Interoperability Is Not Binary

EIA-014 is not asking whether sustainability frameworks can be compared. It asks whether an AI Agent can move one evidence object across GRI, ESRS, TNFD, COSO, GHG Protocol Scope 3 and the UN SDGs without changing the institutional meaning of the relationship. The six DOI publications start from the same 128 canonical MME task positions, but a common origin does not create a common conclusion.

For Singapore capital markets, this becomes a sequence of different decisions. An operational emissions record may support an SGX sustainability disclosure, enter an SFRS S2 workflow, inform a Singapore-Asia Taxonomy assessment, contribute to an ASEAN Taxonomy classification, affect a lender's transition-plan review and later enter limited assurance. Each step has a different authority, threshold, scope and decision owner.

Relationship-Governed Interoperability therefore requires more than a crosswalk. The Agent must know the relationship type, strength, activation condition, evidence state, applicable period and claims boundary. Only then can it distinguish a permissible route from an unsupported promotion of the claim.

Singapore's disclosure architecture is already differentiated. SGX Listing Rule 711A and Practice Note 7.6 place annual sustainability reporting within issuer governance and make the board ultimately responsible. ACRA's roadmap then differentiates STI constituents, larger non-STI issuers, smaller issuers and large non-listed companies, including different timing for ISSB-based climate disclosures, Scope 3 and external limited assurance.

The proposed Singapore Sustainability Disclosure Standards add another boundary. The consultation places mandatory climate-focused requirements in SFRS S2 while SFRS S1 remains voluntary. A data point relevant to IFRS S1 or a broader sustainability topic cannot therefore be treated as if it automatically carries the same mandatory status as an applicable SFRS S2 climate disclosure.

The five relationship states are governance instructions, not probability scores. HIGH identifies strong routing relevance. MEDIUM may identify supporting use. CONDITIONAL requires specified facts. LOW limits the route. NO-DEFAULT-ROUTE tells the system that silence is safer than approximation. None of them establishes compliance, materiality, taxonomy alignment or assurance.

Consider one Scope 3 supplier record received by a Singapore-listed group. For an STI constituent reporting from FY2026, the record may be relevant to a mandatory Scope 3 process. That does not establish that the supplier belongs in a particular category, that the group inventory is complete, that estimates meet the selected methodology or that the final disclosure is ready for assurance.

The same record may also enter a Singapore-Asia Taxonomy assessment or a bank's transition-planning review. Taxonomy classification requires activity-level criteria and thresholds; transition planning requires the financial institution to evaluate the customer's strategy, dependencies, engagement and risk response. A disclosure mapping cannot perform either decision on behalf of the authorised institution.

The board, disclosure committee, risk function, credit committee and assurance provider therefore need different views of the same evidence. A Singapore control design should preserve one canonical record while attaching separate routes, thresholds, decisions and limitations for each use.

ASEAN Taxonomy Version 4 is intended as a regional reference framework, but its inclusiveness is implemented through differentiated pathways. The Foundation Framework, Plus Standard, technical screening criteria, Green tier, Amber transition tiers and entity or portfolio guidance do not make classification binary. They make the basis of classification more explicit.

An ASEAN borrower may operate assets across countries with different national taxonomies, grid factors, sector definitions, data maturity and verification capacity. A Singapore lender can reuse the underlying activity, asset, energy, emissions and transition evidence, but must re-perform the applicable criteria for the jurisdiction, sector, financing instrument and assessment level. A facility result cannot silently become an entity result, and an entity result cannot silently become portfolio alignment.

This is where NO-DEFAULT-ROUTE becomes economically useful. When the local factor, technical threshold, legal classification or transition condition is missing, the correct Agent action may be to hold the assessment and request evidence. Choosing the nearest regional category would create false comparability and potentially misdirect capital.

Relationship Inflation becomes material when it changes eligibility, pricing, mandate compliance, use-of-proceeds allocation, sustainability-linked financing terms, portfolio alignment, risk appetite or investor communication. It is the mechanism by which a valid upstream relationship becomes an invalid downstream capital-market claim.

The inflation paths are concrete: Scope 3 relevance → category applicability → inventory completeness → emissions conclusion; ASEAN Taxonomy Amber → Green; eligible activity → aligned entity → aligned portfolio; transition plan → implemented action → financed-emissions reduction; ISSB relevance → disclosure compliance → limited assurance. Every arrow requires a separate test and decision record.

A capital institution should therefore ask four questions before an AI Agent uses mapped evidence: What exact decision is being made? Which threshold and period apply? Who has authority to make it? What stronger conclusion remains prohibited? If any answer is missing, automated execution should stop before the claim enters a term sheet, investment memo, product label or public disclosure.

The minimum machine record must include the evidence identifier, originating entity and facility, issuer or borrower tier, financial year, activity and sector code, source system, methodology, estimate status, data quality, destination framework and requirement, relationship type and strength, activation condition, threshold, review owner and version.

Decision-specific states must remain separate. For disclosure, retain materiality, applicability, reporting boundary and board approval. For taxonomy, retain activity eligibility, technical screening, essential criteria, Green or Amber tier and entity or portfolio level. For finance, retain mandate, credit or product decision. For assurance, retain scope, level, evidence sufficiency, exceptions and conclusion.

The controlled path is: Operational Evidence → Framework Route → Eligibility or Applicability Test → Entity or Portfolio Context → Decision Authority → Disclosure, Financing or Assurance Outcome. The system must log whether the Agent routed, held, escalated or stopped, together with the evidence and rule version that produced that action.

The first implementation test should use a real Singapore-listed group or ASEAN borrower, not a conceptual crosswalk. Select one material operational evidence object and route it into four workflows: SGX or SFRS S2 disclosure, Singapore-Asia or ASEAN Taxonomy classification, credit or investment review, and assurance preparation.

For each workflow, record the relationship strength, activation facts, applicable threshold, decision owner, permitted use and prohibited claim. The test succeeds only if the evidence is reused without any system promoting relevance into applicability, Amber into Green, activity eligibility into portfolio alignment, a plan into performance, or disclosure readiness into assurance.

Monitor Singapore's tiered climate-reporting implementation, mandatory Scope 3 reporting for STI constituents from FY2026, preparation for limited assurance, MAS transition-planning implementation, and Version 4 ASEAN Taxonomy use cases. The decisive signal will be whether live systems preserve relationship and decision boundaries after AI-enabled automation begins.