Wellbore Genius
Help
Dashboard

Core product status

The 2026-09-04 health pass of the frac simulation path, kept live. Move a finding as its fix lands and the report updates here; the recorded baseline stays visible underneath so you can always see what the pass originally found.

2 landed · 1 in progress · 1 open · 0 won't fix of 4 findings

50% closed

Health pass driven through a real browser, signed in and signed out, across the six routes that make up the core simulation path. All six load, every runnable page runs a case, and no NaN leaked into a results panel.

Route health

Core simulation routes checked in a real browser
RouteHTTPRuns a caseConsole errorsNotes
/pad-parallel200yesnoneInstant preview 0.8 ms across 20 stages; verified solve matched the committed settings. CPU kernel used — no WebGPU adapter in headless.
/pad-orchestrator200yesnoneRun strip present once a case is selected.
/field-pace200yesnoneGuided step lane intact end to end.
/job-simulation200yesnone
/physics-verify200n/anonePass/fail range table rather than a solve.
/dashboard200n/anone
  • First cold navigation to /pad-parallel and /field-pace exceeded 45 s in dev because Vite compiles the route on first visit; both were instant on the second load. That is a dev-server artifact, not a production path.
  • The WebGPU compute path was exercised as CPU fallback only — headless Chromium exposes no adapter.

What the simulator does well today

Explicit fidelity contract

Verified physics tiers (depletion field, stage-to-stress coupling, fibre forward model) are named on screen and stamped into exports, so a fast preview can never be mistaken for a signed-off result.

Instant and verified are separated

The approximate preview is labelled as approximate and has its own compute selector (CPU / Mixed / WebGPU) with graceful CPU fallback when no adapter exists.

Whole-pad parallel solve

The pad converges in waves rather than serially per stage, and the compare view diffs Instant against Verified on identical inputs.

Drift-guarded physics

Physics libraries are covered by unit tests and snapshot drift-guards, with /physics-verify surfacing pass/fail ranges in the app.

Findings, highest leverage first

Verification breadth beyond single-fracture cases
In progress
leverage #1
/physics-verify covers bedding-plane (He-Hutchinson) and TIV (Thomsen) cases only. Extending verification to the pad-scale path — stress-shadow superposition, proppant transport, and the Instant vs Verified delta — turns it into a single credibility page.
See the evidence

Showing the note recorded in the health pass.

No published benchmark case
Landed
leverage #2
A skeptical engineer had no one-click, reproducible public dataset run to compare against the literature. This was the strongest missing trust asset.
See the evidence

Showing the note recorded in the health pass.

WebGPU path unverified in CI
Open
leverage #3
Render and compute quality tiers are pinned to balanced for CI, so the cinematic/WebGPU path has no automated coverage; a regression there is only visible by hand.

Showing the note recorded in the health pass.

Guest banner promised access the gate refuses
Landed
The banner claimed "full features, works without an account" while AccessGate blocks simulation, results, 3D and fleet routes for anonymous visitors.

Showing the note recorded in the health pass.