Archived. Superseded by v2.0 (Business Shape) wherever the two conflict; kept frozen for traceability. Fifty sections plus the v1.0 lock. The design partner is never named.
Superseded (Sept 2 2026): v2.0 (
MASTER-THESIS-2026-09.md, Business Shape) is canon wherever the two conflict. This file is frozen as the v1.1 archive; Appendix E of v2.0 maps every section here to its destination.
v1.1 — Aug 30 2026. Revision log (per §50 variance discipline): v1.0 frozen Aug 29. v1.1 (Aug 30) graduates two queued candidates on the improves-GTM criterion and founder ruling: (1) identity resolution — strategic planning that executes, entered per-template; (2) Shape and Shape Templates as named objects. Six surgical edits (§1, §2, §3, §4, §6, §50); nothing else reopened. Expected validating evidence: template-wedge resonance in cross-vertical channel calls. Otherwise still frozen: Freeze the narrative. Run the experiment. Let evidence earn the next revision. New ideas enter the backlog first and graduate into canon only if they change the product architecture, improve present GTM, reveal a new compounding asset, or carry customer/market evidence. Taylor's verbatim master thesis, banked as canon. Supersedes prior category-ladder framing where they differ; PLATFORM-2026-08.md carries the integration deltas and open tensions. Fact-checks on external claims (a16z Machine Age Fund, o9, Aura by Alvéole, G2 categories) recorded in the backlog entry of the same date.
| Horizon | Sections | Status |
|---|---|---|
| Product thesis — now to seed | §1–3, 5–8, 12, 16–18 | Canon. What we build and sell today. |
| Platform expansion — seed to Series A | §4, 9–11, 13–14, 19 | Sequenced optionality. Each stage gated on the prior. |
| Physical-AI infrastructure — long horizon | §15, Part II (§22–27) | Directional thesis. Investor narrative, not a build plan. |
| Part III addendum (Aug 29) | §28–35, 40, 42–43 → product · §36–39, 44–46 → expansion · §41 parked, §47–48 → long horizon | Same gating applies; the ladder's Product 1 (Revenue Recovery) is the post-pilot pricing evolution, not a mid-campaign switch. |
Category rule refined (v1.1): identity vs door. The identity is strategic planning that executes; category listings are distribution tactics that follow each launched template's wedge. One external category per stage — never presented simultaneously:
Allometry builds a living economic model of a business. It connects demand, pricing, customers, costs, capacity and realized performance into a self-correcting decision graph. This helps companies determine: which work to pursue, which work to decline, what to charge, whether capacity exists, which resources should execute it, whether the expected return justifies the commitment, and how the outcome should change the next decision.
Entry through margin-aware quoting and capacity validation for 20M–200M physical operators. They have accounting, CRM, ERP, FSM — but the most consequential commercial decisions still happen in spreadsheets, inboxes and estimator intuition.
Defense: true cost of work · customer/job-level margin · historical margin leakage · prevent quotes below the floor · flag capacity, payment and execution risk. Offense: which opportunities to pursue · recommend the right price · customers with expansion potential · allocate scarce capacity · CAC/LTV/strategic account value · respond to demand, cost and market changes.
Customer-facing headline: Know what to sell, at what price, and whether you can deliver it.
The per-template wedge principle: every Shape template ships with its own entry module — asset-heavy → pricing + the floor · agencies → engagement margin · hospitality → rate and occupancy economics · food producers → SKU contribution. Same brain, same arc (Profile → Improve/Reshape → Plan → Operator); a different first module per Shape.
Allometry governs the moment information becomes an economic obligation: submitting a bid, setting a price, accepting an order, signing a contract, reserving capacity, assigning labour or equipment, deploying capital, dispatching a human, asset or robot.
Full operating sequence: Sense → Plan → Decide → Authorize → Commit → Execute → Account → Learn. Allometry owns decisioning, authorization and commitment.
The atomic object: the Commercial Commitment — an authorized agreement to deliver a defined unit of work, at a price, against specific capacity, risk and economic constraints. Quotes, contracts, orders, capacity reservations, asset dispatches and financed jobs are different states or forms of a Commercial Commitment.
The spine of the product:
| Object | Meaning |
|---|---|
| Signal | Something economically relevant changed |
| Hypothesis | What Allometry believes the change means |
| Decision | The selected response |
| Commitment | The authorized economic obligation |
| Execution | Work performed by a human, agent or machine |
| Outcome | What happened operationally and financially |
| Learning | How the next hypothesis or policy changes |
The common denominator is not industry — it is businesses with variable economics by customer, unit, job, location or moment. Universal sequence: demand → price → capacity → cost-to-serve → expected margin → commitment → execution → realized margin → next decision.
| Vertical | Economic unit |
|---|---|
| Physical services | Job, route, asset-hour |
| Construction | Bid, project, crew-day |
| Logistics | Shipment, lane, vehicle-hour |
| Agencies | Engagement, retainer, deliverable |
| B2B SaaS | Account, contract, seat, usage |
| Hospitality | Room-night, table, event |
| Restaurants | Menu item, order, table-hour |
| Retail and CPG | SKU, order, promotion, location |
| Professional services | Engagement, person-hour |
| Lending | Borrower, facility, capital deployment |
Shape, defined (graduated v1.1): a company's Shape is its economic morphology — the unit of work, cost structure, capacity constraints, demand pattern, cash cycle, and scaling behavior. Asset-heavy operators are the first family of Shapes; the ontology is Shape-agnostic. One shape-agnostic brain, one shape family at a time.
Product and GTM stay asset-heavy today. The data model is built around a universal economic unit of work — and the Node data model never hardcodes asset-heavy.
Connects: customers, opportunities, quotes, jobs/SKUs, locations, labour, equipment, capacity, input costs, acquisition costs, payment behaviour, market conditions, margin policies, approval authority, expected outcomes, realized outcomes. Every relationship carries: source, current value, confidence, last-updated, historical versions, permissions, expected relationship, realized evidence.
The enduring moat is not a fine-tuned model. It is the accumulated record of what the company believed would happen, what it decided, what it committed to, what it executed, and what happened economically.
Do not train a foundation model per operator. Create a sovereign operating model of each company using frontier models, open models and deterministic systems selectively:
| Intelligence layer | Function |
|---|---|
| Deterministic calculations | Accounting and economic truth |
| Small/open models | Extraction, classification, private routine work |
| Statistical models | Cost, demand, capacity, conversion forecasts |
| Frontier models | Ambiguous reasoning, novel scenarios |
| Decision graph | Company-specific context and relationships |
| Policy engine | Margin, risk, permissions, authority |
| Economic router | Cheapest reliable decision method |
| Attested ledger | Decisions, evidence, outcomes |
Customer data stays operator-controlled. Allometry owns: economic ontology, decision architecture, policy engine, Shape Templates (née deployment templates), connectors, evaluations, cross-company learning methodology.
Models understand the world. Allometry understands this company.
Request → MCP → commercial router → decision graph → policy engine → approval or execution → attested outcome.
Allometry is not another agent builder. It is the economic governance layer used by agents.
Closed loop: Observe → structure → predict → decide → authorize → act → attest → learn. Per quote: context, expected cost/margin, confidence, recommendation, approval/exception, won/lost, actual execution cost, realized margin, expectation-vs-reality delta, correction to future assumptions.
The network supplies the prior. The operator's ledger supplies the truth. Safe assumptions update automatically. Margin floors, strategic policies, capital decisions and material pricing changes require evidence, versioning, shadow testing, approval and rollback.
Quality-gated build: map 10,000, deeply structure the first 1,000, verify the first 100, connect the first 10–20. A large shallow directory is reproducible; verified capacity, quotes and realized outcomes are not. Each profile carries a claimable, machine-readable commercial identity: services, locations, service radius, industries, certifications, equipment, capabilities, approximate capacity, typical project sizes, historical work, response times, procurement requirements, public pricing signals, queryable agent endpoint.
Not 10,000 bespoke websites. Instead: generate 10,000 structured profiles → discoverable by people, search engines and agents → claim and verify → auto-generated commercial page → connect private systems → activate governed quoting and capacity responses.
Progression: Directory → verified operator graph → decision software → transaction network → financial infrastructure.
The Index discovers who can do the work. The decision graph determines whether they should. The ledger proves what happened.
Buyer-agent: "Find three operators capable of completing this project within 200 km, by October, with the required certifications and available capacity." The Index identifies; each operator's sovereign model evaluates (cost, capacity, customer quality, payment risk, strategic value, margin, delivery probability); returns a governed quote, declines, or escalates.
API/MCP actions: search_suppliers · check_capability · request_quote · check_capacity · reserve_capacity · verify_margin_policy · dispatch_work · finance_job.
Every priced and completed job strengthens an attested record of what the operator can deliver profitably. Lender questions answered: contract real? priced profitably? capacity? estimate accuracy? end-customer payment? projected→realized margin? equipment ROI? backlog financeable?
Products: job-specific working capital, PO financing, receivables financing, utilization-tied equipment finance, insurance underwriting, acquisition financing, capacity-expansion financing, continuous lender monitoring.
Flow: quote approved → contract verified → margin and capacity validated → financing offered → work executed → outcome attested → receivable repays facility.
Begin as the intelligence, verification and monitoring layer — not a balance-sheet lender. Revenue: origination fees, underwriting-data fees, monitoring subscriptions, lending-partner rev share, API fees. Permissioned/ZK proofs can eventually verify backlog quality, utilization or margin consistency without revealing every transaction.
| Platform | Primary system | Core question |
|---|---|---|
| o9 | Enterprise planning brain | How should demand, supply and resources be planned? |
| Palantir | Operational ontology | How should the enterprise understand and act on data? |
| Rillet | Accounting general ledger | How should activity be recorded and reported? |
| Allometry | Commercial decision and commitment graph | Should we accept this work, at what price and against which capacity? |
Combined: o9 plans the business. Palantir models and operates the enterprise. Allometry decides and authorizes the economic commitment. Execution systems perform the work. Rillet records the financial consequence.
Physical-AI answers: how can machines perceive, reason and execute work? Allometry answers: which work should be executed, at what price, by which capacity and with what expected return?
a16z Machine Age thesis (physical infrastructure supporting AI): the fund finances new physical supply; Allometry determines how physical capacity is priced, committed and deployed. As intelligence becomes abundant, physical capacity becomes scarce — Allometry governs its economic allocation.
Machine Age Index: data-centre builders, electrical contractors, cooling providers, energy infrastructure, precision manufacturers, robotics integrators, component suppliers, installation/maintenance operators — both ICP and first structured supplier network. Not a hardware company today: the commercial software, data and commitment layer serving the Machine Age supply chain.
Lead with the understandable outcome ("biodiversity data for real estate" → "margin intelligence for physical operators"), not the sensors/AI underneath.
Website: headline "Know which work to take, what to charge and whether you can deliver it profitably." · interactive entry: "Enter your company website. See where margin may be leaking." · flow: Measure → Decide → Improve. Sovereign graph, MCP, router, agentic architecture stay underneath the promise.
| Package | Indicative pricing |
|---|---|
| Margin Scan | Free |
| Operator | ~$30K annually |
| Portfolio | 60K–150K annually |
| Network and API | Custom |
Implementation 7, 500–20,000 initially, until deployment is reliably repeatable. Eventual pricing dimensions: operating units, revenue under governance, decisions evaluated, protected margin, API volume, transactions, realized value.
Initial category: Pricing Software (dynamic pricing, margin optimization, real-time recommendations, ERP/CRM/CPQ integration). Positioning: "AI pricing and commercial decisioning for physical operators." Progression: Pricing Software → CPQ (once the full quote is created/managed) → Decision Management Platforms (once policy engine/automated authorization/APIs mature) → Commercial Decision Intelligence (category we define). Avoid G2's generic Decision-Making Software (collaboration/boards). Initial badges: High Performer Mid-Market, Easiest Setup, Best Support, Fastest Implementation, Highest Adoption, Most Implementable. Eventual claim: the best AI pricing platform for physical operators.
No "AI CEO" framing · don't lead with agents/MCP/models · no generic horizontal workflow platform · don't replace accounting systems · no hardware without data/distribution advantage · no 10,000 bespoke websites · never expose private customer data · paid placement never corrupts recommendations · LLMs never establish accounting truth · no unrestricted self-modification of economic policy · don't broaden GTM beyond asset-heavy operators too early.
Allometry builds a living economic model of physical operators. It connects demand, pricing, capacity and realized performance into a self-correcting decision graph. This allows companies, agents and autonomous systems to determine which work to accept, what to charge, where to deploy capacity and whether the expected return justifies the commitment.
Today: margin-aware quoting and capacity validation. Next: the commercial decision graph for physical operators. Then: the transaction and data network for the operating economy. Eventually: the economic commitment layer through which work is priced, allocated, executed, attested and financed.
Shortest complete expression: Allometry governs the moment a business commits.
The wave to ride is not "humanoids." It is physical capacity becoming callable. Today companies call APIs for compute, messages, payments, models. Next, buyers and agents call APIs to request physical work from humans, conventional equipment, robots and eventually humanoids. Allometry becomes the commercial protocol and economic control plane for those work calls.
Every wave creates a new abundance, then a coordination problem — and rails capture the value:
| Wave | New abundance | Rails that captured value |
|---|---|---|
| Internet | Websites, online users | Search, hosting, DNS, payments, analytics |
| Cloud | Programmable compute | AWS, identity, observability, APIs, billing |
| Mobile | Connected users, apps | App stores, notifications, maps, mobile payments |
| Gig economy | On-demand human capacity | Marketplaces, dispatch, identity, ratings, insurance |
| SaaS | Specialized business systems | Integration, warehouses, billing, identity |
| Agentic AI | Autonomous digital labour | Model routers, MCP, sandboxes, observability, agent payments |
| Physical AI | Programmable physical capacity | Work protocols, dispatch, verification, insurance, settlement, financing |
a16z's Machine Age Fund finances the supply side. As supply grows, the scarce layer becomes: who should do which work, for whom, at what price, under whose authority, with what proof of completion. That is Allometry's opening.
Not a robot marketplace, humanoid dashboard, payments API, fleet manager or voice agent — the layer connecting them: The Work API for the physical economy. Allometry makes physical capacity discoverable, queryable, priceable, authorizable, reservable, dispatchable, verifiable, payable, financeable.
Core platform objects: Actor · Capability · Capacity · Work · Quote · Commitment · Outcome · Settlement.
The first protocol stays narrow — six endpoints: describe_capability · check_capacity · evaluate_work · authorize_quote · record_commitment · attest_outcome. Search, dispatch, settlement and financing follow only after the core economic loop works.
Do not wait for humanoid adoption. Build the same rails on human field teams, contractors, fleets, industrial equipment, warehouses, service vendors, conventional automation. A robot later becomes another capacity type inside the existing graph — as payments served human web transactions before subscriptions, platforms, APIs and agents. Accumulate customers, operating data, economic policies and transaction history before physical autonomy is mainstream.
Beachheads: (1) Industrial and field services — aligns directly with the current quoting wedge. (2) Data-centre / Machine Age supply chain — electrical contractors, cooling/HVAC, generator and power suppliers, maintenance, precision manufacturers, robotics integrators, commissioning firms: rapid demand growth, severe capacity-allocation problems. (3) Warehousing and material handling — early mixed environment (humans, forklifts, AGVs, AMRs, arms, 3PLs), but device-control is competitive: stay on commercial allocation and economic outcomes.
Own: the Operator Index, capability/capacity graph, Physical Work Call schema, commercial decision graph, economic policy engine, quote and commitment logic, Attested Operating Ledger, cross-operator benchmarks, underwriting signals, marketplace trust and reputation. Rent/integrate: payments (Stripe), models (frontier + open), motion control (robot OEMs), fleet communication standards, cloud, identity verification, teleoperation providers, lending capital, insurance balance sheets.
The previous wave made intelligence callable. The next wave will make physical capacity callable. Allometry governs the economic commitment behind every call.
Alternative positioning: Stripe settles machine commerce. NVIDIA powers machine intelligence. Robot makers provide execution. Allometry determines which work is worth doing.
Strongest future headline: The Work API for the physical economy. Supporting blurb: Allometry makes human, machine and vendor capacity queryable. Its decision graph prices and authorizes work, its router dispatches the right capacity, and its ledger verifies outcomes for settlement and financing.
Credible because it begins with today's operator pain and compounds into infrastructure for tomorrow's autonomous economy.
Allometry starts as margin intelligence for asset-heavy operators, helping them decide which work to accept, what to charge and whether they can deliver it profitably. Each commitment and completed job strengthens a self-correcting decision graph and attested operating ledger. That foundation expands into the Allometry Index, a machine-readable map of human, vendor and machine capacity, and ultimately the Work API through which physical work is discovered, priced, authorized, executed, verified and financed.
Strategic caution (standing): the customer never wakes up wanting a Work API. They wake up asking: why did this job lose money, can we quote this, do we have the crew, what should we charge, which customer deserves the capacity. Solve those repeatedly; the Work API emerges from the accumulated decision infrastructure.
New developments; core thesis intact. Naming discipline applied throughout: Atlas → the Allometry Index. §19–23 of this addendum consolidate with Part II (§22–27) — the Physical Work Call and Actor ontology are defined there; this part adds the modes and negotiation envelope.
Do not sell the agent. Sell the economically valuable work the agent completes. Purchasable units: verified target account · qualified opportunity · completed RFP response · approval-ready quote · recovered invoice · collected receivable · prevented supplier overpayment · profitable capacity filled · acquisition target underwritten · commercial decision executed. Begin as a managed outcome business; progressively automate underneath.
The two wedges, defined (not a pivot): the product wedge is margin-aware quoting and capacity validation (§2); the commercial entry offer is retrospective Revenue Recovery, which proves value quickly and funds deployment of the forward-looking product. Sequence: scan historical work → recover identifiable value → construct the operator's cost and margin model → activate the floor on live quotes → close the estimated-versus-realized loop. Revenue Recovery is the implementation and trust mechanism for the core product, not a separate company.
Recommended initial offer — Revenue Recovery: "Allometry finds revenue hiding between your contracts, operations and accounting systems, then helps convert it into invoices and cash." Recovers: completed-but-unbilled work, missed change orders, unapplied contractual escalators, usage above contracted limits, missing travel/equipment/fuel/rush charges, old renewal pricing, duplicate supplier payments, unclaimed credits/rebates, receivables blocked by missing evidence. Pricing: $5–15K deployment (credited toward annual platform) + 15–25% of verified recovered cash, defined attribution rules, success-fee cap. Expansion sequence: recover yesterday's revenue → close today's truth → protect tomorrow's quotes → direct future growth.
Prepaid economic-work credits ($25K / $50K / $100K packs) instead of seats — elastic commercial capacity, not another app to learn. Ladder: verified account $100–400 · qualified opportunity $750–2,000 · attended meeting $1,000–2,500 · completed quote $250–1,500 · complex RFP $1,000–5,000 · acquisition screen $2,500–10,000 · recovered cash 15–25% · incremental gross profit 3–10%. The closer to realized economics, the more it can charge.
Not 1,000 leads: "the 73 accounts most likely to buy, why now, what to offer, and the expected economics of pursuing them" — each account carrying profile, footprint, systems, buying committee, pain, triggers, opportunity value, cycle, win probability, recommended product and price, opening angle, confidence and evidence. Connects the Index directly to monetization.
Open-loop tools recommend; a closed-loop system observes → hypothesizes → simulates → acts/requests approval → measures → explains variance → updates the model → improves. The learning unit: decision + expected outcome + realized outcome + variance explanation.
The Hypothesis Ledger (new canonical object): the Attested Operating Ledger records what happened; the Hypothesis Ledger records why the company acted — hypothesis, facts, assumptions, evidence, expected result and range, confidence, threshold, action, measurement period, realized outcome, variance explanation, resulting policy change. Applies to quotes, hiring, pricing, advertising, expansion, equipment, launches, acquisitions, company formation. The operating ledger records the outcome; the hypothesis ledger records the reasoning; the decision graph connects them; the policy engine determines what can be executed.
Facts (invoices, payments, contracts) are source-backed; margin and cash are deterministic logic; forecasts and scenarios are probabilistic; constraints and authorization are deterministic policy applied to uncertain evidence; learning is outcome-based calibration. AI may estimate whether an acquisition succeeds; it may not silently redefine the hurdle rate, invent accounting truth, or exceed delegated authority.
Deterministic when known · specialized when repeated · open-weight when private or frequent · frontier when novel · human when consequential. The durable asset is not the model — it is company context, economic ontology, decision history, policies and outcome evaluation.
Record prediction and outcome → detect recurring variance → propose change → back-test → shadow mode → compare vs incumbent → approved promotion → rollback and version history. Improvement at three levels: decision (better next estimate), company (how this operator scales, breaks, produces cash), network (permissioned, privacy-preserving priors).
The initial product is ~80% reaction, 20% planning. Every event (new RFP, lead, cost move, freed capacity, completed work, missing invoice, overdue payment, renewal, margin variance, acquisition opportunity) becomes an economic action queue: detect → explain → recommend → prepare → approve → execute → learn. Planning sets the field. Reaction plays the game.
The Node (new canonical object): the sovereign economic computer for one company — private operating graph, company economic model, Attested Operating Ledger, Hypothesis Ledger, policies, specialized models, model router, local memory, MCP/API tools, approval system. Deployed in our cloud, customer VPC, or on-prem; purpose-built hardware only when customers require local inference or sovereign operation. The Index: the permissioned intelligence layer across a market/portfolio/network, receiving standardized authorized outputs from Nodes without centralizing raw data — demand, pricing dispersion, capacity scarcity, benchmarks, payment behavior, acquisition targets, capital needs, intervention performance. One sovereign Node per company. One compounding Index across the network.
Node + Index for PE firms, independent sponsors, holding companies: underwriting → ownership → intervention → realized value → improved underwriting. Before acquisition (map, rank, reconstruct profitability, estimate integration, compare acquire/augment/rebuild/partner/spawn); during diligence (normalize records, validate claims, find leakage, measure owner dependence, score data readiness); after (deploy a Node, encode the thesis, track the 100-day plan, measure realized value, update the Index). A living investment-thesis ledger, not a static IC memo — underwritten vs actual on retention, pricing, synergies, cross-sell, DSO, schedule.
The capital-and-effort allocation engine: detect unmet demand and ask "what business should exist here, and what is the most efficient vehicle for creating it?" Options: spawn AI-native · acquire and scale · acquire and add AI · acquire and rebuild · partner/license · wait · ignore — scored on demand, willingness to pay, competition, distribution, cycle, unit economics, build complexity, capital, founder fit, adjacency, downside recovery, nonlinear upside. Founder curiosity stays an explicit variable: 70% evidence-led experiments, 30% founder-led asymmetric exploration.
The stack: Operator runs and improves a company · the Index maps demand, operators, capabilities, gaps · Foundry decides build, buy or transform. Loop: the Index discovers unmet demand → Foundry chooses the vehicle → Operator runs the company → outcomes improve the Index.
Do not compete with Merge.dev on connectors; normalize economic meaning, decisions, policies, actions, outcomes. Connect (ERP/CRM/FSM/accounting/ecommerce/banking) → Graph (canonical objects) → Act (governed tools): allometry.check_margin · price_quote · check_capacity · find_unbilled_work · prepare_invoice · rank_accounts · simulate_acquisition · request_approval · get_company_pulse. Claude, ChatGPT, Slack and vertical apps become interfaces; Allometry remains the trusted economic context and authorization layer. Give every approved agent access to the company's economics without giving away the company's data.
Approval cards in Slack/Teams before a standalone dashboard: value, expected margin, downside margin, capacity, payment risk — Approve | Reprice | Review assumptions. Every interaction teaches what management approves, which exceptions matter, which assumptions get modified, what autonomy the company accepts. The application becomes the control room; the work occurs inside the systems customers already use.
Allometry Profit Guard — "know what every order actually earns before you discount, fulfill or advertise it" (contribution after COGS, CAC, shipping, fulfilment, duties, fees, discounts, returns, support, expected LTV). Products: Order/Discount Guard, SKU Morphology, Ad Profit Router (send contribution-margin-adjusted conversion values back to Meta/Google instead of revenue — the strongest product), Inventory Capital, Returns Intelligence, audiences, agentic merchandising. Pricing 99–5K/mo + 3–10% of verified incremental profit. Parked per the not-do list ("don't broaden GTM beyond asset-heavy too early") — a Foundry-era or post-A wedge.
Factory spans the agentic software-development lifecycle; Allometry is the agentic quote-to-margin lifecycle. Copy the architecture (one shared context, specialized agents, adjustable autonomy, model routing, local+cloud execution, existing tools as interfaces, production outcome measurement, self-observation) — not the "Droid" character language. Operators want one accountable economic service.
Human-to-agent ("price this RFP and prepare the approval package") → agent-to-agent (buyer agent queries a supplier Node: capability, capacity, indicative price, delivery confidence, terms, approvals, attestations) → autonomous execution within policy (discover, qualify, quote, negotiate, reserve, commit, finance, dispatch, verify, invoice, settle, update). Agentic workflow helps a human; agent-to-agent lets systems coordinate; autonomous commerce lets them commit within delegated authority. Allometry is the economic policy and authorization layer across all three.
Negotiable variables (price, timing, quantity, service level, terms, sequence, reservation, financing, insurance, cancellation) inside a company-defined economic envelope (minimum margin, max discount, customer risk, capacity, working-capital exposure, required return, approval threshold). Agents negotiate inside the envelope; they cannot change the envelope. Models negotiate. Policy governs. Humans delegate the authority. Every proposed and accepted commitment lands in the decision graph.
The Attested Operating Ledger provides forward-looking evidence accounting lacks: profitable backlog, verified capacity, estimated-vs-realized margin, payment history, contract quality, utilization, change-order behavior, concentration, cash-conversion timing, execution reliability. The general ledger reports what happened; the AOL explains what was promised, whether the operator could deliver, why it was priced that way, what was consumed, completed, invoiced, collected, and whether the economics repeat. Underwrite individual commitments, not just the historical balance sheet.
Financing at the moment of commitment: when approved profitable work meets a capital constraint (e.g., $600K contract at 31% contribution needing $180K before first payment), the Node generates the financing package (verified contract, approved quote, cost schedule, capacity evidence, payment history, downside case, conversion date, repayment source) and approved lenders compete for that job. Products: contract/PO/invoice financing, working capital, leasing, RaaS financing, capacity expansion, dynamic insurance, acquisition, RBF, supply-chain finance — increasingly granular: company → business unit → contract → job → asset → a machine's next work call. Capital becomes callable at the same moment profitable capacity becomes callable. Lender canon still applies: relationships, not automated flows, until the ledger earns the rails.
Demand detected → buyer agent issues a Work Call → operator Nodes respond → Allometry calculates price, margin, capacity → agents negotiate inside policy → customer and work underwritten → financing attached if required → commitment authorized → human, agent or machine executes → completion attested → invoice and payment triggered → ledger records realized economics → company and network models improve. Discover → price → authorize → finance → execute → verify → settle → learn.
Integrated: Allometry builds a sovereign economic model of a company, reacts continuously to demand, cost, capacity and cash signals, and turns them into governed commercial actions. Every action is reconciled against its realized outcome. One Node operates each company; the Index learns across a permissioned network. The same infrastructure can recover revenue, price work, operate acquisition theses, direct capital, and eventually coordinate autonomous physical capacity.
Shortest: Allometry turns economic signals into governed actions, then learns from what actually happens.
Product architecture: Node operates the company. The Index understands the market. The Work API makes its economics callable. Foundry decides what should exist next.
Commercial entry: Recover the revenue hiding between contracts, operations and accounting. Pay when the money is recovered.
PE version: Underwrite the thesis. Operate the integration. Prove the value creation.
Infrastructure analogy: Stripe provides financial settlement. Allometry provides economic authorization and operational attestation.
Financing line: Finance the work, not merely the company.
The complete product system: Node · Index · Work API · Decision Graph · Hypothesis Ledger · Attested Operating Ledger · Financing Layer · Foundry.
Ultimate thesis: A self-correcting economic intelligence network for building, buying, financing and operating the businesses of the physical economy.
CME Group and Silicon Data plan to launch H100 and B200 Rental Index Futures on October 5 2026, pending regulatory review — compute becoming a financially managed input. The structural lesson: CME trades contracts on Silicon Data's index; the exchange rents the benchmark. Silicon Data CEO: "two companies buying the exact same GPU capacity could pay wildly different prices with no way to know who got the better deal" — which is operator quoting today, verbatim. Even nominally identical H100 capacity varied up to ~34.5% across providers: quantity without grade is not fungible.
The deep insight: a scarce operational resource becomes financeable and tradable only after someone makes its quality, availability, price and delivery verifiable. Commodity markets were built on grades — No. 2 yellow corn made corn tradable; WTI made oil tradable. The Work Grade makes capacity tradable. One technician-hour is not equivalent to another (trade, certification, geography, equipment, crew, quality, speed, reliability, insurance, complexity, mobilization, history) — exactly as one GPU-hour is not another.
The legibility ladder, completed: legible → callable → financeable → tradable. Neither capital nor capacity becomes callable until capacity is graded. Extends the banked legibility frame: the ledger makes operators legible to capital; the grade makes capacity legible to markets.
The standardized Work Grade (what a Physical Work Call resolves into): capability + region + time window + capacity volume + quality rate + schedule reliability + insurance + expected cost range + minimum commitment + verification standard. Supports price comparison, reservation, long-term contracts, financing, insurance, agent purchasing, and eventually forward markets.
The mapping: GPU-hour → crew/machine-hour · benchmark performance → attested operating performance · compute price index → work and capacity price index · futures contract → forward capacity commitment · Silicon Data → the Allometry Index · cloud provider → operator Node · compute delivery → verified job completion.
Strategic restraint (canonical): become Silicon Data before CME. Do not pitch futures — build the benchmark a future market could not operate without. The staged path: Measure (normalize job types, realized cost, delivery, reliability, quoted-vs-realized, provenance) → Index (permissioned price ranges, capability grades, regional capacity, forward-looking benchmarks) → Contract (standardized Work Calls, bilateral reservations, take-or-pay, attested delivery) → Finance and insure (materials/labor against commitments, equipment against utilization, performance insurance, capital against contracted capacity) → Market (matching, forward commitments, netting, potentially with regulated operators). Observation → standardization → benchmarking → contracting → financing → market formation. Same restraint pattern as §13 (intelligence layer before balance-sheet lender).
Sub-product named: the Allometry Capacity Index (beneath the Index): available capacity by capability/region/time, realized cost per standardized work unit, current and forward price, backlog and utilization, reliability and quality, demand pressure, working-capital needs, constraints. One operator: better quoting and scheduling. PE: portfolio allocation and underwriting. Lenders: whether financed capacity has credible profitable demand. Agents: price discovery. Gate: the no-benchmarks-until-attested commitment (10 operators × 12 months) governs any published index — the ambition is the reason the discipline exists.
Near-term forward products (into §46's financing list, all pre-futures and bilateral): capacity reservation · take-or-pay · minimum-margin commitment (supplier commits capacity only if terms preserve the floor) · contract-backed equipment financing · portfolio capacity facility.
The PE precursor: a sponsor with twenty field-service companies runs an internal capacity market — one portfolio company's excess demand becomes another's available work, routed by the Index before any forced operational merger. The best practical rehearsal for the external Work API; belongs in the PE OS (§37) and the HoldCo path.
Signal vs hype, kept honest: signal — compute as financially managed input, benchmarks before financing, quality variance rivals quantity, index providers occupy value without owning the resource. Unresolved — notional "trillions" ≠ economic value, hardware churn undermines fungibility, concentrated supply weakens discovery, futures fail without two-sided natural hedgers, and physical work is less fungible than compute, so grading is materially harder here. That difficulty is the moat: whoever does the grading owns the market's vocabulary.
Banked formulation: Allometry makes physical capacity legible before making it callable. The Node attests the quality and economics of each operator's capacity. The Index standardizes comparable work across the market. The Work API lets that capacity be queried, reserved, financed and ultimately coordinated. The stack: Node = operator truth · Index = price and performance discovery · Ledger = delivery and economic attestation · Work API = contracting and execution · Financing layer = forward capital · Market = eventual capacity exchange.
A company that sells governed commitments should govern its own. This document is a hypothesis, and §31 defines what a hypothesis owes: facts, assumptions, expected outcomes, falsifiers, a measurement period, and a variance discipline. So the fiftieth section is the ledger entry for the first forty-nine.
Hypothesis. A self-correcting economic intelligence network can be built for the physical economy, entered through margin-aware quoting for asset-heavy operators, and compounded through the Node, the Index, the Work API and ledger-powered capital.
Known facts, measured. Analysis of 90 days of one design partner's completed work surfaced approximately $300K in annualized margin leakage patterns. Partner-run exploratory conversations number in the dozens and growing, with booked evaluations on the calendar. Day one of founder-led outbound touched 42 of a curated 1,000. And in a single month — August 2026 — the surrounding stack arrived: $1.1B raised for physical supply, a major model provider released a research preview of a hardware interoperability standard for agents operating equipment, $80M for long-horizon agent inference, and a regulated futures market announced on a rented compute benchmark. Four layers financed or standardized in thirty days. The commitment layer stayed empty.
Assumptions, stated. That operators will pay before the category has a name. That an enforced floor creates switching costs faster than incumbents can descend. That physical work can be graded at all. That agentic procurement arrives within the decade, not the century.
Expected outcomes, dated. Floor live in production at the design partner in September 2026. One thousand operators contacted by September 15. Three to five paying operators by December, against ten slots offered. No published benchmark until ten operators and twelve attested months stand behind it.
Falsifiers — how this dies. Operators stay single-product and attach fails. Deployment never compresses and the founder remains inside every implementation. An incumbent descends before the ledger compounds. Attribution disputes poison outcome pricing. The grade proves impossible — physical work resists standardization and the Index never clears the noise. Or capital arrives too late for the default-alive path to matter.
Revision entry — v1.1, Aug 30 2026. Identity resolved (strategic planning that executes; per-template wedge; pricing = the asset-heavy door) and Shape graduated, on founder ruling and the improves-GTM criterion. Hypothesis attached: cross-vertical channel demand (food, flavors, cosmetics, logistics already on the calendar) will validate the template-wedge structure before any second template launches. If those engagements fail to convert, the graduation reverts to doctrine.
Variance discipline. This section is reconciled against reality quarterly. Revisions are logged, never silently overwritten — the document keeps its history the way the platform keeps its versions: evidence, shadow, approval, rollback. When outcomes diverge from expectations, the explanation lands here before the story changes anywhere else.
A company that asks operators to commit at the moment of truth should be willing to live at its own. This page is that moment, dated and signed.
Allometry governs the moment a business commits — beginning with this one.
— Taylor Gendron, Founder · August 2026
Gates, explicit: No network product before repeatable Nodes. No benchmark before attested density. No financing product before predictive evidence. No market before standardized bilateral commitments.
Parked for the next 12 months (in the thesis, out of the roadmap): exchange-traded capacity futures · Allometry-manufactured hardware · Ecommerce Profit Guard · Allometry-owned lending capital · external autonomous negotiation · public capacity benchmarks · Foundry-created companies · open marketplace transactions.
Build now (the only build order): historical contract/job/invoice reconciliation → Revenue Recovery queue → deterministic job-cost and margin engine → the floor on live quotes → estimated-versus-realized close → Slack approval cards → first Node data model → five narrowly defined MCP actions → deployment automation → evidence-backed ROI report. Everything else emerges from these primitives.
Derived documents (each generated from this source, never the reverse): one-page company thesis · 10-slide fundraising narrative · initial product requirements document · technical architecture brief for a founding engineer.
Sources for external claims: a16z Machine Age Fund — a16z.com/the-machine-age-fund (.1B, announced Aug 28 2026). Anthropic Model Hardware Standard — anthropic.com/news/model-hardware-standard-research-preview (research preview, Aug 27 2026; open sourcing planned). CME × Silicon Data compute futures — cmegroup.com press release Aug 11 2026 (planned Oct 5 launch, pending regulatory review). Sail Research — sailresearch.com/blog/sail-raises-80m. GPU performance variance (~34.5%) — per Silicon Data benchmarking as cited in secondary coverage; treat as directional.
The next revision of this document will be earned by evidence, not analogy: what data is consistently available, what recovery customers pay for, which calculations they trust, which actions they authorize, whether deployment compresses, whether predictions improve after outcomes.
— Taylor Gendron, Founder · Montréal | New York · taylor@allometry.com