Governed evidence platform / Built in Australia

Evidence that holds up under scrutiny.

Convergence turns large volumes of contested source material into reviewed, traceable evidence for high-stakes decisions. AI does the heavy reading. A governance layer outside the model decides what can be relied on, and human experts hold the final say.

Every claim keeps a live link to its source passage.
Disagreement between sources is recorded and reviewed, never averaged away.
Every output carries its review status. Nothing pretends to be settled when it is not.

Request a walkthrough See how it works

Assertion CV-0412 Held: contested

"Post-combustion capture retrofits have achieved sustained capture rates above 95 percent under commercial operating conditions."

Source: Industry technical report (2024), p.34, para 2. Role: proponent-funded.
Tension: T-117: conflicts with independent monitoring data reporting 62 to 78 percent sustained capture at comparable facilities.
State: Scaffolded awaiting expert adjudication

This claim is excluded from synthesis until the tension is reviewed. The disagreement travels with it, on the record, wherever it is cited.

Illustrative assertion record. The platform's core object: a claim that carries its receipts.

The problem

AI can read everything. It cannot be audited on anything.

Institutions making consequential decisions face evidence that is fragmented and contested: peer-reviewed studies, modelling, regulator reports, industry claims, submissions, grey literature. The hard part is not finding material. It is knowing which claims are reliable, where sources genuinely agree, where agreement merely reflects shared assumptions, and what transfers from one context to another.

Generic AI tools fail this test. They produce fluent summaries with unreliable citations, blend conflicting findings into false consensus, and offer no audit trail. Manual systematic review is rigorous but too slow and too expensive for live decisions. So institutions choose between speed they cannot defend and rigour they cannot afford.

Convergence does not make hard decisions easy. It makes them defensible. The difficulty in complex evidence is information, and the platform is built to preserve it: visible, structured, and reviewed, so that people can make the best call available and show exactly how they made it.

How it works

A governance layer outside the model.

Frontier AI models do the extraction and first-pass analysis. They never get the last word. Everything they produce enters a governed structure where provenance, classification, and promotion are controlled by rules and by people.

1 / Frame

Problem spaces

The decision question is framed precisely enough that heterogeneous sources can be compared honestly, without pretending every document answers the same thing.

2 / Extract

Governed assertions

Claims are extracted as structured objects, each linked to its exact source passage, with the source's role and reliability classified on the record.

3 / Test

Tensions

Contradiction, incompatible assumptions, and boundary conflicts are mapped as first-class records, so apparent agreement is tested before it becomes consensus.

4 / Decide

Decision support

Reviewed structure becomes decision-ready synthesis: what is supportable now, under which constraints, and what still needs work before it can be relied on.

Review states

Three states. No shortcuts between them.

Every piece of evidence in the platform carries one of three states, and the state travels with it everywhere it appears. Promotion is earned through review, never assumed.

Scaffolded

Machine-produced and structured, with provenance attached. Useful for orientation and triage. Clearly marked as unreviewed, and barred from decision outputs.

Reviewed

Inspected by a qualified human reviewer: provenance verified, classification checked, known tensions assessed. Corrections are logged, not overwritten.

Governed

Promoted under the platform's governance rules and fit to carry decision weight. Fully traceable from final synthesis back to source passages, with the review history intact.

Validation

One architecture. Three unrelated domains.

The same governed workflow has been exercised across three deliberately different fields, testing whether the architecture transfers across subject matter without being rebuilt around a single use case.

Aged-care housing supply

Policy-relevant synthesis across regulatory, economic, and demographic sources with conflicting supply projections.

Vision correction economics

Health-economics evidence for low and middle income countries, spanning trial data, cost models, and program evaluations.

Carbon capture claims

Assessment of oil and gas CCS performance claims against independent monitoring data, where proponent and observer evidence diverge sharply.

The architecture is being refined for institutional use with a focus on evidence governance, ontology discipline, and reviewable outputs that can withstand external challenge.

Current focus: hardening the platform for institutional deployment on Australian-hosted infrastructure, with audit logging, access control, and secure source handling built in from the start.

Who it serves

Built for decisions that get questioned later.

Policy and government teams

Can this advice survive estimates, audit, and FOI?

Briefings and regulatory positions backed by evidence that is traceable to source, with disagreement on the record rather than smoothed over.

Investment and risk committees

Can we show our work to the board and the regulator?

Transition, allocation, and risk decisions supported by synthesis that documents what was relied on, what was contested, and who reviewed it.

Research and advisory firms

Can we deliver rigour at commercial speed?

Evidence work of systematic-review quality at a pace and price that fits live engagements, with the audit trail as part of the deliverable.

Contact

Looking at a high-stakes evidence problem?

Start with a walkthrough of the platform on a question that matters to you, or a conversation about whether your evidence base suits governed synthesis.

james@convergencehq.io