Narrative memo, September 2026. Everything measured is labeled measured; everything modeled says so. One loop organizes everything here: read the company continuously → measure its shape → set the objective → diagnose what breaks next → explore what it could become → choose one move → authorize it → execute it → reconcile what happened → learn. The first door into that loop is a margin floor on quotes, because that is where the leak is. Companion: the master thesis, which this memo compresses.
Picture a composite of the operators we have interviewed: a $60M Québec manufacturer, family-built, trucks rolling every morning. The ERP was installed nine years ago. The price book was last touched fourteen months ago. Quotes go out priced from that book plus the general manager's gut, twenty years of good gut, but gut doesn't know that lumber moved eleven percent since March, that one production line is committed to 91% for the next three weeks, or that the customer asking for a discount pays in 66 days. Every detail in that picture comes from a real interview; no single company is described.
The research behind this memo: 74 partner-run exploratory conversations run by a channel partner, plus structured operator interviews and founder-led commercial calls, more than one hundred in total, across pallet plants, distributors, fabricators and field-service companies from $5M family shops to $590M platforms. We asked one question: where does the money actually leak? The answers were four, over and over. Margin is tracked after the fact. Everything is keyed twice between systems that disagree. Capital cannot see the operation. Automation is arriving with rate cards and nobody can compute the cost-per-task truth to judge them against.
None of this is a missing-features problem. It is a shape problem. Every company has a shape: the configuration of products, customers, channels, crews, assets, systems and cash through which it creates value. In biology a trait scales with body mass as Y = a·Mb, and an organism is not a size, it is a set of exponents. A business obeys the same law. Revenue is the mass; cost, capacity, working capital, complexity and margin each scale with it at their own exponent, and almost nobody measures them. A company whose margin scales at 0.8 while its complexity scales at 1.3 is an elephant-sized mouse. The systems record the jobs. They have never measured the exponents, and they were never asked to decide anything at the moment the commitment is made.
Live today. Allometry connects read-only to the systems an operator already runs, ERP, spreadsheets, field software, as they are. A continuous reading layer we call the Pulse resolves the cost that is actually true: executed cost from closed work, not the price book, with every number carrying its source, its method and its confidence. From closed work it measures the operator's coefficients, the exponent for every job type, customer and product, and turns the P&L into a shape. That is the Margin Scan: ninety days of history, the leak computed, the shape drawn, free to the first cohort.
Entering production October 1. The floor goes live on our design partner's daily quoting. Every new quote is checked against p ≥ (1 + f) × c(t), price against today's true cost plus the margin floor, before it leaves. Below floor does not send; it gets repriced, or a human approves it with their name attached. Repricing propagates when input costs move. This is the first door of the product, and it is also the first pricing hypothesis: $2,500 a month, measured against the December gate.
Next. The same reading played forward. A living Profile of the company that management attests. The objective it is pricing toward, stated as the enterprise value the owner wants. A diagnosis of what breaks next, read from the exponents rather than from revenue alone. Three credible adjacent shapes the company could become, each with its effect on cash, capacity, risk and value, and one bounded, measurable Move to test first. Then the reconciliation: what was expected, what happened, and what the company learned. These are designed and specified in the thesis; they are not yet operating in production, and we say so.
The evidence so far, stated precisely. Analysis of ninety days of our design partner's completed work surfaced leakage patterns representing approximately $300K on an annualized basis, measured on real operating data, n=1, and we say n=1 out loud. During that shadow analysis the system also identified a $3.97M prospective over-commitment on a single SKU five weeks before the capacity conflict would have surfaced operationally, identified in analysis, not prevented in live operation; the floor was not yet governing quotes. Allometry is pre-revenue. Data access is not revenue, and nothing here counts as revenue until an invoice is paid. We publish no cross-operator benchmarks until ten operators have twelve months of attested ledger, a public commitment, because this company is a bet on evidence over narrative.
Attested means exactly this: every calculation remains linked to its source records, calculation version, approval history and realized outcome. No blockchain is implied; provenance and reconciliation are.
Agents can hold a setpoint. Control loops that once needed a control engineer now run on a language model with tools: read the sensor, compare to the target, act, repeat. A margin floor is a setpoint. Holding a price to a floor while costs drift is the same loop, run on the books instead of a laser. And the reading that makes it possible is new: models can now resolve an ERP, a spreadsheet and a field system that disagree, without a two-year migration. The cost of truth collapsed, and it was always the barrier.
Software is about to buy physical work by API. It already buys compute, payments and intelligence as a call with a scope, a price and a receipt. The next thing a customer's software will ask for is a pallet run, a crew, a machine-hour, with a spec, a deadline and a budget, expecting an answer in seconds. Stripe collapsed "accept a payment" into one call. Allometry collapses a harder question into one: is this work worth doing for this company, at what price, and can it deliver. The answer comes from the company's shape, its objective, its real calendar and its policies, not from a rate card. Operators whose businesses can answer get found and get paid first.
The systems of record stopped being the moat. ERP, CRM and accounting are cheap to read from and write to now. The value moved from recording the company to deciding for it, and capital has noticed: large rounds are going to companies whose pitch is that the record is not enough and leaders need a live picture they can act on. That thesis is being validated in revenue teams and SaaS. Allometry is the same thesis for the physical economy, with two differences the market has not yet funded: a law that says what the picture should be, and enforcement at the moment of commitment rather than a report after it.
The first door is priced simply and stated as a hypothesis. The Margin Scan is free to a cohort of ten. A read tier, the living Profile without enforcement, is $800 a month. The floor is $2,500 a month, $30K a year, plus $5 per address past five hundred. Design partners enter at $2,500 a month with no fees, weekly sessions with the founder, and the price locked. Enterprise and portfolio are custom and are not in the plan until one closes. We price the commercial value governed and package it as predictable annual software: never per seat, per token or per agent action. Inference is included in the contracted workflow and metered internally per decision.
The math, stated so it can be falsified. One hundred operators on the floor is $3M of annual recurring revenue, the end-2027 milestone, and it requires no enterprise deal and no second vertical. The market is counted bottom-up, not modeled: roughly 17,800 North American physical operators in the revenue band, identified by name, of whom about 2,300 fit the floor precisely. One hundred is four percent of that set. The binding constraint is deployment time per operator, not compute, and the only mechanism that compresses it is the one we measure: every operator on the floor adds attested coefficients to the template, so the second operator deploys faster than the first.
Distribution is one machine, counted end to end. Sixty-five operators a day from the counted list, worked by the founder at the top and by sequenced email below it. A story on day 0, free work on the operator's own quotes on day 3, terms on day 6, a binary close on day 9. A channel partner runs exploratory conversations in its own book and hands over the ones with a live pricing problem, reported as partner-run. No paid acquisition until the funnel is instrumented. No inbound is assumed; referrals are asked for.
The gates, stated so we can be judged against them:
| When | Milestone | Gate it opens |
|---|---|---|
| October 1, 2026 | Floor live on the design partner's daily quoting; 1,000 operators contacted | Pricing hypothesis in market |
| October 2026 | Funnel counted end to end; every lab experiment has a verdict | Paid acquisition permitted |
| December 2026 | Three to five paying design partners; estimated-versus-realized margin reported separately | Second vertical may open |
| Q1 2027 | First complete loop reconciled in a second vertical | Template contract frozen |
| End 2027 | One hundred operators on the floor; $3M ARR | Portfolio and capital products open |
If a gate misses, the plan changes, not the truth.
The kernel is horizontal and the usefulness is specific. Underneath the floor sits one object model, the same for a pallet plant, an agency, a food producer or a hotel: sources and claims with provenance, the company's coefficients, a Profile, an objective, shapes and scenarios, one bounded Move, the policies that govern it, the tools that execute it, and a single Ledger that records what was expected, what was authorized, what happened and what was learned. What makes it specific is a template: the economic units, breakpoints, typical moves and tools for one family of business. Physical and asset-enabled services is the first template because its exponents are the most violent, costs move weekly and prices move annually, so the evidence is largest and fastest there. Templates for professional services, product and CPG, hospitality and recurring B2B are named and closed until the first one proves out.
Three named parts carry the platform. One sovereign Node per company: its graph, its coefficients, its policies and its Ledger, in our cloud or theirs, exportable in full. The Pulse reads the Node's sources continuously and keeps the floor, the coefficients and the Profile from ever going stale; it has no authority, it reads, resolves and raises. One permissioned Index across companies that receives standardized outputs, never raw data, and learns the shape of the market: which interventions moved which exponents, in which family. The Index supplies the prior; the operator's Ledger supplies the truth; the outcome improves the next prior. That is the compounding loop, and it is why the second operator deploys faster than the first.
The durable asset is the Ledger: years of decisions with realized consequences, linked to the coefficients that predicted them. A workflow can be copied in a quarter. That cannot. From it, two extensions are earned rather than assumed. A deep Ledger makes the company's capabilities and prices trustworthy enough for other software to transact against, which is the agent-to-agent call in section three. And the same Ledger is lender-grade evidence: an operator can prove the predicate a lender underwrites, margin above floor for twelve months, receivables performing, concentration bounded, without opening the books. Finance the work, not merely the company. Allometry prepares the package and monitors the outcome. Money never moves through Allometry, and no credit decision is ours.
We are raising $1M to retire three risks in sequence, repeatable deployment, paid conversion and channel distribution, and we do not purchase valuation ahead of evidence. Use of funds, in order: one full-stack engineer so the founder leaves the implementation loop; the coefficient engine and tool contracts as production code; depth in the first template until deployment compresses; then the second template, chosen by evidence. It does not fund paid acquisition before the funnel is instrumented, a second product surface, benchmarks, or any lending. If fewer than three design partners are paying at the December gate, that is the kill condition for the current door, not for the thesis, and everything built for the first template is reusable by the second because the objects are shared.
Three principal risks, with falsifiers. Deployment does not compress: the founder is inside every implementation past operator ten; falsifier, the second operator does not deploy faster than the first; response, the template carries the priors and the first hire is the engineer, not a seller. Operators like the diagnosis and do not act: scans delivered, no moves authorized; falsifier, free work accepted and terms declined; response, the floor is the move, not a report, and day-3 work is done with the estimator on the call, since the buyer is the owner and the user is the estimator. An incumbent descends: an ERP or CRM ships a margin floor; response, the decision lineage and the coefficients live in the Ledger, not in the record system, and their systems attest to their own mutable records while ours attests executed outcomes across systems.
What an investor can verify without trusting the founder: the design partner's ninety-day scan and the coefficients that reproduce it; the floor's shadow-run log, every quote and the floor it was checked against; the funnel counted end to end with dates; the partner-run conversation log, labeled as such; and the December result, reported separately as verified recovered cash, recoverable invoiced value, identified leakage and annualized opportunity. Anything not on that list is narrative, and narrative is not evidence.
Founder-market fit, plainly: I owned a revenue number inside operating businesses, including enterprise software at Coveo, that I could not decompose to the job level, which is this company's origin story. Allometry is solo-founded, pre-revenue, with embedded engineering capacity through a channel partner, live design-partner data, and domain advisors across finance, industrial operations and technical architecture. The first hire internalizes the core after financing. It is a founding system, not one person performing every function.
Every business has a shape. The shape is its exponents. Allometry measures them, decides against them, and enforces the decision at the moment the company commits. See the business. Shape what comes next. Make the change.
— Taylor Gendron, Founder · Montréal | New York · taylor@allometry.com
Companions: the master thesis · the physics of profit · the open letter · the free scan