Actuary — DAG Editor

You can model anything.
You just have to learn how.

As the Actuary, you are the architect of the logic engine. You don't write formulas — you build graphs. Each graph is a precise, unambiguous representation of a clause that the Manager can apply to real contracts.

🕸️ DAG graphs

🔵 10 node types

➡️ 6 link types

🧩 Dimensions

🔗 External parents

👻 Phantom nodes

Your role in the chain

Four roles — one direction. Each step unlocks the next.

Step	Profile	What they do	What you receive
1	Junior	Deconstructs clause text — creates structured clauses with wording, family, treaty types, dependencies	Clauses in draft
2	Senior	Validates clauses and designs the dimension catalogue — axes you will attach to graph nodes	Validated clauses + dimension palette
3	Actuary ← you	Builds DAG graphs per validated clause. Attaches dimensions node by node. Links graphs across clauses.	Graphs ready for Manager
4	Manager	Creates contracts, instantiates a graph, fills in dimension values per node	—

💡

You only see validated clauses. If a clause is missing, the Senior must validate it first. You cannot modify clause content.

Clause sidebar

The left sidebar is your primary navigation — a scrollable list of all validated clauses with their graphs nested inside.

Each clause card shows its identifier (CL_NNNN), title, treaty type pills (coloured by type: FAC, XL, TREATY_XL, QUOTA, SURPLUS), and 🏷 RI term tags with definitions on hover. Click a clause card to expand it and reveal its graphs.

Creating a graph

Click + New graph directly on a clause with no existing graph, or click the + Graph button at the top of the expanded clause in the sidebar. Give it a descriptive label — the Manager uses this to choose which formalisation to instantiate.

💡

One clause, multiple graphs. You can create several graphs for the same clause — alternative formalisations or versions for different treaty types. The Manager chooses which to instantiate per contract.

Graph status indicators

Each graph row in the sidebar shows two badges: a ♻ cycle warning if the last save detected a cycle, and the current save state ✓ saved / ● unsaved.

Building a DAG

A Directed Acyclic Graph encodes the computational logic of a clause as a traversable structure of nodes and directed links.

A reinsurance clause contains nested formulas — "net retained loss, net of recoveries, in excess of the retention, capped at the limit." Each element becomes a node. Each dependency becomes a directed link. The root node holds the clause's final value.

The graph is not a calculator. It is a documentation structure. The Manager reads it, understands exactly what to fill in, and the result is an unambiguous record of the clause's parametrisation on a contract.

Workflow

Select a graph in the sidebar

Click a graph row to load it into the canvas. The canvas area is a pan-and-zoom SVG editor. Use the toolbar to fit the view, add nodes, save, or export CSV.

Add nodes

Click + Node in the toolbar. A temporary node appears on the canvas and the right-hand inspector opens on its Properties tab. Set label, operand, and value type, then click Apply.

Draw links

Hover a node until the edge handle appears at its border, then drag to a target node. A link is created and an operand selector appears — choose PLUS, MINUS, FACTOR, etc.

Attach dimensions

Select a node, open the Dimensions tab in the inspector, and click ＋ Add dimension to pin documentary axes to this node.

Save

Click 💾 Save in the toolbar. The entire graph — nodes, links, dimension attachments — is written atomically. The state badge switches to ✓ saved.

⚠️

Save is all-or-nothing. Unsaved changes exist only in the browser. Refreshing without saving loses your work.

Inspector panel

Clicking any node opens the inspector on the right — three tabs, one node at a time.

Properties

Dimensions

Ext. links

Tab	What you do here
Properties	Set the node label, operand, and value type. Click Apply to confirm — changes are local until you save the graph.
Dimensions	Add or remove documentary dimension attachments. Each dimension you attach here becomes a fill form the Manager must complete for this node.
Ext. links	Manage incoming contributions from nodes in other graphs. See the External parents section for the full workflow.

💡

Apply vs Save. Apply commits changes to the in-memory graph state. Save (toolbar) writes the entire graph to the database. Always Save after Apply.

Node operands

The operand defines the role of a node and determines how it is rendered — colour and border style.

INPUT

Value entered by the Manager at instantiation time. Leaf node — no parents. Dashed border. Primary documentation target.

CONSTANT

Fixed contractual value set at graph construction time. Not filled in by the Manager. Use for rates, thresholds, caps, counts — any value fixed by the clause wording.

SUM

Additive aggregation. Accepts PLUS and MINUS parents. Most common aggregation node.

DIFFERENCE

Explicit subtraction — a base reduced by one or more deductions.

PRODUCT

Multiplication of parent nodes. Models rate × base patterns.

QUOTIENT

Division. Expects exactly one NUMERATOR parent and one DENOMINATOR parent.

Evaluates an expression formed by its incoming links. The expression and its comparison operator are documented in the node label (e.g. "Loss Ratio ≤ 85%"). Emits to exactly one THEN and one ELSE via FACTOR links.

THEN

Value returned when the IF condition is true. Receives the result value from its sources and a routing link from IF via FACTOR.

ELSE

Value returned when the IF condition is false. Same structure as THEN. Can receive a CONSTANT (e.g. zero) when the false branch yields nothing.

LOOKUP

Reads a value from a scale or schedule (sliding scale, reinstatement table). Receives the lookup key via an ARGUMENT parent; returns the corresponding value. Attach the Table dimension to document the schedule.

RESULT

Terminal output node of a graph. Documents the final computed value and serves as the SUBGRAPH hook for cross-graph links. Every graph should end with a RESULT node.

💡

INPUT and CONSTANT are always leaves. They never have parents. Every other node must have at least one parent — otherwise the Manager has nothing to aggregate at that node.

💡

IF / THEN / ELSE are node types, not link types. The condition is the label of the IF node. Incoming links to IF use the same operands as any other node (PLUS for addition, FACTOR for multiplication…). IF always emits to THEN and ELSE via FACTOR links.

Every node must have a value type — the kind of value the Manager will enter: amount, percentage, duration, count, etc. Value types are configured by the Senior.

🔬

All these operands, live in your account from day one

The Profit Commission example ships with every new account — 5 graphs, 46 nodes, covering bottom-up cascade, sliding scale, loss corridor, carry-forward sub-graph and LOOKUP.

→

Link operands

A link connects a parent node to a child node. Its operand describes how the parent contributes to the child's expression.

PLUS

Parent is added. Standard contribution to a SUM, DIFFERENCE, or IF node.

MINUS

Parent is subtracted. Models deductions within a DIFFERENCE or SUM.

FACTOR

Parent is a multiplier. Use with PRODUCT nodes, and for IF→THEN / IF→ELSE routing links.

NUMERATOR

Parent is the numerator of a division. Use with QUOTIENT.

DENOMINATOR

Parent is the denominator of a division. Use with QUOTIENT.

ARGUMENT

Lookup key input. Use with LOOKUP nodes — the ARGUMENT carries the key that drives the scale lookup.

SUBGRAPH

Cross-graph link. Connects a RESULT node in one graph to an INPUT node in another. Declares the inter-graph dependency explicitly.

💡

Link operand = the role of the parent in the child's expression. An IF node evaluating "A + B > 0" receives A and B via PLUS links — same operand as a SUM, because the arithmetic is additive. The IF label carries the semantic ("> 0").

Node type	Expected incoming link operands	Example
SUM	PLUS + MINUS	Gross loss + ALAE − salvage
DIFFERENCE	PLUS (minuend) + MINUS (subtrahend)	Losses − Retention
PRODUCT	FACTOR × n	Rate × Base premium
QUOTIENT	NUMERATOR × 1 + DENOMINATOR × 1	Net loss / Days in year
IF	Any operand — forms the expression in the label	PLUS A, PLUS B → label "A ≤ B"
THEN / ELSE	FACTOR from IF (routing) + any operand from value sources	FACTOR from IF + FACTOR from CONSTANT Rate
LOOKUP	ARGUMENT × 1 (lookup key)	Loss ratio → sliding scale rate
RESULT	PLUS from upstream output node	PLUS from SUM Event Aggregate
INPUT / CONSTANT	None — leaves	—

Dimensions — the creative engine

Dimensions are why RI-TOOL adapts to any organisation without imposing a template. This is where your tenant's identity lives.

🎯

The dimension system is RI-TOOL's core differentiator. Two tenants formalising the same clause can use completely different axes, different granularity, different instructions — and both will be exactly right for their own organisation.

Four layers, three roles

Senior

Categories + Dimensions

Creates named dimension categories and the dimensions themselves, each with fill instructions the Manager will follow line by line.

→

Actuary — you

Node attachments

Pins specific dimensions onto specific nodes via the Dimensions tab of the inspector. One decision per node: what must the Manager document here?

→

Manager

Filled values

Fills one value per node × dimension × instruction line. Result: a fully documented clause instance on a contract.

Attaching a dimension to a node

Select a node → open the Dimensions tab → click ＋ Add dimension → choose from the catalogue built by the Senior. The attachment is local until you save the graph. You can attach multiple dimensions to the same node.

💡

Instructions are the Manager's form. A dimension Amount with three instructions produces three fill fields per attached node. The Senior controls the granularity; you decide where to apply it.

Typical attachment choices by node type

Node type	Typical dimension choices	Reasoning
INPUT (amount)	Currency + Risk + Time	Leaf values are the primary documentation target — always contextualise them.
INPUT (date / count)	Temporal (event window) or Time (period)	Dates and day counts document the temporal scope of the value.
INPUT (market rate)	Time (fixing period)	A market rate is always associated with an observation period.
CONSTANT (rate)	Rate + Table (if sliding scale)	Document the contractual rate and any associated scale.
SUM (event aggregate)	Peril + Spatial + Temporal	The Cat event aggregate is defined by its peril, geographic boundary and time window.
SUM / DIFFERENCE (other)	Amount (result) — optionally scope axes if context changes	Aggregation nodes may need a result annotation; context is usually inherited from children.
IF / THEN / ELSE	Usually none — logic lives in the label and link structure	Attach a dimension only if the condition itself needs documentary context.
LOOKUP	Table (the schedule)	The schedule lives in the Table dimension; the key is documented on the ARGUMENT parent.
RESULT	Usually none — context lives upstream	Terminal node — documentation is on the nodes feeding it.
PRODUCT / QUOTIENT	Usually none	Structural nodes — documentation is on INPUT/CONSTANT parents.

The protection chain

The platform prevents silent data loss in both directions. You cannot remove a dimension attachment if the Manager has already filled values against it. The Senior cannot delete a dimension you have attached to a node. Every layer protects the one below it.

External parents

A node in one graph can receive a contribution from a node in a completely different graph — the mechanism for cross-clause reuse.

Example: Clause A — Net Retained Loss and Clause B — Reinstatement Premium. The reinstatement formula depends on the net retained loss. Instead of duplicating nodes, declare the root of graph A as an external parent of the relevant node in graph B.

How to add an external parent

Select the child node

Click the node in the child graph that will receive the incoming value. The node must already be saved (not a TMP_ node).

Open the Ext. links tab

In the inspector, click the Ext. links tab. Existing external parents are listed. Below them is the Add parent form.

Fill the form

Choose the source graph, the source node, and the operand (PLUS, MINUS, FACTOR…). Click ＋ Add external parent.

⚠️

The child graph must be open when you add the parent. The link is created immediately — no separate save needed for the link itself. However, the canvas refreshes automatically to show the new phantom.

To remove an existing external parent, click the ✕ button next to it in the Ext. links tab. If the child graph is instantiated, a warning is shown — the phantom node referencing this link becomes invalid for existing instances.

Risk links

A Risk link connects an Actuary node to a Risk Modeler node, bringing exposure description into the clause graph.

The Risk Modeler builds graphs of exposure — sites, perils, coverage conditions. Your role is to bridge their work with your clause graphs by declaring which of your nodes relates to which Risk node, and how. This connection is stored in m_graph_risk_link and is qualified by a Risk link operand.

Operand	Meaning	Example
`EXPOSED_TO`	Actuary node relates to a peril	Cat XL loss node ← Windstorm peril
`LOCATED_AT`	Actuary node relates to a site	Property node ← Lyon industrial plant
`COVERED_BY`	Actuary node relates to a coverage condition	Retention node ← Property All Risks coverage
`AGGREGATES`	Actuary node aggregates multiple risk objects	Accumulation node ← Portfolio site

How to add a Risk link

Select a node

Click any saved node in your graph. It must already exist in the database (not a TMP_ node).

Open the Ext. links tab

In the inspector, click the Ext. links tab. Scroll to the 🗺️ Risk links section at the bottom.

Select Risk graph, node, operand

Choose the Risk Graph from the first dropdown, then the Risk node, then the operand that qualifies the relationship. Click ＋ Add Risk link.

Once created, the Risk node appears as a coloured phantom on the canvas — style varies by node type (blue for site, red for peril, green for coverage), with a dashed red-brick border and a label showing the Risk Graph name. The phantom appears immediately, without page reload.

💡

Pre-loaded example. The Risk Modeler has seeded a graph called European Industrial Portfolio — a French industrial site exposed to windstorm and flood, covered by a property all-risks policy. Use it to practice Risk link creation before connecting real exposure data.

⚠️

Risk link operands are tenant-managed. The four default operands (EXPOSED_TO, LOCATED_AT, COVERED_BY, AGGREGATES) cover standard situations. Your Admin can add custom operands in the Risk link operands tab of the Reference section.

Effect on the Manager

Every Risk link you create generates a Risk Dimensions block in the Manager's fill drawer for that node. The Manager sees the Risk node label, the operand, the Risk Graph name, and the exact fill fields defined by the Risk Modeler (address, peril type, insured value…). The Risk dimensions are saved separately from the Actuary dimensions but in a single save action.

Cycle detection

A DAG must be acyclic — no node can be its own ancestor. RI-TOOL detects cycles automatically on every save.

After saving, the platform checks the full graph structure. If a cycle is found, an orange banner appears at the top of the editor and the graph row in the sidebar shows a ♻ cycle badge.

⚠️

Detection is non-blocking. A cyclic graph is still saved — but is semantically invalid and the Manager cannot instantiate it. Fix before use: find the node feeding back into one of its own ancestors, remove the offending link, resave.

Instance lock

Once a Manager instantiates a graph on a contract, that graph's structure is frozen.

The platform blocks any structural modification to an instantiated graph — you cannot save changes or delete it. This protects the Manager's existing data entry from silent invalidation.

🔒

Need to amend an instantiated graph? Create a new version. Use + Graph on the same clause with a new label (e.g. "Net Retained Loss v2"). The Manager instantiates the new version on future contracts; existing contracts keep the original.

Graph Explorer

A production-level view of all your graphs in one place — filterable, inspectable, and exportable.

Click 🔭 Graph Explorer in the Analytics section of the sidebar to open the fullscreen explorer. The inspector panel on the right stays visible throughout — click any card to load its detail.

Filter bar

Six combinable filters apply in real time:

Filter	What it targets
Clause Family	Family of the clause linked to the graph (from the tenant's reference list)
Clause Status	Validation status of the clause — e.g. filter to VALIDATED only
Treaty Type	Any treaty type code associated with the clause
Instances	Instanced (≥ 1 active Manager instance) or Not instanced
Cycle	Graphs flagged with a reference cycle by the last save
Dimensions	Coverage level — Full (100%), Partial, No coverage, or Empty graph (no nodes)

Use ↺ Reset filters to clear all at once. Filter state is persisted in localStorage and restored on your next visit.

Cards and inspector

Each card shows the graph id and label, its clause, and a row of semantic badges — node count, instance count, cycle flag, inter-graph links, and dimension coverage percentage with colour coding. Click a card to load the full detail in the right-hand inspector: graph identity, clause wording, treaty type pills, and a visual coverage bar.

From the inspector, click ↗ Open in DAG editor to jump directly to that graph on the canvas.

Export

The ⬇ Export CSV button in the Explorer header exports the currently filtered view as a CSV — not all graphs, only what the active filters show. Columns: graph id, label, clause id, clause title, clause family, clause status, treaty types (pipe-separated), node count, nodes with dimensions, dimension coverage %, instance count, inter-graph links, cycle flag.

💡

Explorer data is loaded once per session and cached. If you make structural changes in the DAG editor and return to the Explorer, refresh the page to reload.

Data Checks

Live queries on your tenant data, each exportable as CSV — accessible from the Analytics section of the sidebar.

Click 📊 Data Checks to open the fullscreen panel. Each configured view shows a description and a ▶ Run button. Results display as a paginated table with a ⬇ Download CSV link. The DAG toolbar also offers a per-graph ⬇ CSV export of the current graph's nodes and links.

Maintenance

Four cleanup operations accessible from the sidebar footer. Always run the diagnostic first — it shows exactly what will be removed without touching anything.

⚠️

All purge operations are irreversible. Review the diagnostic list carefully before confirming.

Code	What it cleans	When it occurs
A1	Nodes whose parent graph was deleted — cascades to their links and dimension attachments	Graph deleted outside normal flow
A2	Orphan intra-graph links whose parent or child node no longer exists	Node deleted but links not cascaded
A3	Ghost inter-graph links pointing to a deleted graph (parent or child side)	A graph involved in a cross-link was deleted
A4	Orphan dimension attachments (node or dimension deleted) — may also remove associated Manager values	Node or dimension deleted leaving stale attachments

Recommended order: A1 → A2 → A3 → A4. Running A1 first eliminates nodes whose links A2 and A3 would otherwise also catch.