Troubleshooting
Scans that don't run, plan limits, the wrong account, numbers that look off, and known limitations.
A pull request was not scanned
Work down this list; the first two cause most cases.
- Your GitHub name does not match. TrustLedger matches pull requests to your organisation by the GitHub name saved in setup: the part after
github.com/in the repository address. If it is missing, misspelt (for exampleacmeinstead ofacme-corp) or belongs to another TrustLedger organisation, the pull request is ignored and no check appears. Open/onboarding, go to Connect GitHub and correct it. Capital letters do not matter. - The app is not installed on that repository. In GitHub, open the installation settings for the TrustLedger app and check the repository is selected, and that it is installed on the account that owns the repository.
- Check webhook deliveries in the GitHub App settings: a 401 means the webhook secret does not match, so the signature check rejected it.
- Scans run from a queue, so a result can take a little while on a large pull request. Pushing a new commit queues a fresh scan.
The check says “Plan limit reached”
The free early-access plan includes 3 pull-request scans a month and 1 repository. When either is used up, TrustLedger does not scan the pull request and says so on the check instead of failing silently. Existing scans can still be viewed and attested. The scan count restarts on the first day of the month (UTC). Need more now? Write to trustledgerapp.dev@gmail.com.
Setup shows the wrong account or organisation
- Check Signed in as at the top of the setup screens. GitHub sign-in uses whichever GitHub account the browser is signed in to, so a different GitHub login in the same browser signs you in as someone else. Use Not you? Sign out, sign out of github.com, and sign in again, or use a separate browser profile for each account.
- The GitHub name can belong to only one TrustLedger organisation. “Already linked to another TrustLedger organisation” means a different one holds it: sign in as the account that created that organisation, or contact us.
I was signed out on one device after signing in on another
One session is active per person at a time. Signing in on a second device ends the first. This protects the account; sign in again where you want to work.
The numbers on the pull request page do not match
The Security findings card and the Findings column count the same thing: vulnerabilities. Signals are AI-provenance and code-hygiene hints and are never added into the finding counts. A file's risk rating comes from its worst finding, so a CRITICAL file can also hold lower-severity findings.
No check run on the pull request
On GitHub the app needs Checks: write. If permissions were changed after installation, accept the updated permissions in the installation settings. GitLab and Bitbucket report a commit status instead of a check run.
Known limitations
- Detection is not equal in every language. The benchmark report shows recall and false-positive rate per corpus and language. Read the row for your language, not only the headline.
- Generated, minified and vendored files are skipped or treated as third-party code.
- Files over 5,000 lines, or with syntax nested over 1,000 levels deep, are covered by line-level detectors only.
- Flows through code outside the files in the pull request, and through frameworks or libraries the engines do not model, can be missed. See each language's known limitations.
- TrustLedger is not yet SOC 2 or ISO 27001 certified and has not had an independent penetration test. The Trust Center says exactly what is and is not in place.
A finding looks wrong, or a vulnerability was not reported
Read the finding's data flow: every data-flow finding shows how the value got from the source to the sink. If a step is wrong (for example a validation the engine did not recognise), mark the finding a false positive with a reason. Check the coverage matrix for the rule and language before reporting a miss. Then tell us through Contact and feedback: false positives and missed vulnerabilities are the most useful feedback we get.
A file was not analysed by its language engine
See the known limitations above: generated, minified, vendored and very large files get line-level detectors only.