Allometry's 16 modules aren't 16 operational tools. They're 16 evidence guardrails. Each module enforces the data shape, time-stamp, and audit trail that a specific class of capital provider needs to underwrite. Operators adopt them because they fix urgent ops pain — margin leak, dispatch chaos, churn. The byproduct is structured, hash-chained, unfakable.
▸ 3 months in the vault → basic visibility · 6 months → working-capital lenders · 12 months → infra debt · 24 months → insurance-linked capital · 36 months → rated paper. Lenders connect via API and consume the vault. Same shape Plaid took for consumer fintech — for the physical economy.
Every module enforces a specific data shape and depositing rhythm. Operators experience them as ops tools. Lenders experience them as evidence standards. Below: what each guardrail deposits, and which lender tiers will eventually consume it.
▸ Each module generates two outputs: an operational action for the operator, and a time-stamped evidence deposit that feeds the Pulse underwriting model (next section). The table below shows the deposit and the lender tier it eventually unlocks — not the model inference itself.
| # | Module | Operational job (the wedge) | Vault deposit (evidence produced) | Tier unlocked |
|---|---|---|---|---|
| M.01 | Quoting Engine | Win quotes faster | Pipeline visibility · win-rate · deal velocity · forward revenue | T1 |
| M.02 | Margin Protect | Stop margin leak | Margin stability · override discipline · price-floor enforcement history | T1T2 |
| M.03 | Cost Engine | Track unit cost live | Unit economics · cost-to-serve · input volatility · gross-margin trace | T1T2 |
| M.04 | Contract Intelligence | Read every contract | Counterparty quality · renewal probability · liability exposure · clause-level decomposition | T2T4 |
| M.05 | Location Intelligence | Score addresses | Geographic risk · concentration · address-level density · weather-correlation surface | T3T2 |
| M.06 | Job Costing | Track every job | Estimated-vs-actual variance · gross-margin history · cost-input drift | T1T2 |
| M.07 | Deployment Design | Optimize schedule | Capacity utilization · time-series of throughput · forward capacity commitment | T1T2 |
| M.08 | Active Operations | Live crew tracking | SLA compliance · operational performance · uptime · claim-incidence proxy | T3T2 |
| M.09 | Deal Room | Pipeline mgmt | Conversion rates · forward booked revenue · stage velocity · cohort retention | T1 |
| M.10 | Outbound ABM | Target accounts | Demand generation efficiency · CAC stability · channel ROI | T1 |
| M.11 | Customer Health | Account scoring | Churn risk · account LTV · sentiment–payment correlation | T1T2 |
| M.12 | Account Substrate | Unify sources | Data completeness · stakeholder graph · governance maturity index | T4T1 |
| M.13 | Market Intelligence | External signals | Macro alignment · concentration risk · demand-curve fit | T2T4 |
| M.14 | Control & Governance | Policy enforcement | Compliance posture · audit-readiness · evidence-completeness score | T4T3 |
| M.15 | Inventory Management | Stock levels | Asset visibility · working-capital cycle · stockout exposure | T1T3 |
| M.16 | Pulse Workflows | Workflow agents + audit ledger | Hash-chained decisions · WORM evidence trail · tamper-evident provenance | T4T3 |
The vault deposits don't go to lenders raw. They feed the Pulse Engine — a RAG-augmented, recursively-updated underwriting model that converts time-stamped evidence into an attested operator fingerprint. The API doesn't serve data. It serves inferences with cryptographic provenance. This is what separates us from a passthrough fabric like Plaid.
The model pulls historical evidence from the operator's vault — last quarter's margin variance, three years of route economics, every contract clause, every audit log — and assembles a contextual frame for inference.
An LLM reasoning layer reads the retrieved evidence and extracts the underwriting signal: revenue quality, margin defensibility, counterparty risk, operational maturity. Reasoning is structured, traceable, and explainable.
Each new deposit doesn't just trigger inference — it updates the model's understanding of that operator's cohort, vertical, and address-level pattern. The system relearns continuously. Yesterday's score is not today's score.
Every inference output carries a zero-knowledge proof: the lender can verify the score was computed from the operator's actual vault data, with the published methodology, without seeing the raw underlying data. Cryptographic provenance.
// Pulse Engine inference · attested · cohort-aware
{
"operator_id": "ALM-0042",
"pulse_score": 87, // composite 0–100 underwriting score
"confidence_interval": [83, 91], // 95% CI · widens at low vintage
"revenue_quality_tier": "A-", // A / B / C / D · margin defensibility
"ebitcac_ratio": 4.2, // EBITDA / CAC · efficiency signal
"vault_tier_eligible": ["T1", "T2"], // lender tiers unlocked today
"cohort_percentile": 0.78, // vs. comparable operators · vertical-aware
"vault_age_months": 14, // vintage · gates which tiers are eligible
"loop_coverage": {
"price": 0.94, // % of guardrails active in each loop
"deploy": 0.88,
"expand": 0.71,
"govern": 0.92
},
"last_attested": "2026-05-13T14:22:11Z",
"attestation_hash": "0x4F2A9C...A7E1", // ZKP proof · verifiable on-chain
"methodology_version": "pulse-v2.4.1" // model version that produced this inference
}Allometry is a pre-seed company. Pretending we've shipped everything would destroy credibility. The thesis above is the destination — and the maturation path is staged honestly. Lenders evaluating us should know exactly which row of this table they're contracting against.
The Pulse API response below is from our flagship anchor design partner. Identity, industry, and geography confidential; metrics confirmed with the operator. This is the kind of structured profile a balance-sheet partner pre-screens against today, and the kind of profile that thickens into Tier-2 readiness over the next year.
The Pulse profile is the input. The output is a ranked, match-scored shortlist of capital partners whose mandates fit your stage, KPIs, and ZKP-attested vault. Bank-style DD takes six weeks per lender. Pulse-driven matching closes the same loop in ~200ms — and emits a cryptographic attestation the lender can verify before they decision.
You can't fake six months of hash-chained job-level margin. You can't backfill a year of route economics. The moat is time-stamped, audit-trailed, and unfakable. Each month an operator runs on Allometry, the vault thickens — and the next tier of capital becomes consumable.
Allometry is the operator's day-to-day system of record for quoting, dispatch, contracts, and audit. Every datum is a byproduct of work the operator was going to do anyway. There's no separate "underwriting reporting" workflow — there's no opportunity to fabricate.
▸ first-party, by constructionM.16 (Pulse Workflows) writes every decision into a WORM ledger with cryptographic time-stamping. You can't backfill margin you didn't have, route economics you didn't run, or compliance you didn't perform. The chain breaks if you try.
▸ tamper-evident provenanceThe Vault exposes a standardized API. Pre-approved capital partners query an operator's vault directly, see the time-stamped evidence, and underwrite in seconds — not the weeks it takes for traditional DD. Plaid's shape, applied to asset-heavy capital.
▸ underwriting fabricThe first question any sharp lender or investor asks. Here's the honest comparison against the four things that look adjacent.
| Bank DD | Plaid | ServiceTitan / FSM | Allometry Vault | |
|---|---|---|---|---|
| Data shape | unstructured packets | bank txn feed only | ops record | underwriting-grade · structured |
| Generation | manually requested | API passthrough | byproduct of ops | byproduct + attested |
| Time horizon | snapshot | rolling 90 days | rolling | full vintage · multi-year |
| Hash-chained | no | no | no | yes |
| ZKP-attested | no | no | no | yes (Stage 2) |
| Lender-API consumable | no | partial (consumer fintech only) | no | yes |
| Asset-heavy operator coverage | case-by-case | no | vertical-specific | native |
| Time-to-decision | ~6 weeks | n/a | n/a | ~200ms (Pulse inference) |
▸ The defensible position: hash-chained, attested, lender-API-consumable, asset-heavy-native — and the only system that's all four at once. ServiceTitan and FSMs can't add the vault layer without disrupting their system-of-record business model. Plaid won't ever expand into asset-heavy ops data. Banks can't industrialize bespoke DD.
This thesis didn't work in 2020. It works now because four independent maturity curves crossed inflections in the last 18 months.
Structured reasoning over heterogeneous evidence is no longer research. We can run an LLM over a 14-month vault and extract underwriting signal at <100ms per inference. Five years ago this required custom ML infra; today it's a primitive.
Risc0, Aleo, Halo2 and zk-rollup tooling are now mature enough that proving an inference cryptographically — without revealing the underlying data — is shippable in production. This was a research project two years ago. Now it's a deployment pattern.
Climate-aligned and energy-transition capital is chasing novel underwriting signals. Pension funds need long-duration BBB-equivalent paper; reinsurers need asset-performance risk data they currently underwrite blind. Demand exists; supply infrastructure doesn't.
Banks pulled back from middle-market asset-heavy lending post-SVB. Fintech ignored field services and physical-revenue operators in favour of consumer credit. The wedge is wide and widening, and the data infrastructure to underwrite this population doesn't exist yet.
Allometry is two businesses operating under one trojan horse. The SaaS layer is how we acquire operators and generate the data. The Vault API is how that data eventually monetizes into the multi-billion underwriting infrastructure play.
Per-seat · per-module · sold today
Per-query · per-facility · rev-share on capital deployed · sold tomorrow
The data isn't entered — it's generated by operations. When an operator quotes through Allometry, dispatches through Allometry, and signs contracts through Allometry, the vault deposits are byproducts of work that had to happen anyway. There's no separate "underwriting reporting" workflow to gerrymander.
On top of that, M.16 writes every decision into a WORM ledger with cryptographic time-stamping. Backfilling a margin you didn't have, or a route you didn't run, breaks the hash chain — and the chain break is publicly verifiable by lenders consuming the API.
The operator owns the vault. Full export, full portability, no lock-in on the data itself. If you churn off Allometry, you take your vault with you.
What we license to lenders is read-access against an operator's vault while it's active. If the operator goes dormant, the vault stops thickening, and the eligible-tier signal in the Pulse API will reflect that. We never sell or share an operator's vault without their explicit per-lender consent.
They could — but they won't, because the vault layer requires re-architecting at the system-of-record level, and that would disrupt their existing business model.
Their data is structured for ops reporting, not for underwriting. Their decisions aren't hash-chained. Their data isn't third-party-validatable. Their architecture optimizes for "what should the operator see in the UI today" — not "what should a lender be able to verify three years from now." Adding our layer means redoing their write-path on every module, which costs more than they'd ever recover. We start from the underwriting requirement and work backward.
Honest answer: in pilot, not production. Today, vault deposits are hash-chained (real) and time-stamped (real). The ZKP attestation layer that proves "this inference was computed from this operator's actual vault data with the published methodology, without revealing the underlying data" is in Stage 2 of our model maturation roadmap.
We've benchmarked Risc0, Aleo, and Halo2 against our inference workload. Production rollout is 12–18 months out, gated by anchor-lender pilot demand. We'll never claim "ZKP-attested" on a production API response until the proof is real and verifiable on-chain.
The API spec isn't the moat. The vault is — and vaults are time-locked. A competitor can copy the GET /v1/operators/<id>/profile shape in an afternoon. What they can't copy is 14 months of hash-chained job-level margin from an operator who's been running on their system that whole time.
The deeper moat: two-sided network. Operators choose Allometry because the lender side is pre-integrated. Lenders choose Allometry because the operator coverage is wide enough to underwrite a portfolio against. Each side strengthens the other. A late entrant has to bootstrap both at once.
We're solving the smaller problem first. Plaid started by selling consumer-fintech bank-account access — a data shape lenders had never trusted via API before, and one that took years of compliance work to legitimize.
Our Stage 1 is Tier-1 working capital — where the data requirements are lightest (6 months of cashflow, contracted revenue, audit log) and where balance-sheet banks are actively asking for novel underwriting signals to extend cleaner credit lines post-2023. We're not asking Munich Re to underwrite a parametric pool on day one. The harder tiers come after we've operationalized the easier ones.
They don't have to trust us. They have to trust three things we don't own:
(1) Cryptographic proof. ZKP attestations are mathematically verifiable — not "trust us, here's our score" but "verify the score yourself against the published methodology." (2) Independent validators. Big-4 audit and equivalent independent engineering partners sign attestations against the same vault data lenders consume. (3) Open methodology. The Pulse model version, scoring weights, and inference logic are published. Lenders evaluate the methodology, not the company.
The point of the architecture is that institutional trust doesn't require trusting Allometry as a company. It requires trusting the math + the audit + the methodology. Those are all third-party-verifiable.
▸ The Vault is the architecture. These three are how the architecture earns value.
Watch the Pulse Engine ingest 8 sources, attribute each signal back to its loop, and continuously re-rank accounts by address × LTV.
/pulse →Operators with vintage'd vaults receive cryptographically-attested capital pre-approvals — capital that lands in days, not quarters.
/capital-request →The 16 modules operators adopt to fix urgent ops pain. Underneath each one: a vault deposit they didn't know they were making.
/operating-system →If you're an operator: we'll close your urgent loops. If you're a capital partner: we'll be your underwriting fabric. Either way, the vault thickens.
Open a vault →