Order likely causes of a crash from its error trace

A ranked table of possible crash causes with the check that would confirm or reject each.

Choose one of three approaches. The steps for that option appear below.

Choose an option above. On a phone, swipe the tabs left or right to see all options →

Claude: maps the trace, ranks possible causes

Claude maps the trace frames, then drafts possible causes with checks. You check the map and run the checks.

  1. You

    Gather trace

  2. Claude

    Map frames

  3. You

    Check frame map

  4. Claude

    Draft possible causes

  5. You

    Run checks, decide

  1. Gather the trace and commit summaries

    Use: You

    Have ready: the stack trace and the recent commit summaries.

    Check

    Check the trace still shows every frame and line number.

  2. Map the trace frames to modules

    Use: Claude

    Ask Claude to parse the pasted trace frames and map them to modules.

    Details

    What to paste into the tool

    Task: Frame-to-module table (frame, module, line). Here is a stack trace. Parse every frame and map it to a module. Output a table: frame, module, line. Mark anything that does not map as unresolved instead of guessing.

  3. Check and correct the frame map

    Use: You

    Compare the frame map with the repository and correct it.

    Check

    Check each frame's module path and line against the repository at the deployed version; fix wrong or unresolved mappings.

  4. Draft possible causes with a check each

    Use: Claude

    Ask Claude to draft possible causes, each with a confirm-or-reject check. Mark the ranking as unchecked reasoning.

    Details

    What to paste into the tool

    Task: Ordered hypothesis table: hypothesis, evidence from frames/commits, the specific check that would confirm or reject it. Using this checked frame map, the stack trace and the commit summaries, draft hypotheses for the crash. Order them by how well they fit the frames and recent changes, and attach to each one a specific check that would confirm or reject it. Mark the output as unchecked reasoning, not a diagnosis (source constraint: 'Mark ranking as unchecked reasoning.').

  5. Run the checks, then approve or hold

    Use: You

    Run the confirm-or-reject checks, top cause first, and record each outcome. The list stays not-yet-proven until a check confirms a cause.

    Check

    Check each cause names a specific, runnable check. Run the top check, and more if time allows, and record each outcome. Approve the ranking only for causes a check has confirmed. Possible causes are not proven diagnoses: do not apply a fix before its check has been run, and treat the ranking as a best guess until checks confirm it.

Before you paste into any AI tool, remove any confidential information. Put it back at the end.

Tools in this workflow

Deciding between ChatGPT and Claude for this? See the comparison.

Final check

Developer verifies each frame's module path and line against the repository at the deployed version.

The output is marked as unchecked reasoning in every path.

Each hypothesis must name a specific check; the developer records the outcome of the top check, and more if time allows.

Hypotheses are not diagnoses; no fix is applied without verification.

Check the result against your original inputs. Approve it before sharing or acting.

What still needs checking
  • Exact paste/input limits for long traces are not published on the cited pages.
  • Whether a Copilot Free, Pro or Pro+ user has opted out of training use is an account-level setting not verifiable here.

Share this page