01 / Project overview

An ALM platform for a liability that has no matching asset.

CaptureNow is an asset and liability management platform for durable carbon removal. It measures a corporate's residual emissions liability, vetted removal assets through a three-proof gate, structures a bilateral financial instrument, and manages the book over a 10–25 year horizon. Built on Helium's orchestration layer with a new deterministic computation layer.

The liability
Forward tCO₂e
Scope 1, 2 and 3 residual emissions to net-zero year. Mandatory to disclose under CSRD, ISSB/IFRS S2, SSBJ.
The instrument
CLS / NRW
Carbon Liability Swap or Net Removal Warrant. A claim on custodied verified removal tonnes, structured under ISDA.
=
The metric
Carbon Solvency Ratio
Custodied NRTs divided by forecast residual liability. The platform's headline dashboard number.
The structural problem

A long-dated, regulated liability meets a short-dated market

Companies carry a growing clean-up debt — tonnes they will be legally required to remove by their net-zero year. Today it is treated as an annual expense: buy cheap offsets, retire them, book the cost. New rules are forcing a shift toward long-lived removal assets held on balance sheet.

Liability side

Mandatory, growing, long dated

Every large manufacturer and exporter carries a forward emissions liability. CSRD, ISSB/IFRS S2, SSBJ, SBTi V2.0 and CBAM are converging to make holding durable removal assets mandatory — not optional.

Market side

Short, low durability, no balance sheet fit

The voluntary carbon market trades mostly short, avoidance-based credits at registry face value. No standardised, duration-matched, balance-sheet-recognisable removal asset exists to stand against the liability.

The 2028 inflection

Demand becomes forced

SBTi V2.0 raises the durable removal share of neutralisation. CORSIA Phase 2 and CSRD assurance converge around 2028. Demand shifts from voluntary to regulatory-mandated.

Dates to verify
Border price gap

CBAM arbitrage delta

A structural spread between domestic carbon price and EU CBAM border cost forms part of the floating leg economics. The platform computes this delta per credit and carries it into fair value and lease rate.

What the platform delivers

Five capabilities, end to end

Measure
Model the residual carbon liability to net-zero year across multiple scenarios.
Vet
Run every removal credit through a three-proof gate — science, legal, financial — to mint a Net Removal Tonne.
Match
Pair certified NRTs to liability tranches across duration, durability, geography, lot size and cost.
Structure
Generate a draft CLS/NRW instrument under ISDA with bank-standard FpML messaging.
Manage
Daily mark-to-market, margin, reversal detection, cure workflow, and disclosure packs.
The one rule

The orchestrator sequences — it never decides

Foundational constraint

Every routing, proof-gate and matching decision is a deterministic Python function in the computation layer, reproducible from state alone. No AI model makes a routing, gate or matching decision. Break this and the audit story breaks.

Enforcement is not a comment in a doc — it is a CI test gate. Every decision function carries an fca_critical marker, is 100% unit-tested, and CI blocks merge if any test fails.

02 / Proposed solution

Score the asset, certify it, match it, manage it.

The platform sits above registries, brokers and marketplaces. Its job is to convert a spot-traded, registry face value carbon position into a duration-matched, balance-sheet-recognisable asset book and keep it solvent against a moving liability over 10–25 years.

Signature capability

The three-proof gate

No asset can be matched or placed into an instrument until it independently clears all three tests. Each produces a signed certificate that travels with the asset for its full lifecycle.

Three-proof gate · signature capability
PROOF A

Science

Confirms actual net removal quantity — not registry face value. Includes measurement, audit status and a durability tier.

PASS · score ≥ 75 · no adverse audit · no open reversal
PROOF B

Legal

Confirms clean title, registry lock-out, defined cure obligation, adequate replacement pool and jurisdictional permissibility.

PASS · clean title · instrument eligible · buffer adequate
PROOF C

Financial

Establishes fair value at inception, IFRS-9 accounting classification, lease rate and CBAM border delta.

PASS · fair valued · classified · lease rate set
Output → Proof-ready NRT inventory Signed, hash-chained, independently verifiable. Rejections recorded, never silently discarded.
Core building blocks

Asset, instrument, structure

Unit of account

Net Removal Tonne (NRT)

A tonne verified by measurement, confirmed by audit, durability-tiered, held without retirement, legally clear and fair-valued. Also framed commercially as a Net Removal Warrant (NRW).

SpecifiedTerminology to reconcile
Primary instrument

Carbon Liability Swap (CLS)

A 10-year bilateral derivative, extendable to 25 years, under ISDA with a bespoke Carbon Product Supplement. Fixed leg: lease rate by durability tier. Floating: permanent removal price index.

Mechanics specifiedLegal supplement pending
Operating shape

Three balance sheets

Client treasury deposit stays at its bank. Cure reserve sits in a ring-fenced Durability Trust vehicle. The fee-earning platform stays capital-light — a manager, not a principal carrying last-resort risk.

Model definedConsolidation opinion pending
The asset spectrum

Four durability tiers

Tier
Permanence
Indicative lease · min score
Tier 1 · Geological
Mineralisation, deep geological sequestration
1,000+ years
6.5–8.0% pa · ≥ 90
Tier 2 · Durable
BECCS with geological storage, direct air capture
100–1,000 years
5.5–6.5% pa · ≥ 80
Tier 3 · Long-lived
Biochar, enhanced weathering
50–100 years
4.8–5.5% pa · ≥ 75
Tier 4 · Transitional
Jurisdictional, sensor-verified, sovereign-backed
20–50 years
4.2% pa · ≥ 75
Product areas

Seven platform capabilities

Area 01

Liability workbench

Connect a corporate system, ingest GHG inventory, model the full liability to net zero, compute residual exposure, and display the live Carbon Solvency Ratio.

Specified
Area 02

Asset intelligence

Continuously refreshed, asset-level view of every monitored removal credit — registry status, removal score, durability tier, title and forward price, reversals flagged in near-real-time.

Specified
Area 03

Proof certification

The science, legal and financial gate producing signed certificates, a rejection register and a public verification endpoint for counterparties.

Specified
Area 04

Matching and structuring

Optimised liability-to-asset matching across five dimensions. Draft instruments with alternatives. Mandatory human approval before execution.

SpecifiedObjective function pending
Area 05

Lifecycle and cure

Daily valuation and margining, automated reversal detection, 90-day cure workflow, buffer pool monitoring and escalation when replacement is not found.

Specified
Area 06

Disclosure and reporting

One-click ISSB and IFRS S2 packs, board reports, border reports and certificate registers generated from platform data.

Specified
Area 07

Forward strategy

Client-specific guidance on optimal instrument size, cheapest tier mix satisfying regulation, and when to settle versus roll — in plain language and as structured data.

Spec is a draft
03 / Architecture

Two layers, three sides, one matching core.

Helium's orchestration layer is the conductor — it sequences workflows, holds durable state, inserts human approvals, calls AI models and pulls data from connectors. The computation layer is the new carbon-specific brain — deterministic math only, no model in any decision path. Together they drive 17 modules across three sides.

The two layers

Helium orchestration over new computation

Layer 1 · Helium orchestration — reused

The conductor

Sequences multi-step workflows, holds durable long-running state, inserts human checkpoints, calls AI models with fallback and cost control, pulls registry data via connectors, enforces multi-tenant RBAC isolation, and logs every step. Decides flow — never decides numbers.

Reused from HeliumLong-state spike needed
Layer 2 · Computation — new build

The calculators

Six deterministic engines: Liability Modeller, Three-Proof Scorer, Matching/Optimiser, Pricing & NAV, Margin & Collateral, Carbon Solvency Ratio. Plus regulated audit signing: immutable hash-chained INSERT-only ledger, JOSE/JWS certificate signing, and CASS client-asset segregation.

New buildRegulated audit — not reuse
Reuse framing

~40% of code volume may be reused from Helium (accounts, roles, sign-in, database, cache, API skeleton, RBAC). But only ~15–20% of the effort and risk. The cost and failure risk sit entirely in the new computation layer. Frame it as a large head start on plumbing, not “half the product.”

17 modules

Three sides and a matching core

Side 1 · Buy side

Liability ingestion

Bounded multi-agent group
BS-1
Liability Aggregator
Normalises corporate feeds to a structured liability ledger.
BS-2
Forward Modeller
Projects the liability curve to 2050 by scenario.
BS-3
Residual Calculator
Computes the irreducible notional. Deterministic. fca_critical.
BS-4
Obligation Classifier
Tags regulatory profile and minimum asset quality. LLM reads messy ERP data only.
BS-5
Solvency Ratio Calculator
Emits the live Carbon Solvency Ratio. Deterministic. fca_critical.
Side 3 · ALM engine

The matching core

Stateful orchestration · swap and lease book
AO-1
Three-Proof Orchestrator
Master quality gate. Issues proof bundles. Deterministic routing only.
AO-2
Matching Engine
Five-dimensional optimisation, liability to NRT. fca_critical.
AO-3
Swap Structuring
Generates CLS/NRW term sheets, machine and human readable. FpML output.
AO-4
Forward Strategy Engine
Scenario projection and client guidance.
AO-5
Collateral and Margin
Daily valuation, variation margin, cure triggers. fca_critical.
AO-6
Reporting and Disclosure
Dashboard, ISSB/IFRS S2 packs, certificate registers.
Side 2 · Supply side

Registry and producer intelligence

Deterministic data pipeline
SS-1
Registry Monitor
Polls Verra, Puro.earth, TGO and others. Detects reversals. Auditable asset lineage.
SS-2
Quality Scorer
Computes NRCS removal score with gradient-boosted deterministic model. fca_critical.
SS-3
Audit Integrator
Extracts VVB audit fields, reconciles quantities and adverse audit flags.
SS-4
Durability Atlas Builder
Assigns tier, half-life and reversal risk.
SS-5
Legal Title Verifier
Chain of title, lock-out, transfer eligibility. LLM for PDF parsing only.
SS-6
Pricing Intelligence
Forward price curves by tier and CBAM border delta.
End-to-end data flow

Seven steps from intake to live instrument

01
Client submits GHG inventory via ERP/treasury connector.
02
Buy side models residual liability and emits Carbon Solvency Ratio.
03
Supply pipeline produces a proof-ready NRT inventory.
04
Three-proof gate clears assets and issues signed certificates.
05
Matching engine pairs NRTs to liability tranches.
06
Term sheet generated → mandatory human approval checkpoint.
07
Instrument executes. Daily valuation, margin and reporting begin.
Data contract

Four data layers, one shared schema

Layer 1
Raw source
Validates registry and MRV inputs at the adapter boundary. Nothing crosses without validation.
Adapter boundary
Layer 2
Canonical
Normalises all inputs into typed, versioned records. Shared schema package is the single contract between subsystems.
Shared schema
Layer 3
Proof certificates
Issues signed science, legal and financial certificates and proof bundles. JOSE/JWS signing.
Signed + hash-chained
Layer 4
Swap book
Holds the instrument, daily valuation and margin records. INSERT-only immutable ledger.
Immutable · INSERT-only
Stack summary

Technology choices

Language
Python throughout, with strict typed schemas as the contract between every system boundary.
Orchestration
Stateful long-running orchestrator for the ALM engine, bounded multi-agent group for buy side, deterministic pipeline for supply side.
Durable state
Helium's runtime or a LangGraph + Postgres checkpoint engine embedded for 10–25 year instrument lifecycle. Spike required first.
ML models
Gradient-boosted trees for the deterministic removal score. Convex optimisation for the matching engine. No model touches a financial decision.
Trust layer
JOSE/JWS signed certificates, append-only hash-chained ledger. Mathematical tamper evidence — no blockchain.
Storage
Transactional store for records and ledger, NRT time-series store for prices, object storage for raw files and MRV documents. Never mixed.
Interfaces
JSON API for clients and internal calls. FpML/XML for derivative confirmation with bank counterparties.
Quality gates
Blocking fca_critical test marker. 100% coverage on all financial-critical modules. Property-based tests on scoring and residual logic.
04 / Delivery timeline

Five phases to a pilot-ready platform.

The software platform can be delivered in roughly five months by a dedicated squad. The build is sequenced so the reused orchestration lands first and de-risks the new computation work. Each phase closes with a working, tested increment. Legal, accounting and regulatory enablers run in parallel on longer external tracks and gate the move from pilot to live product.

Build phases · sequenced

Five phases in order

Phase
Deliverables
Helium reuse
Risk
Phase 0
Foundation
Stand CaptureNow on Helium: accounts, roles, sign-in, database, cache, observability, API skeleton, RBAC. Add corporate-client account type, regulatory tiers, maker-checker approvals. Spike: validate durable long-running instrument state before relying on it.
High plumbing reuse
State spike critical
Phase 1
Supply side
Registry connectors on Helium's connector framework. MRV time-series store with buffered ingestion. Auditable asset lineage pipeline (registry → normalise → proof-ready with provenance). Three-Proof Scorer: science, legal and financial vetting engines.
Medium reuse
New compute
Phase 2
Liability + matching
Liability Modeller. Matching/Optimiser (objective function must be authored and signed off before this phase). Orchestrated intake → proof → match workflow with human approval. Deterministic routing only — fca_critical test gate enforced in CI from day one.
Medium reuse
fca_critical path
Phase 3
Instrument + compliance
CLS/NRW instrument with FpML messaging. Pricing & NAV, Margin & Collateral, Carbon Solvency Ratio. AUM fee engine. Compliance workflows and reserve/cure-pool. Regulated audit: immutable hash-chained INSERT-only ledger, JOSE/JWS signing, CASS segregation — all new build.
Low / new build
Regulated audit new
Phase 4
Hardening
Client ERP/treasury API. Penetration-test the audit chain and signing. Load and scenario testing. Security review and independent audit trail verification. Pilot-ready handover on sandbox data.
Medium reuse
Pilot-ready
Month-by-month build

Sequenced from the data outward

Month 1
Foundations
Infrastructure, data contracts and the first inputs.
  • Stand up environments, pipeline, secrets and observability.
  • Define and publish the shared schema package across all four layers.
  • Build the storage layer, event bus and adapter interfaces.
  • Build the first registry adapters and MRV ingestion skeleton.
  • Put the fca_critical test harness and coverage gates in place from day one.
Month 2
Supply side
Scoring, audit and pricing on sandbox data.
  • Complete registry adapters and deterministic reversal detection with auditable lineage.
  • Build the NRCS removal scoring model, permanence haircuts and adverse audit override.
  • Build audit extraction (VVB parse), durability atlas and legal title verifier.
  • Build pricing engine, forward curves by tier and CBAM border delta.
  • Run the supply pipeline end-to-end against sandbox registry data.
Month 3
Buy side
Liability modelling and solvency.
  • Build liability ingestion and normalisation from corporate ERP/treasury systems.
  • Build forward modelling across base, accelerated and stress scenarios.
  • Build the residual calculator and obligation classifier with 100% coverage.
  • Build the Carbon Solvency Ratio engine and its event publishing.
  • Wire up the buy-side orchestration end-to-end.
Month 4
Gate and matching
Certificates, matching and structuring.
  • Build the three-proof orchestrator with deterministic routing only.
  • Build JOSE/JWS signed certificates and the hash-chained INSERT-only audit ledger.
  • Build the matching engine once its objective function is authored and signed off.
  • Build CLS/NRW term sheet generation in machine-readable and human-readable form with FpML output.
  • Build mandatory human approval checkpoints with maker-checker workflow.
Month 5
Lifecycle and hardening
Valuation, reporting and pilot readiness.
  • Build daily valuation, variation margin, cure workflow and buffer pool monitoring.
  • Build dashboard, ISSB/IFRS S2 disclosure packs, registers and board reports.
  • Finalise the client ERP/treasury API and the counterparty verification endpoint.
  • Security hardening, audit trail review and end-to-end integration tests.
  • User acceptance on sandbox data, then a pilot-ready handover.
Enabler tracks · parallel

What gates the move from pilot to live

The software build is the short, controllable track. These external enablers are longer and largely outside the team's control. Production go-live with real client assets and live instruments cannot happen until they clear.

Software platform buildthis concept
Phases 0–4 · ~5 months
ISDA Carbon Supplementexternal counsel
Drafting and working group · ~18 months
IFRS 9 accounting opinionBig Four firm
~6 months
FCA authorisationauditor and regulator
12–18 months from application
Registry and counterparty mandatescommercial
Ongoing through early phases
0369121518 mo
What five months delivers — and what it does not

Delivered. A working platform that ingests data, scores and certifies assets, computes solvency, matches liabilities, generates a draft instrument and produces disclosure packs — all on sandbox and test data with a verifiable audit trail. Enough to demonstrate the concept, run an internal pilot and support diligence.

Not delivered. A live regulated product. Executing a real instrument with real client assets requires the ISDA supplement, a signed IFRS-9 accounting opinion, FCA authorisation, and counterparty mandates. Those are the enabler tracks above.

05 / Assumptions and risk

What the build assumes and where risk sits.

Every assumption below is either drawn directly from the source documents or is a standard condition for a build of this scope. They are the items most worth resolving before committing to scale. The five-month timeline holds only while these hold.

Delivery assumptions

What the five-phase plan depends on

A dedicated squad Assumed

The plan assumes a multidisciplinary team: backend and data engineers, an ML engineer, a quant developer for matching and valuation logic, a front-end engineer and a delivery lead — with access to legal and accounting subject matter input. A lean operations-only group is insufficient for the build window.

Stable specifications Assumed

The documented architecture and schemas are assumed stable for the build window. Material changes to the instrument design or data model during the build ripple through every layer and will move the dates.

Sandbox and test data Assumed

The pilot runs on sandbox registry access and representative test data, including synthetic edge cases for reversals and adverse audits. Live registry feeds are not required to reach pilot readiness, but they are required to go live.

Objective function authored first Build pending

The matching engine's optimisation objective and constraints must be specified and signed off before Phase 2 / Month 4. The source documents flag this as the team's own work to author. It is on the critical path for the matching feature.

Long-state spike completed first De-risk required

Holding a 10–25-year instrument's state — with a daily mark-to-market loop and resumable maker-checker approval interrupts — is heavier than chat-style runs Helium was built for. A small spike must validate this before Phase 0 numbers go into any plan or pitch. If Helium can't support it, a LangGraph + Postgres checkpoint engine is embedded instead.

Scoring training data Build pending

The source, format and update cadence of the dataset that trains the removal scoring model are not yet defined. The model can be built and tested on representative data, but real-world accuracy depends on this dataset being sourced and validated before go-live.

External assumptions

Outside the software — but gating the product

ISDA Carbon Supplement External sign-off pending

The supplement that defines the CLS swap does not yet exist in standardised form. External counsel must draft it and engage the relevant working group — estimated at ~18 months. Instrument execution is blocked until it exists and FpML bank messaging is confirmed.

IFRS 9 accounting opinion External sign-off pending

The claim that custodied credits qualify for held-to-maturity treatment is under review and requires a partner-level opinion from a major firm. The materials note this is load-bearing for the business model and a precondition for a bank credit committee. Not yet issued.

FCA authorisation Authorisation pending

The platform targets custody, arranging and managing investment permissions, 12–18 months from application. The design anticipates the controls required. The permissions are not yet held. This is the critical path item for a live regulated product.

Named counterparties Prospective

Banks, distributors, anchor clients, insurers and investors appearing in the materials are described as in negotiation, with named counterparties requiring signed mandates before external citation. Integration points are in scope as technical work — not as confirmed partnerships.

Registry access agreements Pending

Commercial agreements for registry API access (Verra, Puro.earth, TGO and others) are pending. The legal structure of the joint custodian Durability Trust entity referenced in the anchor deal is also unresolved. Both are listed as critical-path items in the source documents.

Commercial and model risk

Where to focus diligence

Financial model is illustrative Assumption risk

The accompanying financial model is explicitly driver-based and illustrative, with several key inputs as placeholders. The base case reaches approximately 15% of the stated revenue threshold. Only an accelerated, liability-linked path — which books fees on undelivered forward mandates — approaches the target. Read as structure, not forecast.

Going concern and operating risk Risk

The sources list principal risks explicitly: reversal and cure performance under stress, an early-revenue going concern position dependent on later funding, and an arm's-length and consolidation status that is subject to opinion and audit and not yet established.

Recommended sequence before scaling spend

The dependencies are sequential. The legal supplement precedes the accounting opinion, which precedes the assurance engagement, which underpins the first client onboarding and the authorisation evidence. The engineering build can proceed in parallel — but the matching engine objective function, the scoring model training data, the long-state spike, and the consolidation/arm's-length opinion are the four items most worth resolving before committing to scale.