A marketplace built on human-made goods named agent-readability as a reason for its discovery spend — and attached no number to it (Etsy, 5 Aug 2026)

Why it matters (spine). RDKG's wager is that as intelligence collapses toward free, verified and hard-to-synthesise ground truth becomes the scarce input — and that machines will be the customer. The 31 July sweep produced a sharp correction to that wager: Reddit, the company whose entire asset is human conversation, monetises it through advertising rather than licensing to AI (reddit-q2-2026-human-corpus-ad-premium). This card is a small counterweight, and it is honest about how small it is. Etsy — a marketplace whose own description of itself says "In a time of increasing automation, it's our mission to keep human connection at the heart of commerce" — is restructuring its catalogue so that machines can read it.

Where this sits against what the garden already holds. It is not the first evidence in the cabinet for agentic commerce, and it should not be read as such. The cabinet already carries the rails — x402-linux-foundation, x402-foundation-operational-launch, ap2-agent-payments-rails, visa-stablecoin-platform-open-usd — and one demand-side projection from a regulator: autorite-ai-agents-concentration-opinion (17 July 2026, verified), in which France's competition authority puts agent-driven e-commerce traffic at "less than 5%" today and "between 20% and 25%" by 2030. What is new here is the supply side, from an operator: a marketplace saying, in its own investor letter, that it is reshaping how its inventory is structured partly because machines will be reading it. Rails, projection and posture are three different kinds of evidence, and none of them is a measurement of agent-driven commerce actually happening.

What the letter says. On the purpose of the discovery spend: "Our foundational investments include work on our inventory signals, content feeds, and product listing ad campaign structures. This work is critical to ensure that the long-tail of Etsy inventory is highly visible and consumable across all external platforms, including as agentic experiences become more important over time." And on the buyer-facing side: "Moving to other Discovery investments, we continue to test how and where conversational 'agentic' shopping fits as a durable, distinctly Etsy experience." Two things are worth naming. The first quotation is about supply-side legibility — making inventory machine-consumable off-platform — which is structurally the same problem RDKG's machine lane solves for verified signals. The second is a test, in Etsy's own word, not a shipped product.

Why this is verified with low confidence, rather than signal. Tier is provenance; confidence is whether the claim holds (L6, L52). The provenance here is sound: a first-party investor document, dated inside the window, read from the main thread, with a falsifier stated below — so the tier is verified for the claim the card actually makes, which is "Etsy states that agent-readability is one reason for its discovery investments." The confidence is low, and that belongs in prose rather than in the tier. The document does not support "agents are driving commerce at Etsy," because the letter contains no agent-attributed traffic figure, no agent-attributed GMS, no spend allocation and no timeline.

This was drafted at signal and corrected during the run's self-audit, and the correction is worth recording: the drafted rationale — that the claim was weaker than the document — is a confidence judgement wearing a tier, which is precisely the split L52 exists to enforce. It also mis-cited the Replimune card (replimune-rp1-ctgt-advisory-committee, dated 30 July 2026) as precedent. That card was held at signal because its 8-K described a future, scheduled event and so could not evidence an outcome at all; that is a provenance gap. Etsy's letter fully evidences what this card claims. The two are not the same call.

One exclusion, deliberate. A share-of-traffic figure for agentic experiences circulated during this sweep; its only sources were an earnings call and a scout-relayed snippet, neither of which I opened, and a direct search of the letter found no such figure. Neither the number nor an approximation of it appears anywhere on this card, which is why it is described here without being stated.

Garden-code note (L41, flagged not resolved). This card's domains include machine-layer, which the beat file would tag TKG; the Frontier Map reads TKG as Toxicology, and this same run's USDA card uses that second sense. As on the eBay, USDA and Epic cards in this run, TKG is deliberately left out of gardens rather than resolved by guessing, and domains carries the link.

Money angle and metacrisis lever. Money angle: if marketplaces must make catalogues machine-legible to stay discoverable, then structured, verified, citable inventory data becomes a purchased input rather than a nice-to-have — which is RDKG's machine lane, described from the buyer's side. Metacrisis lever: thin. Etsy's stated mission is human-centred commerce, and there is a real tension between "keep human connection at the heart of commerce" and optimising the catalogue for machine consumption. This card notes the tension and does not resolve it; the letter does not address it.

What would prove it wrong (falsifier). The quotation fails if Etsy's shareholder letter for Q2 2026 does not contain these sentences, or if the company withdraws or restates the letter. The implication — that agent-readability is becoming a real budget line for marketplaces — fails if Etsy's next two shareholder letters drop the agentic framing entirely, or if the company reports agent-driven traffic and the figure is negligible and flat. It also fails in a subtler way that the letter's own text hints at: Etsy attributes its take-rate expansion to advertising, saying "Etsy marketplace year-over-year take rate expansion was primarily driven by Etsy Ads, as machine learning-driven improvements continued to enhance relevance and support more effective seller budget pacing." If the money keeps coming from ads while agent-readability stays an unpriced investment, this card describes a posture rather than a market — which is exactly the Reddit finding repeating.

Cross-link, and the corporate fact that ties the run together. The same letter states "On July 30, 2026 we completed the sale of Depop to eBay," with Depop presented as a discontinued operation for all periods — the seller's side of the transaction whose price is established on this run's eBay card (ebay-q2-2026-provenance-layer-depop-price). Etsy also discloses a new $2 billion share repurchase authorization and a Restructuring Plan, and says it intends to "deepen our expertise in strategically critical areas across product, engineering, and customer operations, with a particular focus on expanding and strengthening our team's machine learning skills." A company selling a marketplace, buying back stock and hiring machine-learning talent is reallocating toward the machine layer, though the letter never frames it that way and neither does this card.

Provenance (what I actually read). Fetched Etsy's "Second Quarter 2026 Shareholder Letter" PDF from Etsy's investor-relations asset host, linked from the dated release page investors.etsy.com/news-events/press-releases/detail/225/etsy-inc-reports-second-quarter-2026-results, from the main thread on 21 August 2026. The release page carries published_time: 2026-08-05 and the visible line "August 05, 2026 4:05pm EDT", and the letter's own filename encodes 8.5.26. Every quotation and fact above is read from that PDF: the inventory-signals and long-tail sentence; the conversational-agentic-shopping sentence; the Depop completion sentence and its discontinued-operations treatment; the take-rate/Etsy Ads attribution; the machine-learning skills line; the $2 billion share repurchase authorization; and the Restructuring Plan. The mission sentence is quoted from the "About Etsy" section of the same day's release page, which also supplies the 5 August date. This card contains no derived figures. I did not open the Form 10-Q, the 8-K, the company memo exhibit, or the earnings call, and I searched the letter directly for an agent-traffic figure and found none — which is why no such figure appears here.