Every source, understood together.

Each document is understood. The whole never is.

Every source gets read in isolation and filed away — so nobody sees that two reports name the same company differently, that July's call already superseded May's forecast, or that the conclusion everyone is citing rests on a source another one contradicts. The understanding that matters lives between the documents, and nothing holds it.

VERI reads one source. MESH reads the whole.

ORIS structures each source. VERI analyses one source at a time and never sees another — which is exactly why its results are reproducible. MESH is where those independent results meet: it evaluates each new source-level result against the relevant Corpus state and discovers the relationships no single document declares, each one traceable back to the sources that support it.

MESH updates the Corpus with traceable relationships and states — it never turns an inference into a fact, never averages competing views, and never overwrites what a source actually said.

Contradictions are knowledge, not errors.

Two institutions forecast the same indicator and disagree. VERI analysed each source on its own and never saw the other; MESH is what notices they are about the same subject. It records the conflict and keeps both — with author, date and evidence — instead of averaging them, overwriting one, or manufacturing consensus.

# report — May 2026
source: macro_outlook_may.pdf · p.8 · Institution A
claim:  "IPCA forecast for 2026: 4.2%"

# audio — July 2026
source: weekly_call_jul.mp3 · 14:32 · Institution B
claim:  "IPCA forecast for 2026: 5.1%"

# neither claim is lost or merged
├── two independent claims
├── same subject · same reference period
├── different authors and dates
├── relationship: contradicts · scope: cross_source · discovered by MESH
└── evidence preserved for both

A later source may update an earlier position, contradict another institution, refine a prior claim or introduce an unrelated view on the same topic. VERI captures these relationships within each source, carrying scope: within_source. MESH discovers them across the Corpus, carrying scope: cross_source — the same vocabulary at two scopes, never mixed and never silently promoted from one to the other.

More than files. More than vectors. More than a graph.

A Corpus is the persistent collection of source-level results analysed by MESH — the sources themselves, what VERI established about each one, and the cross-source understanding built on top. You create a Corpus, add sources to it and govern who can reach it; MESH is what studies it. The knowledge graph is a capability inside MESH — not the product.

Entities and claims are not connected because they look semantically similar. Every relationship retains provenance, source references, temporal context and the evidence that produced it.

"IPCA forecast for 2026: 4.2%"
├── asserted_by   → Institution A
├── supported_by  → macro_outlook_may.pdf · p.8
├── concerns      → Brazil · IPCA
├── valid_for     → 2026
├── contradicts   → claim kn_517 · Institution B
└── created_at    → 2026-05-14

MESH is built on VERI. Every relationship leads back to evidence.

MESH does not read documents. It works exclusively on knowledge grounded by VERI — claims that already carry their span, author, qualifiers and epistemic state. That dependency is the guarantee: a cross-source relationship is only as defensible as the source-level results underneath it, so MESH never accepts raw text, loose chunks, embeddings or claims without provenance.

MESH is a separate product, not a separate errand: send a source and the chain runs for you; bring your own parser output and it starts at VERI; bring a VERI result you already have and it starts at MESH. You are charged only for the steps that actually run — MESH never charges twice for an existing VERI result. What never changes is the contract: nothing reaches MESH ungrounded.

MESH never charges twice for an existing VERI result.

MESH runs on VERI-grounded results, so the honest question is what actually has to run for the request you sent. You are charged for those steps and no others — work already done is never redone, and never re-billed.

A Corpus operation is work over the whole: adding a grounded result to the Corpus, comparing it against the relevant state, relating it to what is already there, or querying across everything. Rates are published at availability — never before.

Served to people, applications, models and agents.

MESH is being built to serve knowledge through Web, REST API, SDK and MCP — retrieval designed to return the relevant claim with its evidence, competing views and history in one traceable response. Never disconnected chunks, never locked to a single AI provider.

# how a Corpus is served — in development
Web       # explore, search, inspect evidence
REST API  # applications and pipelines
SDK       # your language, one contract
MCP       # models and agents, permission-aware
Export    # your asset, portable

# conceptual MCP tools — subject to change
search_knowledge · get_claim · get_evidence
trace_claim · find_conflicts · list_changes

MESH provides the understanding — with the sources behind it. Your applications and models decide what to do with it.

Isolation per organization, permissions at Corpus and source level, full history and audit. Each control is announced when it ships — never before.

Built for knowledge that must survive scrutiny.

Anatomy of a cross-source relationship.

Not a consolidated truth: an addressable relationship between claims that came from different sources — with everything needed to check it, and both sources still behind it.

Watch understanding change.

A new source arrives, VERI establishes what it says, and MESH re-evaluates the subject against the Corpus. Nothing is ever overwritten — every state keeps the source behind it.

The same subject across four sources — every prior state and the source that produced it preserved.

Retrieval finds text. The layer preserves why it is true.