# PH-CON-002

**Name:** Energy conservation violation

**Severity:** error

**Input modes:** adapter+dump

**Scope separation (read first):** PH-CON-002 validates the production
rule's ability to measure **wave-energy drift on analytically
controlled conservative snapshots**. It does not certify the accuracy
of a wave-equation time integrator.

The F2 fixture splits into two distinct layers:

- **F2 harness-level (authoritative):** `E(t)` computed directly from
  analytical `u_t, u_x, u_y` — roundoff-only drift.
- **Rule-verdict contract:** analytical `u` snapshots fed to the
  rule; rule computes `u_t` via 2nd-order central FD internally →
  `O(Δt²)` drift with method-dependent tolerance.

Plan §17's "leapfrog time-stepper over 1000 steps with 1e-8
tolerance" is **not implemented** in V1 — PH-CON-002 does not use a
leapfrog stepper (it uses central-FD on supplied `u` snapshots), and
any numerically-evolved fixture would be supplementary with its own
tolerance per user's revised Task 9 contract.

Pure numpy — no mesh assembly, no torch / scikit-fem.