Skip to content

PHP: clearly-wrong calls edges in the repro neighborhood (.parent() --calls--> .setUp(); .handle() --calls--> test methods) — failure mode #3 evidence #39

Description

@filipechagas

Observation (evidence capture, not yet diagnosed)

Measured 2026-08-06 on the api.lawnstarter.com repro graph @ d2d4bed7ae (48,529 nodes), probing call-context edges around ChargeCustomerService:

  • app_servicecategory_servicecategory_parent (.parent()) --calls--> .setUp() of both ChargeCustomerServiceTest variants and of BalanceCustomerAccountServiceTest — a model accessor does not call test fixtures.
  • app_services_billing_chargecustomerservice_chargecustomerservice_handle (.handle()) --calls--> several ChargeCustomerServiceTest test methods (e.g. .test_integration_execute_payment_failure_charge_successfully_charges_the_customer()) — production code does not call tests; the direction/target is wrong.

Repro: load the graph and list edges with context == "call" whose endpoint ids contain chargecustomerservice / balancecustomeraccountservice (probe script preserved in #37's step-through notes).

Why it matters

These edges feed the "all test noise" impression in #37's phrasing-A output and inflate every call-filtered traversal near billing. This is the PE-78803 spike's failure mode #3 (node-identity / wrong-edge territory) — kept deliberately separate from #37's seeding fix per its scope decision.

Next step

Diagnose the minting path (bare-name legacy resolution vs. member-call resolver vs. stub shadowing) before proposing a fix; this issue exists to carry the measured evidence and keep #37 unconflated.

Metadata

Metadata

Assignees

No one assigned

    Labels

    ready-for-agentFully specified, ready for an AFK agent

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions