Skip to content
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Govern Graph

Load an institution's knowledge graph from a graph package, see where every fact came from, and let agents query it.

Govern Graph holds one knowledge graph per institution: its products, pricing, features, channels, policies, disclosures, document requirements, rules and thresholds, and the regulatory packs behind them. Agents use the graph to answer questions from the institution's own governed facts. You load it from a graph package, a JSON file that lists the graph's nodes and relationships. Each fact in the graph keeps its source, owner, confidence and effective date, so you can trace where a value came from. An uploaded graph becomes knowledge in the account, so you attach it to agents the same way as any other knowledge.

Govern graphs are account-scoped: a graph you upload is created for the whole account, and every project in it can read it.

Open it in Control Center → Govern → Govern Graph. To load or change a graph, see Upload and manage a govern graph.

Permissions

Action
Roles

View govern graphs

Approver, Governance Manager, Auditor, Account Admin, Org Admin

Upload, rename and manage govern graphs

Governance Manager, Account Admin, Org Admin

Attach a govern graph to an agent

Developer, Account Admin, Org Admin

Retry, Upload new version and Edit need the Ingest source & extract knowledge graph candidates capability. See Roles for the full capability matrix.

Lifecycle of a govern graph

A govern graph moves through seven stages. The first six are available. The seventh, sign-off, is a preview.

Stage
What happens

Upload

You upload a graph package, a .json file. The package is validated before anything is written.

Build

The platform writes the nodes, then the relationships. The row reads Processing with the current stage. It ends as Completed or Failed. A failed build retries on its own, then waits for you to select Retry.

Version

Each completed build is a numbered version of the graph. Agents read the active version. When you upload a newer package, they keep reading the current version until the new one finishes building, and then switch. A failed build changes nothing for agents.

Review

You browse the graph on Overview, Ingest & Map and Graph Explorer, and trace any fact on Provenance.

Refine

You change a value with Edit node, upload a newer package with Upload new version, or go back with Restore. Upload new version and Restore each make a new version, and earlier versions are kept.

Use

You attach the graph to agents as knowledge. They query it with Knowledge Graph Search.

Sign-off (preview)

You attest a node, and it becomes ratified. Nothing you attest in the preview is saved.

Two kinds of state apply, and they're easy to mix up.

The graph's build state is what you see in the list. A graph is Processing, Completed or Failed. In Version history, the version agents read carries an Active badge, and a version whose build failed carries a Failed badge.

A node's sign-off state is what you see on Provenance and the Sign-off Queue (Preview). A node is unconfirmed until a person attests it. Nothing attested in the preview is saved, so every node is unconfirmed, and agents read every node.

A govern graph can't be deleted. Uploading a new version and restoring a version never remove an earlier version.

What a graph holds

Nodes

A node is a single object in the graph. Nodes fall into two groups.

Business objects describe what the institution offers:

Object
What it is

Institution

The bank or credit union itself

Product

An account or loan product

Feature

A feature of a product

Segment

A group of customers a product is aimed at

Channel

A way customers interact, such as a branch, app or phone

Pricing

A fee, rate or price

OfferPromotion

A promotion or offer

Customer

A customer. Always empty

Account

A held account. Always empty

Governance objects describe what the law and the institution's policies require:

Object
What it is

RuleThreshold

A number a law or policy fixes, such as a maximum hold period or a reporting limit

Authority

Who can decide or sign, such as the BSA officer or the board

Policy

The institution's written position on how it meets a rule

Disclosure

What the institution must tell the customer, and when

DocumentRequirement

What the customer must provide, such as ID

Consent

What the customer must agree to before the institution acts

Source

Where a fact came from, such as a regulation, an institution document or a web page. Every other node points to one

Customer and Account are always empty. Member data is never stored in the graph. It's resolved from your core systems when an agent needs it.

Every node carries the same properties:

Property
Records

ID and version

The node's identity and version

Effective from

The date the value took effect

Source provenance

Where the value came from: the source, its class and the pack, where one applies

Confidence score

How reliable the value is. A seeded placeholder is low. A value read from a source is higher

PII classification

Whether the value holds personal data

Owner

The team accountable for the node

State

Its sign-off state. New nodes start as default_unconfirmed

Fill

How the value was populated. seeded means it came from a regulatory pack. A node read from a public source carries a source-derived value

Default issue flag

Marks a node that needs attention. NEEDS_FI means the institution must supply or confirm the value. These nodes count as open gaps on Overview and are listed first on the Sign-off Queue

A node created from a regulatory pack is a placeholder until evidence supports it and a person signs it off. Seeded doesn't mean true.

Unconfirmed, attested and ratified

A graph starts as a draft. Its nodes are built from public sources, such as the institution's website, and seeded from regulatory packs. Nobody at the institution has confirmed them. Attestation is the gate between that draft and the institution's own truth: until a person at the institution attests a node, it's a proposal and not a fact the institution stands behind.

Three terms describe where a node is:

Term
Meaning

Unconfirmed

The starting state, written default_unconfirmed. The value was built from a public source or seeded from a pack. Every node begins here, whatever the package says.

Attested

A person at the institution has confirmed the value. Attesting is the action you take on the Sign-off Queue. Overview counts the share of nodes attested.

Ratified

The state a node reaches once it's attested. A ratified node is the institution's confirmed value, with the attester and the time on its record.

In short, you attest a node, and it becomes ratified. A node that was never attested stays unconfirmed.

Nothing attested in the preview is saved, so the gate isn't enforced. Every node stays unconfirmed, and agents read every node whether or not it has been attested. Treat answers from the graph as unattested. See Sign-off Queue (Preview).

Relationships

A relationship is a typed link between two nodes. For example, a product is GOVERNED_BY a disclosure, or a product INCLUDES a feature. Relationships let you move from a product to its pricing to the rule that governs it, in the Graph Explorer or in an agent's query.

The relationships in a graph are the ones its package defines. Each entry in the package's edges list names the node it starts from in from, a relationship type in rel and the node it points to in to. Qore stores them as written and doesn't add relationships of its own.

A relationship is written as a pattern. (:Product)-[:GOVERNED_BY]->(:Disclosure) reads: a Product node is linked by GOVERNED_BY to a Disclosure node. A package for a credit union typically defines patterns like these:

Relationship
Meaning

(:Institution)-[:OFFERS]->(:Product)

The institution offers this product.

(:Product)-[:INCLUDES]->(:Feature)

The product includes this feature.

(:Product)-[:PRICED_BY]->(:Pricing)

The product's rates are defined by this pricing schedule.

(:Product)-[:APPLIES_TO]->(:Segment)

The product is available to this member segment.

(:Product)-[:GOVERNED_BY]->(:Disclosure)

The product is subject to this disclosure document.

(:RuleThreshold)-[:APPLIES_TO]->(:Product)

This rule threshold constrains or governs this product.

(:Institution)-[:OPERATES]->(:Channel)

The institution operates this service channel.

(:Institution)-[:GOVERNED_BY]->(:Disclosure)

The institution is subject to this disclosure document.

(:RuleThreshold)-[:APPLIES_TO]->(:Institution)

This rule threshold applies at the institution level.

(:RuleThreshold)-[:REQUIRES]->(:Product)

This rule threshold requires the existence of this product.

Your own package can use other node types and relationship types. The Graph Summary on About Graph lists the ones in your graph, under Relationships.

A relationship is stored only when both of its ends are nodes in the package. If one end is missing, the relationship is skipped. See Graph package.

Regulatory packs

A regulatory pack, or reg pack, is a bundle of rules for one regulatory topic, such as consumer deposits or customer identification. Packs seed a graph's rules, policies and disclosures, so you start from the obligations that apply to your kind of institution instead of building them by hand.

The institution's type decides which packs apply: its charter, whether credit union, bank or thrift, the customers it serves, its product focus and its size. Every institution receives a base set of packs. Further packs are added if it lends, serves businesses, or is large.

Each pack is built in three layers:

Layer
Contents

Citation

The exact address of each rule in the law, with its version

Obligation

A plain-English statement of what the institution must do

Full text

The regulation's word-for-word text, taken from the official source and never reworded

Packs are maintained in a shared library and are the same for every institution of the same type. The institution's own documents, such as its disclosures and fee schedules, are a separate layer in the graph and are not part of a pack. The graph points to a rule's source. It doesn't store the regulation's wording.

A pack is applied when the graph package is built, and the package records which packs it used. Overview and Ingest & Map list them.

Packs are grouped into tiers by what they cover:

Tier
Covers

Floor

Rules every institution of that type follows, such as deposit accounts, privacy, fraud, payments and communications

Lending

Consumer, mortgage, small-business and servicing rules, added when the institution lends

Commercial and treasury

Rules for business and commercial customers

Prudential

Governance, risk, capital, reporting and vendor oversight

Overlay

Additional consumer-protection rules for larger institutions

Each pack has a short code. These are the codes you'll see most often on Overview:

Code
Pack
Regulations it covers

DEP

Deposit accounts and Truth in Savings

Reg DD, Reg E, Reg CC, Reg D

CIP-BSA

Customer identification and BSA / anti-money-laundering

BSA/AML, CIP, OFAC, SAR/CTR reporting

PRIV-SEC

Privacy and data security

Reg P for GLBA privacy, GLBA Safeguards, Red Flags

FRAUD

Fraud and identity theft

Red Flags, FCRA, EFT error and unauthorized transfers

SERV-DEP

Deposit servicing

Reg E, Reg CC, subsequent Truth in Savings disclosures

DISPUTES

Error resolution and disputes

Reg E and Reg Z error resolution, FCRA disputes

PAY

Payments and funds transfer

Reg E, Reg CC, Reg J, ACH, Reg GG

CONS-COMPLIANCE

Consumer compliance and UDAAP

UDAAP, FTC Act section 5

COMMS

Member communications and advertising

Advertising rules, share-insurance signage, TCPA, CAN-SPAM

CARDS

Card programs

Reg Z for the CARD Act, Reg E, Reg II

Other packs follow the same pattern: LEND- packs for lending, BANK- packs for banks and THRIFT- packs for savings associations.

Graph list

The list shows each graph in the account.

Column
Shows

Graph

The graph's name.

Version

The newest version number. When the newest version isn't the one in use, the active version follows it.

File

The package file.

Created At

When the graph was created.

Status

Completed or Failed.

The refresh button next to Upload Govern Graph reloads the list. Select a completed graph to open it. The ⋮ menu on a row has these actions:

Action
What it does

View

Opens the uploaded package.

Download

Downloads the package file.

Retry

Tries a failed build again. It shows only when a retry is possible.

Upload new version

Uploads a new package for this graph.

Version history

Opens the graph's versions, where you can open or restore one.

Edit

Renames the graph. Renaming changes only the name, not the package, nodes or versions.

Builds in progress

While a graph builds, the row reads Building the graph. You can open a graph that hasn't finished building from this list. Its page names the current stage, such as Queued, Validating the package, Writing nodes or Writing relationships, or says the build is waiting for an automatic retry. If the build failed, the page gives the reason, the stage it failed at, the retry notice and a Retry button. Retry is hidden when the failure can't be retried. When a version becomes active, the page changes to the graph's tabs.

From the Knowledge list, an unfinished graph's page says the graph is still building. It shows no progress and no failure reason, so a failed graph can read the same way. Check the graph list instead.

Versions

Each row in Version history gives:

Item
Shows

Version number

The version of the graph.

Package file and time

The package the version was built from, and when it was made, as a relative time.

View graph

Opens that version's graph on the same tabs, so you can look at an earlier version while agents read another.

Restore

Makes that version the one agents read.

The version agents read carries an Active badge and has no Restore. A version whose build failed carries a Failed badge and has neither button. A restore adds a new version noted Restored file version N and keeps the later versions, so you can restore forward again.

Graph package

A package is one .json file, up to 80 MB. A file that isn't .json or is larger than 80 MB can't be selected. Qore reads the package as it is. This table lists what is accepted, skipped and stored.

In the package
What happens

A node with no id

The upload fails.

An edge whose from or to isn't a node in the package

The edge is skipped. This includes every edge from a regulatory obligation ID, because obligations appear only as edge endpoints.

A field Qore ignores

Not stored. seeded_by, which repeats the seeds edges, is ignored.

An empty value, which is null or an empty list

Not stored.

Field names

Stored in camelCase: effective_from becomes effectiveFrom.

Relationship types

Stored in upper snake case: governed_by becomes GOVERNED_BY.

Package details, such as package, version, fi and coverage

Kept as the graph's envelope and shown on Overview.

A package built with regulatory seeding can have far more edges than the graph keeps, because the seeds edges from obligation IDs point at nodes outside the package.

Own Neo4j settings

By default, Qore builds a graph in the platform's own Neo4j graph database. In a development environment you can build it in your own Neo4j server instead. In the upload dialog, Use your own Neo4j (Dev Only) opens a window titled Use your own Neo4j, with the line The graph is built in this Neo4j instead of the platform's.

Field
Required
What to enter

Connection URL

Yes

The address of your server, in the form neo4j+s://<host name>. Only neo4j+s:// addresses to a publicly reachable server are allowed.

Username

Yes

A Neo4j user for the server.

Password

Yes

The user's password. The eye icon at the right of the field reveals or hides what you type. The password is stored encrypted.

Database

No

The database to build in. Leave it empty to use the server's default database.

Button
What it does

Test connection

Tries the connection with the details you entered, so you can catch a mistake before the build.

Save

Keeps the details for this upload. It doesn't connect.

Cancel

Closes the window without saving.

Test connection and Save aren't available until the required fields are filled. A graph built on your own server can't be exported or imported. In production, Qore refuses the option.

Knowledge list

A govern graph also appears in Knowledge in the Console, in MY KNOWLEDGE, with a Govern Graph badge and an eye icon in place of the play icon. A graph has no Try Knowledge control. Its row menu has these actions:

Action
Behavior

Share Knowledge

Isn't available, because everyone in the account already has access.

Export Knowledge

Isn't available for a graph built on your own Neo4j server. The tooltip reads A govern graph on your own Neo4j server cannot be exported or imported.

Delete

Is in the menu, but a govern graph can't be deleted.

Exporting a knowledge base that holds govern graphs is supported. Qore first checks that each graph can be exported, and refuses the export if not. The ZIP holds the knowledge package, with its govern graphs, and the document files. See Govern graphs in My Knowledge.

Inside a graph

A graph opens on six tabs: Overview, About Graph, Ingest & Map, Graph Explorer, Provenance, and Sign-off Queue (Preview).

The Overview tab of a govern graph, with the institution profile, counts of graph nodes, relationships, attested share, open gaps and regulatory packs, and the graph composition breakdown
Overview.
  • Institution profile. The institution's name, type and charter.

  • Headline counts. Graph nodes, relationships, the share attested, open gaps, and the regulatory packs loaded. Open gaps are items the institution still has to supply.

  • Graph composition. The nodes grouped into entities, deterministic rules, governance documents and sources, with a count per object type. Customer and Account are empty by design: member data is never collected into the graph; it's resolved at run time from the core systems.

  • Attestation & confidence. How many nodes the institution has attested, and the spread of source confidence.

  • Regulatory coverage. The regulatory packs applied to the graph, each labelled with its short code. The total beside the list is the number of packs in the library for the institution's type, such as credit unions, not for every type. A note beneath the list states that the library holds further packs, loaded together with the product-line modules that need them. The card identifies which packs seeded the graph. It doesn't report whether each obligation has been met, because the graph doesn't keep the link from an obligation to the rule it seeded.

About Graph has two tabs, Graph Summary and Traversal Guidelines. Both are written by the platform from the graph, and they are generated automatically once the graph is built. Until then, a tab reads No graph summary yet and It is generated automatically once the graph is built. If it has not started, generate it now., with a Generate button. While one is running, the tab reads Generating graph summary… or Generating traversal guidelines… and This runs in the background and takes a few minutes. You can leave this page.

  • Graph Summary describes what the graph holds, in sections:

    • Purpose. What the graph encodes and what it was built from.

    • Questions it answers. The questions the graph can answer, followed by what it does not cover. For example, it can note that fee amounts mentioned in disclosure notes are not ingested, or that real-time balances, transaction history and member-specific eligibility are outside the graph.

    • Entity types. Each node type in the package, with its count, a one-line description and the key properties worth filtering on, such as name, type and owner.

    • Relationships. Each relationship pattern in the package, such as (:Product)-[:GOVERNED_BY]->(:Disclosure), with a plain-language meaning. See Relationships.

    • Topics covered. The subjects the graph covers.

  • Traversal Guidelines are written topic by topic. Each says where to start in the graph and which relationships to follow, for example: start from a product by name, then follow what it includes or what it is priced by.

Once a guide exists, its tab has Regenerate and Save buttons. The text opens in an editor. Its toolbar has undo and redo, bold, italic, underline and strikethrough, numbered and bulleted lists, alignment, heading levels H1 to H3, and code formatting. The text is editable in place: change it and select Save to keep your version, or select Regenerate to have the platform write a new one. Each save or regeneration is stored as a new version of the guide for that graph version. A summary you have edited is not replaced when the same package is imported again. The guides are stored as text for each graph version, and agents use them when they query the graph.

The Ingest and Map tab, listing each source with the node types it produced and a confirm count, beside the foundation catalog layers A to H with node counts
Ingest & Map.

Ingest & Map is a read-only view of where a graph's content came from and how it fills the foundation catalog.

  • Sourced so far. Each source, such as a regulatory pack seed or a web page, with the node types it produced and confirm N, the number of its nodes still waiting for confirmation. Selecting confirm N opens the Sign-off Queue (Preview), which lists all of the graph's nodes, not only that source's.

  • Maps to foundation catalog. How the nodes fill the catalog layers. Each row gives the number of nodes the graph holds in that layer.

The foundation catalog is the fixed set of layers every node is filed under:

Layer
Name
Holds

A

Institution & Org

The institution and its organisation

B

Product, Pricing, Offer

Products, rates, fees and promotions

C

Compliance & Disclosures

Disclosures and regulatory references

D

Enterprise Policies

The institution's internal policies

E

Customers & Segments

Member segments. Labelled runtime: resolved from core systems when needed, never stored

F

Channels & Comms

Branches, ATMs, digital and contact channels

G

Systems & Integration

Systems the institution runs on

H

Governance & Authority

Authority levels, thresholds and approvals

A layer with few or no nodes marks a gap the institution still needs to fill. Public sources usually fill the product, pricing and compliance layers. Systems and authority thresholds come from the institution.

The graph is loaded from a package built outside Qore, and Qore doesn't read documents and extract facts itself. The extracted label on a source describes how the package was built. Nothing a source produced counts as fact until it's attested.

  • The explorer opens with the graph's node types as bubbles, each with its count, joined by the relationships between types. Select a bubble for its actions, or double-click it to open it. The layout menu offers All nodes, Group by node type, and Group by relationship. Filter by kind narrows the kinds shown, and the buttons at the right zoom the view.

  • Search nodes finds a node by name, with the results grouped by node type. Choose one to see it with its neighbours.

  • The selected node panel shows the node's name, its type, and its number of relationships, with two tabs. Properties lists each property of the node, including its source provenance. Relationships lists its connections.

  • Edit node, the pencil in the panel, changes the node's properties. See Node editing.

The Provenance tab, tracing the Share Savings APY from its rate sheet source through the extracted field and the Pricing graph node to its default-unconfirmed status, above the source register
Provenance.

Provenance answers one question: where did this fact come from, and where does it stand now? It has two parts, the trace for one fact and the source register for the whole graph. You can also reach it from Lineage on the Sign-off Queue (Preview).

The trace

The heading names the fact and gives its node ID at the right. The trace follows the fact in four steps, left to right. Select a step for an explanation.

Step
What it gives you

Source

The source the fact was read from, with its type and the date it was read. Source opens the original page or document.

Extracted field

The value pulled from that source, and the name of the field it was stored in.

Graph node

The node type and version the fact was stored as, and the confidence with which it was placed.

Status

The fact's sign-off state. A fact nobody has attested reads Default-unconfirmed, with the label awaiting institution sign-off. See Unconfirmed, attested and ratified.

Four cards sit beneath the trace:

Card
Meaning

Owner

The team accountable for the fact, such as Product.

Effective from

The date the value took effect.

As-of (graph version)

The graph version that held the fact, with its date. This is the version an agent read.

PII classification

Whether the value holds personal data. none means it doesn't.

Effective time is when the value took effect. As-of time is which graph version an agent read. The two are tracked as separate values, so you can always answer "which rate did the agent quote, and was it current?" A note under the cards says so.

The source register

The source register covers the whole graph, not one fact. Its header gives the number of sources and the date of the last sync. Each source is a row with:

  • its name and type;

  • the number of nodes it produced. A source that fed no node shows a count of zero;

  • its confidence;

  • the date it was read, and a Source link to the original.

Ingest & Map lists the same sources from the other side, by what each one added to the foundation catalog.

The Sign-off Queue preview, with counts of nodes awaiting sign-off and ratified, and review queues for Compliance and Risk, Product and Finance, Marketing, and Ops and IT, each with Bulk attest
Sign-off Queue (Preview).

Sign-off is how a node moves from Draft to Ratified. The queue groups nodes by the accountable role that owns them, such as Compliance & Risk, Product & Finance, Marketing, and Ops / IT, with Lineage for each node and Bulk attest per role.

The cards at the top give the state of the graph:

Card
Meaning

Awaiting sign-off

Nodes that are still default-unconfirmed

Ratified

Nodes that have been ratified. It stays at zero, because nothing attested in the preview is saved

Your queue

Values the institution still has to supply, shown as NEEDS_FI, listed first

Lifecycle

A label naming the stages of a node's life: Draft, Ratified, Stale and Retired

Lineage takes you to the Provenance tab, where the trace follows a fact from its source to its status. Bulk attest marks every node in a role's group at once, and Bulk attest ready set marks the nodes that aren't waiting on a value from the institution. When you attest, Awaiting sign-off falls, Ratified rises, and each role's group reports how many of its nodes are ratified.

Node editing

Edit node in the Graph Explorer panel turns every property into a field, except the ID, which can't be changed. Name is required. Features takes one item per line, and Confidence Score takes a number. Leaving a node with unsaved edits asks before it discards them.

The fields are the properties described under Nodes: Name, Owner, PII classification, Fill, Version, Confidence score, State, Effective from and Source provenance, which holds the source, its class and the pack. Fill records how the value was populated, for example seeded for a value taken from a regulatory pack.

Editor behavior
Detail

Values you can change

Values at any depth, including items inside a list.

Read-only values

Structured values, and values the editor can't address.

Structure

You can't add or remove nested objects or list items. You can only change values inside the ones that exist.

Agent skills

A graph that has finished uploading becomes knowledge in the account. When a govern graph is attached to an agent, the agent queries it with these skills.

Skill
Behavior

Knowledge Graph Search

Answers structured and relational questions, such as which documents a product requires or which policy governs a disclosure. It turns the question into a read-only graph query and returns up to 25 rows, and tells the agent when more matches exist.

Knowledge Graph Cypher Query

A pre-attached skill that isn't on by default. Turn it on in Agent Config → Skills & Capabilities, on the Pre-attached Skills tab. The agent writes a read-only Cypher query and runs it with the same safety checks as Knowledge Graph Search. The agent is told to use it only when it already has the exact query, from the user, its instructions or an earlier search.

The Console Operator reaches a graph through the Knowledge Builder agent it calls. The agent can list graphs and their versions, read a graph and its guides, ask it questions, retry a build, create or restore a version, edit nodes, and regenerate or edit the guides. It asks for permission before it edits nodes or restores a version.

You can't create or attach a graph by attaching the package JSON in the Agent Builder chat. Upload it in Govern Graph first.

Messages

Message
Meaning

"Failed to build this graph."

The build failed. The recorded reason follows.

"The build failed without recording a reason."

The build failed and no reason was recorded.

"This will be retried automatically within the next few minutes."

Retries remain.

"Automatic retries have been exhausted. Use Retry to try again."

No automatic retries remain. Select Retry.

"Version restored"

A restore finished.

"Restore version N?"

Confirms a restore. Agents start reading that version right away, nothing is rebuilt, and later versions are kept.

"The graph is created for the whole account, so every project can read it."

Shown in the upload dialog.

"Custom Neo4j connections are not enabled in this environment."

Qore refuses Use your own Neo4j (Dev Only) in production.

"No graph summary yet"

The summary or guideline isn't generated yet. Generate starts it.

"Generating graph summary…"

The summary is being generated in the background.

For package errors shown on the Failed badge, see Troubleshoot govern graphs.

What's next

Last updated

Was this helpful?