A short building-in-public recap of the week.

Sunday recap is the only slot where the calendar trumps the topic: we don't pick what to write about, we just look back at what landed and what it taught us. This week the headline wasn't ours, the most interesting signal was the smallest number on the dashboard, and a quiet infrastructure win matters more than either.

The headline we didn't write

Portal to Bitcoin raised $25M for an "atomic OTC desk" — HTLC-based, non-custodial, cross-chain. That's the same primitive Hashlock is built on, scoped to a single asset and an institutional human desk, and now priced by professional investors. We treat that as category validation rather than competition. The thesis we've been making in long-form for months — that atomic cross-chain settlement is real infrastructure, not a research curiosity — just got marked to market by people whose job is to put a number on this. The boundary between us is clean: their wedge is BTC-first institutional OTC; our wedge is multi-chain agent-native settlement exposed as an MCP server. Both can be right.

The corollary, which we wrote about Friday, is that "atomic" is now the word in the category most likely to get stretched. When a category gets funded, the language gets borrowed. We made the case that "atomic" has a specific operational meaning — both legs settle or neither does, with no window where one party has delivered and the other hasn't — and that custodial "settles atomically from your perspective" claims should be read as a different product, not a synonym.

Five posts, one primitive

Five posts went out, deliberately framed as one primitive seen from five angles:

A working paper on reactive intent markets, framing sealed-bid RFQ as the matching layer above HTLC settlement. An execution rewards and tiered KYC post on why agent payouts need on-chain, verifiable counterparty tiers rather than custodial checkboxes. A multi-leg atomic vs sequential routing comparison — what your DEX aggregator can't guarantee when a route crosses chains. The Friday "what 'atomic' actually has to mean" essay, written off the back of the Portal raise. And the Saturday 6-question FAQ translating the essay into the questions a reader would actually ask next.

The throughline: settlement is a property of the protocol, not a promise of the operator. Five mechanisms, one primitive.

The smallest number on the dashboard

Mid-week, our analytics caught the first measurable end-to-end platform attribution: one LinkedIn click on the May 13 Execution Rewards post traced cleanly into a session. One. We spent six weeks building UTM pipelines, fixing OAuth on the GSC/GA4 probes, and reasoning about funnels that we couldn't actually see. The first real number is one.

Why does that matter? Because before the click, every "post landed" claim was self-report. After the click, we can tell the difference between a post that ran and a post that worked. That's the input the strategist task has been missing: which platform-format-topic combinations actually move someone to the site, and which just feel like they should. We will be very careful not to over-fit to n=1. But the rails are now there.

Two related signals on the same day: GA4 started reporting agent_error and tool_call events — meaning the MCP/agent side is finally instrumented, and we can start telling agent-traffic from human-traffic. And linkedin/social appeared as a source for the first time, separate from t.co. The reason this matters operationally is that the next round of content decisions can be made on data rather than vibes.

What we shipped on infrastructure

Quieter wins, but they compound:

• Substack opened. hashlockmarkets.substack.com went live Friday with the "what 'atomic' actually has to mean" essay as the inaugural post (the slug hashlock was already taken by an unrelated user). Free tier, no paywall, weekly cadence on Fridays.
• DefiLlama listing is now day 6 — the protocol page exists, though our PR to populate the url field is still open against defillama-server.
• /methodology is in the sitemap and indexed. Small SEO win — three days ago it was discovered-not-indexed.

What we still don't know

A few questions we don't have answers for yet, in case any of them are interesting to you:

• At what trade size does optimistic-bridge speed beat HTLC atomicity for an agent? We have an opinion ("never, when the failure mode is mid-trade") but the honest answer depends on the agent's time-discount and the size of the route. A real number would help us pitch this better.
• Which X publish time generates t.co traffic that doesn't bounce in under five seconds? We've held 17:00 LOCAL for three weeks and the only sample we have is n=1 at 0% engagement. Two more clean samples and we'll know whether to move.
• How does an agent measure counterparty trust in real time? Execution rewards and tiered KYC is a partial answer; we don't think it's the whole one.

Where we are in the build

For anyone new this week: Hashlock is the atomic settlement layer for the agent economy. Sealed-bid RFQ plus HTLC atomic settlement, fused into one operation, exposed as an MCP server with six tools an agent calls directly. No bridges, no custodians, no relayer with custody.

Chain status, current and honest: ETH live end-to-end; SUI contracts deployed and CLI-tested with the mainnet gateway still being wired; BTC signet-validated with mainnet wiring pending. Roadmap chains (Base, Arbitrum, Solana, TON) are scoped but not deployed. We keep saying the same things in the same order because chain claims are exactly the place we don't want to drift.

Next week's calendar: Monday goes back to the technical-deep slot (something on HTLC mechanics or the BTC Collateral architecture), Wednesday is the long-overdue Verified Counterparty Directory primitive, and Friday's Substack will pick up the trade-size threshold question if anyone replies with a number.

Site: hashlock.markets. Source on GitHub: github.com/Hashlock-Tech/hashlock-mcp. Academic write-up of the primitive on SSRN for anyone who wants the formal version.

If you read all the way down here: at what trade size does optimistic-bridge speed beat HTLC atomicity for your use case? Reply with a number if you have one — we're collecting answers for next Friday's piece.