How Kyroq works
How Kyroq produces investment-grade verdicts.
Methodology, framework alignment, evidence architecture, foundation-model substrate. Everything your compliance, legal and risk teams need to scope an engagement. Read in 10 minutes.
Part 01
Five frameworks. One output.
Between 2023 and 2025, the major nature-disclosure frameworks reached substantive convergence on what evidence an underwrite must produce. Build the evidence graph correctly once, and the same underwrite ships against all five.
FRAMEWORK 01 · DISCLOSURE
TNFD LEAP-FI.
The Taskforce on Nature-related Financial Disclosures, finance-sector guidance (LEAP-FI) released September 2023, sector-specific 2025 update. Locate, Evaluate, Assess, Prepare. Adopted by 500+ corporates and institutions by Q2 2026.
KYROQ MAPS TO: §LEAP-FI sections 1–7, including dependency-and-impact mapping, priority biome flagging and disclosure preparation.
FRAMEWORK 02 · PRUDENTIAL
NGFS three-phase.
The Network for Greening the Financial System nature-risk scenarios. Phase 1, physical risk. Phase 2, transition risk. Phase 3, credit-risk translation (PD/LGD). Adopted by EBA, ECB, PRA, Banco do Brasil and the Federal Reserve.
KYROQ MAPS TO: Asset-level physical risk, transition-trajectory, and credit-risk metric derivation per deal.
FRAMEWORK 03 · BANK-RISK
UNEP FI NBRA.
The UNEP Finance Initiative's Nature Banking Risk Assessment framework. The de-facto bank-risk language for nature-positive lending. Adopted across the multilateral DFI system and the major sustainable-finance banks.
KYROQ MAPS TO: Sectoral nature-risk classification, dependency-and-impact mapping at deal level, mitigation hierarchy.
FRAMEWORK 04 · PROJECT-RISK
IFC PS1–PS8.
The IFC Performance Standards. The 2024 revision sharpened PS6 (Biodiversity) and the offset-design requirements. The most rigorous bar in development finance, and the operating language used by IDB Invest, KfW DEG, FMO, BII and Proparco.
KYROQ MAPS TO: PS1 risk identification, PS5 land tenure, PS6 critical-habitat triage and mitigation hierarchy, PS7 IPs and FPIC.
FRAMEWORK 05 · TA & PRICING
EIB NCFF.
The European Investment Bank's Natural Capital Financing Facility. Technical Assistance bands published with explicit budgets: small (€17–35K), mid (€30–50K), large (€700–935K), flagship (€990K). The benchmark for pricing nature-analytics engagements in Europe.
KYROQ MAPS TO: TA-band documentation, scoping memo and disbursement-readiness pack.
CONCURRENT MAPPING
One evidence graph. Five concurrent verdicts.
The framework-alignment matrix maps every section of every Kyroq memo to all five frameworks in parallel. Source citations point back to the same observations. Your CSO reads TNFD; your CRO reads NGFS; your bank-risk team reads NBRA; your IFC-aligned partner reads PS6; your EIB-facing team reads NCFF. One memo, five readers, five aligned verdicts.
The full framework-alignment matrix, every section of the Fazenda Buriti Verde sample memo mapped to all five frameworks, ships as a downloadable PDF appendix to the sample memo.
Part 02
Every claim resolves to a specific piece of evidence.
Evidence-pinning is the property that turns an AI output into an underwriting decision. We engineered for it from day one. Three columns describe the contract.
WHAT YOU SEE
A claim in the memo.
"Cerrado primary vegetation cover: 35.4% of the polygon, above the Forest Code Article 12 legal minimum for the biome."
WHAT IS PINNED
The source.
The exact satellite tile (Sentinel-2 L2A, tile T22LCJ, acquired 2025-04-18). The classification head version (kyroq-cerrado-v3.2). The methodology paragraph (Forest Code Article 12, in force since 2012, applicable to the Cerrado biome).
CRYPTOGRAPHIC PROOF
The audit anchor.
SHA-256 hash of the source bundle. RFC 3161 timestamp signed under an eIDAS-qualified time-stamping authority. Immutable verdict-registry entry. Replayable forever, bit-identical, by any independent third party.
Tamper-evident by design. Reproducible by independent third parties. Designed for MRM audit. Verifier CLI is open-source. PwC, EY, KPMG and Bureau Veritas have all accepted the replay bundle as audit evidence in commercial engagements we have run.
Part 03
The agent drafts. The underwriter signs.
Every section of the memo carries a data-quality tier (A / B / C) and a percentage confidence score. Sections below 60% confidence auto-route to senior-underwriter review before delivery. The decision and the liability stay with your team.
Audit-grade. Multi-source ensemble. Direct registry data.
investment-committee-grade. Single-source primary data. Validated against priors.
Escalated. Auto-routes to senior underwriter for review.
The liability stance. Kyroq is a tool-provider. The customer-side underwriter owns the decision, signs the memo and carries the personal and institutional liability. Our role is to compress the production cycle of the evidence chain, not to substitute for the human judgement that signs it.
Part 04
Built on commodity substrate. Defended at the workflow layer.
The open-weights Earth-observation foundation models the field is converging on are commodity. We fine-tune for tropical biomes, integrate primary-source administrative data, and build the agentic workflow above. The substrate is open. The workflow is what's hard to replicate.
PERCEPTION
Fine-tuned EO foundation models.
Multi-source fusion of optical, SAR and biophysical layers. Proprietary in-house fine-tunes for the Cerrado, Amazon, miombo, Atlantic-forest and southeast-Asian biomes, calibrated against ground-truth from the 240-cooperative reference network. Cerrado-tune F1: 0.94.
REASONING
Frontier-LLM orchestration.
An agentic AI workflow over a typed tool surface. Versioned prompts. Logged tool calls. Bring-your-own model where customer policy requires it. Private deployment available for tier-one regulated engagements.
ORCHESTRATION
Typed evidence graph + MCP tools.
Every tool the agent invokes is a named, typed contract over the evidence graph. Tools are auditable. Tool-call logs are part of the replay bundle. The orchestration layer is reproducible by any independent third party.
Read the full methodology.
The 38-page methodology paper covers framework alignment in detail, the evidence-graph data model, the model card library, the validation benchmark suite, the MRM Trust Package contents and the regulatory analysis. Co-authored with academic partners (forthcoming Q3 2026).