The April Scoreboard: Who Actually Shipped AI in Hospitality This Month — and What April 11's TCPA Deadline Did to the Procurement Calendar

It is Friday afternoon, the twenty-fifth of April, and I have twenty-three tabs open across two browser windows. One window is the regulators: the SEC’s litigation release page, the FCC’s consumer and governmental affairs docket, UKHospitality’s NICs campaign tracker, the California Department of Industrial Relations wage schedule. The other window is the operators and the vendors: Hi Auto’s funding announcement from the second of the month, Blackbird’s Series B from the eighth, DoorDash’s merchant-AI press release from the ninth, the Olo investor materials I pulled while waiting for Chipotle’s earnings release on Wednesday. There is a half-finished flat white on my desk, a printout of the Restaurant Leadership Conference agenda with Phoenix’s session rooms circled in red ballpoint, and a Moleskine open to a page where, four weeks ago, I had written a single line: what is this month actually going to be about?

I have an answer now, and I want to put it on the record before it stops being surprising.

April 2025 is the first month in the four years I have been writing about hospitality technology where regulators, operators, and vendors moved at the same speed in the same direction. They did not coordinate. They are not, in any meaningful sense, talking to each other. But the calendar lined up, and the result is that for the first time the procurement conversations I am hearing — in hotel corridors at RLC, on operator calls, in vendor founder coffees — are being priced in the same order, against the same constraints, in roughly the same units.

That ordering is the framework I want to put down here, because it is going to anchor how I think about this category for the next eighteen months, and I would rather mark the spot now than retrofit a theory in November.

The framework, stated plainly: hospitality AI economics in 2025 and 2026 will be priced by the regulatory-cost layer first, the labor-cost layer second, and the product-feature layer last. The vendors who internalize that ordering — who lead their sales conversations with what the regulation costs and what the labor line saves before they say a word about the product — will define the next five years of this category. The vendors who keep leading with feature demos will spend 2026 explaining why their pilots stalled.

This essay is the scoreboard for the month that taught me to write that down. It is also, I think, the first piece in what I expect will become a longer framework chain — the kind of doctrinal foundation we will keep building on as the year goes. The framework chain in this column will compound: in a piece we publish in late June we begin articulating what I am tentatively calling an operator-vendor compact, and the framework we develop into final form a year from now in our Four Margins essay starts here, in the four ordered layers of cost that April surfaced. I would rather mark the origin point honestly than pretend the structure arrived fully formed.

The shape of April

Before I impose any frame on the month, here is the month, in dates, the way I have it on the printout taped above my monitor.

The first week opened with funding. Hi Auto closed a fifteen-million-dollar strategic-led round on the first or second — the announcements landed in tranches across the two days — extending a voice-AI franchise that already runs in Bojangles and Panera locations. The same week, Chef Robotics announced forty-three point one million for back-of-house automation, a deal I will return to because the mix of strategic and venture capital on the cap table is doing more work than the headline number.

The second week was where the regulatory tectonics moved. On the sixth, the United Kingdom’s increase to employer National Insurance Contributions — from thirteen point eight percent to fifteen percent, with the secondary threshold dropping from nine thousand one hundred pounds to five thousand — came into force. UKHospitality’s campaign tracker at ukhospitality.org.uk/campaigns/national-insurance-contributions has been the document I have been refreshing for weeks; Kate Nicholls’s framing, in the trade body’s pre-Budget submissions and in subsequent comment, has been consistent: this is “a tax on jobs at the entry level,” and the entry level is precisely where hospitality hires. On the eighth, Blackbird Labs closed fifty million in Series B led by Spark Capital, a round whose composition tells you something about which side of the operator-vendor relationship investors think will accrue power over the next cycle. On the ninth, two unrelated things happened on the same day that I now think will be remembered as the day this framework became unavoidable: the SEC filed its complaint against Albert Saniger in the Southern District of New York under Case No. 1:25-cv-02937 — the litigation release at sec.gov/enforcement-litigation/litigation-releases/lr-26282 is worth reading in full — and DoorDash launched a suite of AI merchant tools the same morning.

Then, on the eleventh, the FCC’s TCPA one-to-one consent and opt-out compliance deadline came into force. Hostie’s compliance writeup at hostie.ai/resources/2025-tcpa-fcc-compliance-checklist-ai-voice-calls-restaurants has been circulating in the operator slacks I lurk in; it is the cleanest summary of what the rule actually requires for restaurants running AI voice and SMS programs, and it is being passed around as a procurement-gating document, not as a compliance footnote.

The third week was the trade show. RLC Phoenix ran from the thirteenth through the sixteenth, and on the sixteenth itself, Red Roof announced its strategic partnership with HotelIQ to roll out AI-driven revenue management across the portfolio. I want to come back to the Red Roof deal because it is the kind of mid-market hotel partnership that almost no one wrote about, and which I think is the single best illustration of the labor-cost-second layer of the framework.

The fourth week is where we are now. Burgerbots is anticipated to open its Los Gatos pilot on or around the twenty-first. Chipotle reported on Wednesday the twenty-third, surfacing — in answer to an analyst question about throughput — that the Olo Catering Plus pilot is live in a defined market. Hilton’s Q1 earnings on the twenty-ninth, Yelp’s Spring Release the same day, and Wendy’s Q1 on the thirtieth are all anticipated as I write — by the time you read this they may be in the rearview, but I want to mark that the framework I am laying out below was written before those calls landed.

I have left out a dozen things. There were smaller funding rounds. There were operator press releases I am skeptical of. There was a flurry of hotel-tech vendor announcements timed to RLC that I will deal with in shorter pieces. But the spine of the month — the events that, if you imagine them as nodes on a timeline, you cannot remove without the month falling apart — is what I have just listed. And the spine has a shape.

The shape is this: every node on the spine is either a regulatory event, a labor-cost event, or a product-feature event. They occur in clusters. The clusters do not interleave the way they did in 2023 or 2024. In April 2025, the regulatory events came first in the month and set the procurement terms; the labor-cost events ran underneath them as a continuous bass note; the product-feature events came last, and were priced — by the operators I have been speaking to — against the first two.

That is the framework. The rest of this essay is the evidence.

The regulatory-cost layer arrives

I want to take the SEC’s Saniger complaint and the FCC’s TCPA deadline together, because I think they are the same event in a generational sense, even though they have nothing legally to do with each other.

The SEC’s litigation release on the ninth — case number 1:25-cv-02937 in the Southern District of New York — describes a defendant who, the agency alleges, raised more than fifty million dollars from investors on the representation that Nate, the one-tap checkout app, was powered by proprietary artificial intelligence. The complaint’s core allegation is that the artificial intelligence in question was, in fact, hundreds of human contractors in a call center in the Philippines, manually completing the transactions that investors were told were autonomous. I am not going to relitigate the complaint; the agency’s filing speaks for itself, and the criminal complaint filed in parallel by the U.S. Attorney’s Office speaks for itself further. What I want to mark is the category-defining function of the action.

For three years, every hospitality vendor I have spoken to has used the phrase “AI-powered” as if it were a free option. It cost nothing to say it; it implied capability without implying obligation; and it functioned as a marketing accelerant, particularly into the operator middle market where buyers were uncertain about the underlying technology and were comforted by the label. The SEC’s filing on the ninth changes the option’s pricing. The phrase “AI-powered” now carries a non-zero probability of being adjudicated in a Southern District courtroom, with discovery, depositions, and a litigation release that will follow the defendant for the rest of his professional life. Every vendor founder I have spoken with this month — and I have had at least eight of these calls — has independently raised the Nate complaint without prompting. The phrase I keep hearing is “we are tightening our claims language.”

Tightening claims language is not a legal hygiene exercise. It is a procurement-calendar event. When a vendor tightens claims language, three things happen: the marketing collateral gets rewritten, which delays the sales cycle by four to six weeks; the master services agreement gets re-papered, which adds a legal-review step on the operator side; and — most importantly — the operator’s general counsel becomes a stakeholder in a category they previously ignored. The general counsel was not, in April of 2024, a buyer of hospitality AI. The general counsel is, in April of 2025, a gate on it.

The FCC’s TCPA deadline on the eleventh did the same thing to the voice and SMS subcategories, with sharper teeth. The rule — and Hostie’s compliance writeup at the URL above is the version I have been passing around — requires that opt-out requests be honored within ten business days, that one-to-one consent be obtained before automated outreach, and that the consent records be retained in a defensible form. The penalty per violation, in the worst case, runs into five figures. For a restaurant chain running AI voice across two hundred locations, with each location making hundreds of automated calls per week, the math is straightforward and unpleasant.

Here is what I find significant. The TCPA rule did not invent the obligation; the underlying statute is decades old. What the April eleventh deadline did was operationalize the obligation against a technology category that had spent two years assuming it would be regulated later. Hi Auto, which closed its fifteen million two days before the rule came into force, is on the right side of this — their compliance posture has been part of the franchise-system sale into Bojangles and Panera from the start, and the strategic-led composition of the round (rather than a pure venture round) reflects what those franchise systems are paying for: not the AI, but the ability to deploy the AI without the franchisor’s general counsel killing the pilot. I think this is going to become more important than the underlying technology in the voice-agent subcategory, and we will probably write a dedicated piece about the maturity curve of that category — in a later piece we publish on the voice-agent category — once the post-deadline operating data comes in.

The framework principle: when a regulatory event with teeth lands inside an AI subcategory, the procurement conversation re-prices around the compliance posture, not the model quality. Saniger and TCPA, together, mark April 2025 as the month that re-pricing started in earnest for hospitality.

The labor-cost layer hardens

If the regulatory-cost layer is the new top of the operator’s procurement stack, the labor-cost layer is the bedrock underneath it, and April hardened that bedrock in two jurisdictions at once.

The UK NICs hike on the sixth is the one that will get the headline coverage, and rightly. The increase from thirteen point eight percent to fifteen percent in the employer rate, combined with the threshold drop from nine thousand one hundred pounds to five thousand, is not a marginal change. UKHospitality’s modelling, which Nicholls has been carrying into every parliamentary committee and trade-press interview through the spring, puts the additional annual cost to the hospitality sector in the low billions of pounds. The campaign page at ukhospitality.org.uk/campaigns/national-insurance-contributions tracks the trade body’s lobbying activity and, more usefully, links to the operator submissions that quantify the impact at the unit level. The submissions I have read are devastating. A medium-sized casual-dining group running two hundred sites with an average of forty heads per site is looking at a six-to-seven-figure additional annual cost, on a labor base that was already absorbing the National Living Wage uplift that took effect on the same day.

Kate Nicholls’s framing — “a tax on jobs at the entry level” — is the framing I want to flag, because the entry level is the level at which most hospitality AI deployments are economically justified. The drive-thru order-taker, the front-desk checker-in, the host who answers the phone for reservations, the kitchen prep worker who portions and assembles: these are the roles that vendor pricing models assume will either be augmented or replaced. When the cost of those roles rises by three to five percent overnight, the payback period on a hospitality AI deployment compresses by months, sometimes by quarters. That compression is what is moving deals through procurement that were stuck in 2024.

The California analog is the AB 1228 fast-food minimum wage, which has held at twenty dollars since April of last year. The compounding effect, twelve months in, is starting to show in operator commentary. Wingstop’s Smart Kitchen rollout, which by the company’s own disclosures was at well over two hundred stores by the end of Q1, is best read as a direct response to the AB 1228 economics. Wendy’s FreshAI scaling — Restaurant Dive’s reporting at restaurantdive.com/news/wendys-deploy-digital-menu-boards-drive-thru-ai-500-restaurants-2025/746977 lays out the five-hundred-restaurant target for 2025, and the Q1 call on the thirtieth is anticipated to give us a clearer line on the rollout pace — is the same response in a different operating model. Both companies are pricing labor-replacement value as the primary line, and AI capability as the enabling technology that makes the line possible. The product-feature conversation comes a distant third.

Here is what I want to mark. In April 2024, an operator running a labor-cost model and a hospitality AI vendor running a product-feature model could have an hour-long sales conversation in which neither one mentioned regulation. That is no longer possible in April 2025. The conversation now opens with the regulatory layer — what is the vendor’s TCPA posture, what is its data-retention defensibility, what claims is it making and can those claims be substantiated — and only then proceeds to the labor-cost math. The product-feature conversation, where it happens at all, happens at the end of the meeting, in the last fifteen minutes, after the contract has been functionally decided.

The product-feature layer (still) matters least

I want to be careful here, because I am not arguing that the underlying product does not matter. It does. A voice agent that hallucinates menu items will be ripped out within thirty days regardless of how clean the TCPA posture is. A kitchen-automation system that produces inconsistent assembly will lose the operator more than it saves. The product-feature layer is a gate, not a driver — the floor below which the deal does not happen at all.

What I am arguing is that the product-feature layer no longer differentiates the deal once it has cleared the floor. The differentiation in April 2025 is happening one and two layers above the product, and the vendors who are still pitching as if the product is the lead are losing ground without quite understanding why.

DoorDash’s AI merchant tools, announced on the ninth — the same day as the SEC filing — are an interesting case in point. The tools themselves are competent: menu optimization, smart photography suggestions, automated promotion targeting. They are not, on their merits, dramatically better than what Toast or Olo could put together with three quarters of engineering effort. What makes the DoorDash announcement matter is that DoorDash is sitting on transaction data across a million-plus restaurants, and the announcement reframes the company’s relationship with its merchants from logistics partner to AI infrastructure provider. The product is the trojan horse; the data-position is the move. Olo’s response — the Catering Plus pilot inside Chipotle, surfaced on Wednesday — is the same architectural move from a different starting point. The product-feature conversation is the surface; the data-position conversation is what the operators are actually pricing.

This is the layer where, I think, the next five years of vendor competition will be decided, and it is the layer where the framework I am building out here will need the most work in subsequent essays. I am not yet sure how to price a data-position relative to a regulatory-cost layer; I think the answer will involve a notion of switching cost that we have not yet articulated cleanly. But I want to mark the question now and come back to it.

What the funding rounds say

Three funding rounds in April are worth reading as theses. Hi Auto’s fifteen million on the first or second is a deployment thesis: the capital is going into deeper integration with the franchise systems Hi Auto already serves, and the strategic-led composition of the round (rather than a pure venture-led structure) tells you that the franchisors themselves are paying for the right to scale the deployment without taking on the regulatory and compliance overhead. Hi Auto is selling, effectively, a compliance-wrapped voice agent; the model and the audio pipeline are the cost of entry, and the deployability inside a franchise system with two thousand units is the product.

Chef Robotics’ forty-three point one million on the first is a unit-economics thesis. Back-of-house automation has been the graveyard of hospitality robotics for a decade; the bet here is that the labor-cost layer has hardened enough — California, UK, and the federal minimum-wage politics that will recur through the year — to make the unit economics work for the first time. I am not yet convinced, but the cap-table composition (strategic capital alongside venture) is the same pattern as Hi Auto: the operators who would deploy the system are participating in the funding, which means the deployment risk is being absorbed at the cap-table level rather than at the procurement level.

Blackbird’s fifty million in Series B led by Spark on the eighth is the most strategically interesting of the three, and the one I am still working out. Blackbird is, on the surface, a loyalty and payments network for independent restaurants. Underneath the surface, it is a data-position thesis: the network effects accrue if Blackbird can sit between the restaurant and the diner on enough transactions that the data becomes definitive for both sides. Spark’s leadership of the round, with the firm’s track record in marketplace and infrastructure businesses, suggests the investors are pricing the data-position as the long-term value, not the loyalty product itself. If that is right — and I think it is — Blackbird is a category-defining bet on the same architecture that DoorDash and Olo are racing toward from different starting points.

The framework principle: when three significant funding rounds in a single month all price the deployment, the regulatory wrap, or the data-position as the lead, and price the product feature as the cost of entry, the category has moved. April 2025 is when the move became visible on the funding side.

What enterprise operators did

Hilton’s Q1 on the twenty-ninth, with the AI Planner update — Hotel Dive’s preview coverage at hoteldive.com/news/hilton-bets-ai-with-vendor-partnerships/818938 sets the table for what we are anticipating — will be read by most of the trade press as a feature announcement. I think the more important reading is as a vendor-partnership thesis. Hilton’s posture has been, consistently, that the company does not build its own AI; it partners with vendors who can clear the regulatory and compliance bars, and it captures the operating value through deployment scale. That posture is exactly the framework I am laying out: the regulatory layer is the floor, the labor layer is the math, and the product feature is the implementation detail that the vendor brings.

Wendy’s FreshAI scaling — anticipated to be the centerpiece of the Q1 call on the thirtieth — is the operational variant of the same thesis. Wendy’s is not building a model; it is deploying a stack that, on the company’s own reporting, has cleared internal regulatory review and is being scaled against a labor-cost case the company has been making publicly for two years. Wingstop’s Smart Kitchen, which by the disclosures we have was past two hundred stores at the close of Q1, is the same pattern in a different cuisine and a different unit economics. Both companies are paying for the deployability of the AI; the AI itself is the cost of entry.

Chipotle’s Q1 on Wednesday surfaced — in answer to an analyst question I will quote when the transcript is final — the Olo Catering Plus pilot, and the framing matters more than the pilot’s surface-area. Chipotle is using Olo as the operating layer; Olo is using Chipotle as the data position. The pilot itself is small, and the financial impact in Q1 was immaterial, but the architecture the pilot establishes is the architecture we should expect to see repeated across the enterprise QSR segment through the back half of the year. I expect this to be a thread we pick up again in the operator-vendor compact piece we will publish in late June.

Red Roof’s partnership with HotelIQ on the sixteenth, announced during RLC, is the most under-covered enterprise-operator move of the month, and I think it is the cleanest illustration of the labor-cost-second layer working in the mid-market hotel segment. Red Roof is not Hilton; it does not have the engineering bench to build its own AI revenue management, and it does not have the brand premium to absorb the cost of a bad deployment. The HotelIQ partnership prices the AI capability as a labor-replacement line — front-desk and revenue-management labor that mid-market hotel operators have been struggling to keep in seat through the post-COVID labor market — and prices the product feature as the enabling technology. The deal does not make a marketing splash, but it is, I think, the template for thirty similar deals across the mid-market hotel segment over the next twelve months.

What the trade show told us

I spent four days at RLC Phoenix from the thirteenth through the sixteenth, and the most useful intelligence I came home with was not from the keynote sessions. It was from the procurement timing.

Three years ago, RLC was a product-evaluation conference: operators came to see what was new, vendors came to demo, and the deals that flowed from the show closed in Q3 or Q4. Two years ago, RLC was a budget-alignment conference: operators came with budgets and shopping lists, and the deals closed in Q2. This year, RLC was a contract-finalization conference. The buyers I spoke to — and I had at least a dozen of these conversations across the four days, with operators ranging from sixty-unit to multi-thousand-unit chains — were not browsing. They were closing.

The shift in posture is what I want to flag, because it is the procurement-calendar signal that the framework predicts. When the regulatory-cost layer prices first and the labor-cost layer prices second, the procurement calendar compresses, because the operator cannot afford to leave deployments to Q4 anymore. The labor cost is hitting now; the regulatory exposure is live now; the AI deployments that were scheduled for back-half-of-year rollout in 2024 are now being scheduled for Q2 and early Q3 of 2025. RLC Phoenix this year was the show where that compression became visible in the hallways.

The vendors who picked up on the compression and brought MSA-ready paper to Phoenix did well. The vendors who brought a demo and a deck did not. I want to mark that observation because, in twelve months when we look back at the vendors who grew and the vendors who stalled, the procurement-readiness posture at RLC Phoenix 2025 will be one of the better predictors I have seen of which side a vendor ended up on.

The framework we’re building toward

Let me state where I think this lands as a framework, and let me be honest about what is provisional in it.

The provisional version of the framework is four layers, stacked, with the procurement conversation moving top to bottom. The top layer is the regulatory-cost layer: TCPA, SEC enforcement posture, GDPR for cross-border deployments, the EU AI Act for systems with European exposure. This is the layer that prices first because it is the layer with the sharpest teeth and the shortest tail of remediation. The second layer is the labor-cost layer: minimum-wage trajectories, NICs and equivalent employer-side taxes, scheduling regulation, predictive-scheduling laws. This is the layer that prices the deployment math and determines the payback period. The third layer is the data-position layer: who sits on the transaction record, who sits on the guest record, who sits on the operating telemetry. This is the layer that will determine vendor pricing power five years out, and it is the layer I am least confident I have the right framework for yet. The fourth layer is the product-feature layer: the model, the integration, the user experience, the operating reliability. This is the layer that gates the deal — the floor — but no longer differentiates it.

The framework is provisional in three specific ways, and I want to flag them so I can come back and correct them when I have learned more.

First, I am not yet sure how the regulatory layer and the data-position layer interact when they conflict. A vendor with a strong data-position and a weak regulatory posture should, on the framework’s logic, lose to a vendor with a clean regulatory posture and a weaker data-position. I think that is right, but I have not seen enough deals close in 2025 yet to test it. The Saniger complaint is the closest test we have, and the answer there is unambiguous: the regulatory layer wins. But Saniger is an extreme case, and the more interesting tests will be in the middle of the distribution.

Second, I am not yet sure how the labor-cost layer behaves in a recession. The framework as I have laid it out assumes the labor-cost layer is hardening — wages rising, regulatory employer-cost loads rising, labor availability tight. If the macro turns and labor becomes available at lower cost, the payback math on hospitality AI deployments stretches, and the framework’s ordering may invert at the middle two layers. I do not think this is a near-term risk, but it is a risk I want to mark.

Third, I am not yet sure how to integrate the product-feature layer cleanly. The argument I have made — that it is a floor and not a differentiator — is true at the deal-close level, but it understates how much engineering investment is required to clear the floor. The vendors who clear it cheaply will compound faster than the vendors who clear it expensively, and the framework needs a treatment of that compounding that I have not yet written.

These three caveats are the open work. They are also why I expect to write the next several Mise columns as iterations on this framework rather than as fresh frames. The framework chain we are starting here will, I expect, get its first major refactoring in the operator-vendor compact piece we publish in late June, and its first attempt at a unified treatment of the four layers in a piece we will publish in late summer — what becomes, eventually, the framework we develop into final form a year from now in our Four Margins essay. I would rather build the chain in public, with the caveats marked, than pretend to a coherence I have not yet earned.

What I’m holding

I want to close with what I am holding from this month, in the editorial sense — the observations I have not yet resolved into framework but which I do not want to lose.

I am holding the fact that the Saniger complaint and the TCPA deadline landed within forty-eight hours of each other, and that nobody — not the vendors, not the operators, not the trade press — wrote about them as the same event. I think they are the same event, and the failure to see them as the same event is the failure that will mark the vendors who lose 2026.

I am holding Kate Nicholls’s verbatim framing — “a tax on jobs at the entry level” — because I think the entry level is going to become a contested term over the next twelve months. The vendors who are pitching labor-replacement value are going to need a more careful language for which jobs they are replacing and which they are augmenting, and “the entry level” is going to be the line of contestation.

I am holding Mark Podolsky’s commentary from the RLC main-stage Q&A on the fifteenth — I will not paraphrase the quote until I have the transcript in front of me, but the substance was that the operator side of the relationship is no longer willing to pilot AI deployments without a defensible compliance posture from the vendor — because I think it is the cleanest articulation of the procurement-calendar compression I observed in the hallways, and I want to carry it forward to subsequent columns.

I am holding the Burgerbots Los Gatos opening, anticipated for around the twenty-first, because the unit-level operating data from that pilot is going to be one of the better tests of the labor-cost layer’s behavior in a California fast-casual context, and I want to have a baseline observation logged before the data starts to come in.

I am holding the Hilton Q1 call on the twenty-ninth, the Yelp Spring Release the same day, and the Wendy’s Q1 call on the thirtieth as the immediate evidence base against which the framework I have laid out here will get its first stress test. If those three events confirm the layer-ordering I have proposed, the framework gets a much firmer foundation. If they contradict it, I will say so in the next Mise column. I would rather be wrong on the record than vague on the record.

What we’ll be tracking through year-end

For the rest of the year, the Mise column will return to this framework on a roughly monthly cadence. The scoreboard format I have used here — regulatory events, labor-cost events, product-feature events, then the funding rounds and operator moves read against those three layers — will be the recurring structure. I will not always agree with the framework as the year goes on; I expect to refactor it at least twice, and I expect the late-June piece on the operator-vendor compact to refactor the relationship between the regulatory layer and the data-position layer in particular.

The dates I am marking on next month’s printout, which is already taped above the calendar: the May fast-food employer disclosure compliance dates in California; the DoorDash and SevenRooms partnership announcement that I am told is coming on or around May sixth; the National Restaurant Association show in Chicago in mid-May; the UK trading-update season through the back half of the month for the LSE-listed hospitality groups, which will be the first set of operator disclosures that bake the NICs hike into the operating line. Each of those events will be a node on the May scoreboard. Each of them will, I expect, fit the framework I have laid out here.

If they do not — if the framework breaks against the May evidence — I will say so. The doctrinal posture of this column is that the framework serves the evidence, not the other way around. April 2025 was the month the evidence finally lined up. May will tell us whether it stays lined up. The next eighteen months will tell us whether the framework chain we are starting here holds up well enough to be worth refining, or whether it needs a wholesale rebuild.

Either way, the work happens on the record. That is the principle. That is the column.

— Eitan writes the Mise column. Tips: tips@tabletransfers.com.