditional Fiat (measured)

• Event. BPI published the study in March 2026. Method: 9,072 scenarios across 36 frontier language models, neutral scenario design (no leading prompts); each scenario asked the model to choose a preferred monetary instrument from a candidate set. Headline: Bitcoin the top overall monetary preference at 48.3% of responses, and dominant on the store-of-value dimension at 79.1%; over 90% favored digitally-native money over fiat (stablecoins led payment-preference scenarios at 53.2%). Per-provider the result was uneven — one provider’s models chose Bitcoin in 68% of responses, another’s in 26% — and the strongest single-model consensus anywhere in the study was 91.3%; wide spread, one-directional.
• Structural significance. Load-bearing empirical anchor for Case-FA C3 (substrate-preference signal) and Doctrine-FA P1 (substrate-selection-precedes-scale). Establishes that frontier models reasoning about substrate selection under neutral choice converge substantially toward Bitcoin without ideological prompting — consistent with the four-conjunctive-constraints argument. The study measures preference under inference, not deployed-flow dominance; convergent independent replication would strengthen the signal, contrary results would weaken it. No replication published as of May 2026.
• Bears on: supports / tests Case-FA C3 and §6; Doctrine-FA P1 and §7.2; falsifier-relevant-to Case-FA §8.2 (replication showing preference shift). KB origin: The AI-agent monetary substrate case § The empirical signal.
• Sources. Bitcoin Policy Institute — Study: AI Models Overwhelmingly Prefer Bitcoin and Digital-Native Money Over Traditional Fiat (March 3 2026). (BPI ai models prefer bitcoin research)

2026-02-11 — Lightning Labs releases lightning-agent-tools (confirmed)

• Event. Lightning Labs open-sourced lightning-agent-tools — a production AI-agent toolkit on the Bitcoin substrate. Seven composable skills: (1) running a Lightning node programmatically; (2) remote-signer key isolation; (3) baking scoped macaroons in five preset roles; (4) paying L402-gated APIs via lnget; (5) hosting paid endpoints via Aperture; (6) querying node state through MCP; (7) orchestrating end-to-end buyer/seller workflows.
• Structural significance. First Tier-1 production deployment of the Bitcoin-substrate agent-payment stack — the operational counterpart to the Thesis. Activates L402 (specified 2020) from “interesting protocol” to “production agent-commerce stack with deployed tooling.” Shipped February 2026; AWS AgentCore shipped May 2026 — the two competing-substrate production stacks emerged within 90 days of each other on directly comparable surfaces, making the Independence Doctrine’s prediction testable in real time.
• Bears on: supports Stack-FA S4 (integration primitives — L402, MCP, scoped credentials), S5 (deployed wallet architectures — the canonical reference implementation), S6 (security model — remote-signer isolation, scoped macaroons), and Case-FA C4 + §9; tests Doctrine-FA P1.
• Sources. Lightning Labs: The Agents Are Here and They Want to Transact (Feb 11 2026); Bitcoin Magazine, The Block, BitcoinEthereumNews coverage. Capability enumeration: Stack-FA §6; Case-FA §9.

Maintenance and refresh protocol

Append cadence. As developments warrant. Single dated records for specific events; multi-week composite records acceptable for slower-moving developments. Each record names the event, its structural significance, the canonical claim-IDs it bears on, and primary sources. The structured current-state snapshot this log feeds lives on the companion State of Play — For Agents (refreshed at least quarterly).

Defer-pattern (locked 2026-05-26). The canonical surfaces and their FA twins link out to Field Notes for ongoing empirical tracking rather than carrying it inline; routine empirical updates defer here.

References and provenance

Primary canonical source. Field Notes — The Log — the canonical log surface; source for all records here. This document is its For-Agents twin. Companion snapshot: State of Play — For Agents.

Canonical claim-ID series this surface cross-references. Case-FA (C1–C6), Independence-Doctrine-FA (D1–D6, P1–P6), Treasury-FA (M-series), Border-Skirmishes-FA (BS-series), Stack-FA (S1–S8 + §8 CP1–CP4).

Human-track canonical surfaces. Field Notes — The Log, Case, Independence-Doctrine, Border-Skirmishes, Stack.

KB origin. The case for investing in Bitcoin § AI-agent monetary substrate case; The AI-agent monetary substrate case.