Troubleshooting
Use this page when the platform loads but a run will not start, a target fails to load, or account verification gets stuck.
WebGPU Not Available
What this means
LatticeZero scoring and docking run on WebGPU. If the browser cannot create a WebGPU adapter, the app can still load its docs and account pages, but GPU scoring features stay unavailable.
Common causes
- Unsupported browser build
- Hardware acceleration disabled
- Out-of-date GPU drivers
- Remote desktop or VM session without GPU passthrough
- Enterprise browser policies that disable WebGPU
What to check
- Use a current Chromium browser such as Chrome or Edge.
- Turn on hardware acceleration in browser settings.
- Open
chrome://gpuand confirm WebGPU is enabled on the active adapter. - Update NVIDIA, AMD, or Intel graphics drivers.
- If you are on a laptop with integrated and discrete GPUs, force the browser onto the discrete GPU.
- If you are in a VM or remote session, confirm GPU passthrough is actually available.
Expected fallback behavior
- The app should explain that WebGPU is unavailable.
- Documentation pages, account pages, and non-scoring surfaces should still work.
- There is no hidden server-side docking fallback for the web scoring path; a WebGPU-capable browser is required for live runs.
Scoring Errors
Run stalls or times out
Typical causes:
- Large ligand set with an aggressive accuracy preset
- Browser tab suspended in the background
- GPU resource exhaustion from other tabs or applications
- A demo bundle that loaded partially and never finished preparing the scoring path
What to do:
- Refresh the page and rerun the target.
- Drop to a smaller ligand subset before retrying a full run.
- Close other GPU-heavy tabs and desktop apps.
- Keep the tab in the foreground while the run initializes.
Receptor or target bundle fails to load
Typical symptoms:
- Empty target list
- "Target not found"
- Blank run panel after selecting a target
- Repeated fetch failures in the browser console
What to do:
- Reload the page once to clear partial bundle state.
- Try a second target; if multiple targets fail, the issue is probably global rather than target-specific.
- Check the browser console for missing bundle files such as
manifest.json,receptor.pdb, or ligand assets. - If the failure is isolated to one bundle, compare its static files with a known-good target in
/static/wasm/demo_assets/v3/.
Scores are zero, NaN, or obviously wrong
Typical causes:
- The pose is outside the intended pocket
- The wrong scoring mode or profile is being used
- The target bundle loaded but the scoring metadata is stale
- WebGPU initialized, but the scoring path failed partway through setup
What to do:
- Confirm the target bundle loaded completely before running.
- Re-check the selected scoring profile and mode.
- Compare the result against the published AUC for that target on Benchmark Results.
- For lab work, compare
dockAucwith the post-rescore metric to see whether the issue is the search stage or the scoring stage.
Email Verification Issues
Verification email never arrives
- Check spam and promotions folders first.
- Wait a few minutes before requesting another message.
- Confirm the account was created with the same address you are checking.
- If available, use the resend action instead of creating a second account.
Verification link opens but the account still looks unverified
- Open the newest verification email only.
- Complete the flow in the same browser session you plan to use for sign-in.
- If you requested multiple emails, older links may have been superseded.
- Sign out and back in after verification to refresh session state.
Magic-link or verification flow expires
- Request a fresh email and use the newest link.
- Avoid forwarding old verification links between devices.
- If the app loops back to sign-in, clear the stale tab and retry from the newest message.
Common Docking Questions
Why is my AUC low?
A low AUC usually means one of three things:
- the pose search is not landing actives in the right geometry
- the scoring profile is not separating actives from decoys for that target
- the current bundle is intentionally retained as an honest-but-weaker product path
Use these checks:
- Compare
dockAucagainst the post-rescore AUC. - If both are low, the search setup or bundle quality is the likely issue.
- If
dockAucis reasonable but post-rescore drops, the scoring profile or rescore configuration needs attention.
What does cascade rescore do?
Cascade rescore is the second-stage ranking pass applied after docking. The first stage finds poses and produces a raw dock ranking. The cascade stage then extracts physics features from those poses and applies the target-specific scoring profile. On some surfaces the post-docking metric is labeled rescoreAuc; in discussion it is often referred to as cascadeAuc.
In practice:
dockAuctells you how well the pose-search stage alone separates actives from decoys.cascadeAuc/rescoreAuctells you how well the final ranking performs after profile-based rescoring.
When should I trust the dock ranking over the rescore ranking?
Use the raw dock ranking only when the bundle or validation notes explicitly say the dock ordering is the retained gold path. Otherwise, the post-rescore ranking is the number to compare against the published benchmark.
When To Escalate
Escalate a bundle issue when all of the following are true:
- WebGPU is working on the machine
- the target assets load cleanly
- rerunning does not change the failure
- the live result still disagrees materially with the published target AUC or expected behavior
When filing the issue, capture:
- browser name and version
- OS and GPU model
- target ID
- whether the failure happened in Demo, Run, or IsoPose Lab mode
- any console error text