add Neo4j heuristics calculator by andyfelder16 · Pull Request #1647 · WebFuzzing/EvoMaster

andyfelder16 · 2026-07-21T13:03:49Z

First of a few PRs splitting #1641, per review feedback.

Computes H(Q,G): how close a graph is to satisfying a parsed Cypher MATCH query, as a Truthness <ofTrue, ofFalse>
Neo4jHeuristicsCalculator, Neo4jStructuralMatcher (matched_elements), Neo4jConditionEvaluator (condition + operand evaluation), Neo4jMapping
Neo4jNode/Neo4jEdge/Neo4jGraph: in-memory graph model
Works in Truthness end-to-end, mirroring the SQL heuristics calculator
11 tests

…ristic

…ound

jgaleotti · 2026-07-31T14:12:31Z

@andyfelder16 is this ready for re-review? Please do not forget to ask for a re-review whenever ready.

arcuri82 · 2026-08-04T07:47:42Z

+     */
+    private Truthness equalityTruthness(Object a, Object b) {
+        if (a == null || b == null) {
+            return null;


in line of princple, i don't like passing null like this when computing heuristics score. in DistanceHelper, we have things like H_REACHED_BUT_NULL which is lower than H_NOT_NULL. i think we should "score a gradient" in these cases, and not rely on null

if it makes sense, you could even introduce a new value that is lower than H_REACHED_BUT_NULL (but still > 0), eg H_INVALID, and have a special new boolean field in Truthness called like invalid. if a truthness object is invalid, its heuristics is H_INVALID. any operation on invalid (eg AND or OR) would result in an invalid result. this would avoid passing around null. @jgaleotti what do you think?

introducing this new H_INVALID could be done in a new PR before this one is merged

Agreed on not returning null, and scoring a low value instead would already help the gradient.

The part I'm less sure about is having invalid propagate through AND/OR. Small example:

MATCH (p:Person) WHERE p.age = 25 AND p.nickname = "bob"

against a node (:Person {age: 27}), which has no nickname property. The conditions score:

condition ρ (ofTrue)

p:Person 1.0 label matches

p.age = 25 0.333 1 - normalizeValue(|27-25|)

p.nickname = "bob" ? property is absent, nothing to measure

Depending on what we do with that third one:

today, skipped: avgOfTrue = (1 + 0.333) / 2 = 0.667, and mutating age to 25 moves it to 1.0

scored low and kept in the average: avgOfTrue = (1 + 0.333 + 0.05) / 3 = 0.461, and the same mutation moves it to 0.683

invalid absorbing the AND: the whole conjunction is H_INVALID, and mutating age changes nothing

So a low ofTrue that still takes part in the aggregation gives us the "no nulls" property without losing the gradient, while an absorbing invalid loses it exactly on the queries where some predicate can't be valuated.

Happy to go either way, just let me know which one you prefer and I'll implement it

@andyfelder16 indeed, we should not lose the gradient! ;)
however, I need some clarifications here. Assume same query but with an OR:

MATCH (p:Person) WHERE p.age = 25 OR p.nickname = "bob"

now, consider the node (:Person {age: 25})... what would be the result here? null? we need to handle such case properly (and have at least 1 test to verify it)

Not null, it gives <1.0, 0.1> and distance 0, which matches Cypher semantics: true OR null is true, and Neo4j does return that node. The unvaluatable branch is dropped and the OR is satisfied by the age one

I also ran the neighbouring cases to see where the current skipping breaks down:

query node distance today Neo4j returns

p.age = 25 OR p.nickname = "bob" {age: 25} 0.0 the node

p.age = 25 AND p.nickname = "bob" {age: 25} 0.0 nothing

p.nickname = "bob" {age: 25} 0.0 nothing

The last two are false positives. In Cypher true AND null is null, so neither query matches anything, but we report them as fully satisfied. That is the worst case for the search

My take is that this settles it in favour of your suggestion. Scoring a low value instead of skipping fixes both cases: taking 0.05 as that value, the AND goes from 0.0 to about 0.107, while the OR stays at 0.0 because the aggregation still picks up the satisfied branch

I'll add tests for the three cases above

jgaleotti · 2026-08-06T20:03:41Z

+    }
+
+    private Truthness computeHeuristicPattern(MatchPattern pattern, Neo4jGraph graph, List<Neo4jMapping> mappings) {
+        Truthness nodes = computeHeuristicMatchNodes(pattern.nodeCount(), graph.nodeCount());


rename to hNodes and hEdges. Otherwise it might be confused with the actual nodes and edges

jgaleotti · 2026-08-06T20:05:01Z

+
+    private Truthness edgesForMapping(List<PatternEdge> patternEdges, Neo4jGraph graph,
+                                      Neo4jMapping mapping) {
+        Truthness[] perEdge = new Truthness[patternEdges.size()];


rename to truthnessPerEdge

jgaleotti · 2026-08-06T20:05:14Z

+
+    private Truthness computeHeuristicPattern(MatchPattern pattern, Neo4jGraph graph, List<Neo4jMapping> mappings) {
+        Truthness nodes = computeHeuristicMatchNodes(pattern.nodeCount(), graph.nodeCount());
+        Truthness edges = computeHeuristicMatchEdges(pattern.getEdges(), graph, mappings);


rename to hEdges or similar

jgaleotti · 2026-08-06T20:06:20Z

+    }
+
+    /** Existence-only edge check: TRUE if some graph relationship matches the edge's endpoints. */
+    private Truthness edgeMatch(PatternEdge edge, Neo4jGraph graph, Neo4jMapping mapping) {


edge -> patternEdge ?

jgaleotti · 2026-08-06T20:06:46Z

+    private Truthness edgeMatch(PatternEdge edge, Neo4jGraph graph, Neo4jMapping mapping) {
+        Neo4jNode source = mapping.getNode(edge.getSourceVariable());
+        Neo4jNode target = mapping.getNode(edge.getTargetVariable());
+        if (source == null || target == null) {


can source or taget be null?

jgaleotti · 2026-08-06T20:06:59Z

+        if (source == null || target == null) {
+            return FALSE_TRUTHNESS;
+        }
+        for (Neo4jEdge rel : graph.getEdges()) {


rel is an actual edge

jgaleotti · 2026-08-06T20:08:57Z

+        if (conditions.isEmpty()) {
+            return TRUE_TRUTHNESS;
+        }
+        List<Truthness> truths = new ArrayList<>();


rename truths to listOfTruthness

jgaleotti · 2026-08-06T20:09:22Z

+        }
+        List<Truthness> truths = new ArrayList<>();
+        for (CypherCondition c : conditions) {
+            truths.add(evaluator.evaluateCondition(c, mapping));


store evaluator.evaluateCondition to a local variable before adding to the list

score cypher match queries against a graph with a truthness-based heu…

588b240

…ristic

andyfelder16 requested a review from jgaleotti July 21, 2026 14:16

cleaned test file

af9a624

jgaleotti requested changes Jul 21, 2026

View reviewed changes

dispatch cypher conditions through a visitor and fix review comments

2d0e890

andyfelder16 requested a review from jgaleotti July 25, 2026 15:00

jgaleotti requested changes Jul 28, 2026

View reviewed changes

Comment thread ...main/java/org/evomaster/client/java/controller/neo4j/heuristics/Neo4jConditionEvaluator.java Outdated

Comment thread ...roller/src/main/java/org/evomaster/client/java/controller/neo4j/heuristics/Neo4jMapping.java Outdated

guard neo4j mapping lookups so a null variable name can't pass as unb…

f1551bc

…ound

andyfelder16 requested a review from jgaleotti July 31, 2026 22:36

jgaleotti approved these changes Aug 3, 2026

View reviewed changes

jgaleotti requested a review from arcuri82 August 3, 2026 14:08

arcuri82 requested changes Aug 4, 2026

View reviewed changes

align the neo4j heuristics with evomaster review conventions

659b89b

andyfelder16 requested a review from arcuri82 August 4, 2026 13:59

score unvaluatable cypher conditions instead of skipping them

85e728e

jgaleotti requested changes Aug 6, 2026

View reviewed changes

condition	ρ (ofTrue)
`p:Person`	`1.0`	label matches
`p.age = 25`	`0.333`	`1 - normalizeValue(\|27-25\|)`
`p.nickname = "bob"`	?	property is absent, nothing to measure

query	node	distance today	Neo4j returns
`p.age = 25 OR p.nickname = "bob"`	`{age: 25}`	`0.0`	the node
`p.age = 25 AND p.nickname = "bob"`	`{age: 25}`	`0.0`	nothing
`p.nickname = "bob"`	`{age: 25}`	`0.0`	nothing

Uh oh!

Conversation

andyfelder16 commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jgaleotti commented Jul 31, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

andyfelder16 commented Jul 21, 2026 •

edited

Loading