Grouping Noisy Errors

The same error can be reported as dozens of separate issues when stack frames or messages contain volatile data — random IDs, dynamic file paths, build hashes, anonymous function names. The fix is two-step: merge the existing issues into one target, then create a grouping rule so future events from the same call site share a single canonical fingerprint instead of spawning new ones.

Important up front: “same error” here is narrow. Two issues that share a name or a sentence of message text but came from different code paths, different SDKs, or different runtimes are different errors and should stay separate, even if the user thinks of them as “the same kind of bug”. Grouping a frontend TypeError together with a backend TypeError because both messages contain “undefined” destroys the signal that lets the team find each one. The criteria in step 1 exist to keep that from happening.

Available tools

Merge vs grouping rule

The two tools solve different halves of the problem:

• Merge is one-shot. It collapses existing issues into a target and re-attaches their events. Future events still group by their original fingerprints — if the same noisy pattern keeps producing new fingerprints, merging is a treadmill.
• Grouping rule is durable. It rewrites the fingerprint of any matching event to custom-rule:<rule_id> at ingestion time, so all future matches share one canonical fingerprint rather than spawning new ones. The first match either creates a new issue keyed off that fingerprint, or routes to whatever issue is already bound to it.

Use both together when the issue is recurring: merge historical duplicates into a target issue, then create the rule. The rule API does not accept a target issue ID — once the rule starts firing, the resulting custom-rule:... issue can be merged into the same target so the consolidation sticks. Use merge alone for historical sprawl that you don’t expect to recur. Use a grouping rule alone for a brand-new pattern you’re getting ahead of, when you don’t need to consolidate with an existing issue.

Workflow

Step 1 — Confirm the duplicates

Search by exception type or message to find candidates:

posthog:query-error-tracking-issues-list
{
  "searchQuery": "TypeError: Cannot read property",
  "status": "active",
  "limit": 50,
  "orderBy": "occurrences",
  "dateRange": { "date_from": "-30d" }
}

For each candidate, pull one sampled exception event to compare stack, type, and message:

posthog:query-error-tracking-issue-events
{
  "issueId": "<candidate_issue_id>",
  "limit": 1,
  "verbosity": "stack"
}

Run this once per candidate. The tool defaults to onlyAppFrames: true, which makes the top in-app frame stand out at a glance. If two candidates share the same top frame and same exception type, they’re likely the same error — but verify against the full checklist below before merging.

Are they the same error?

Treat two issues as duplicates only when every one of these matches:

• $lib is the same SDK (posthog-js, posthog-python, posthog-node, posthog-android, etc.). Errors from different SDKs almost always come from different code paths even when the exception type matches.
• The exception type is identical ($exception_types).
• The top in-app stack frame points at the same file and same function. Line numbers and minor offsets within that function are fine; a different file or a different function on top means a different bug.
• The message follows the same template, with differences confined to volatile data — IDs, hashes, timestamps, dynamic paths. If the difference is a different verb, object, or operation, it’s a different bug.
• $exception_handled agrees (both handled or both unhandled). A caught variant and an uncaught variant are different code paths and benefit from staying separate.

If any single one of those differs, they are not duplicates — investigate separately (investigating-error-issue).

What NOT to group together

These are the failure modes that destroy debugging signal. Do not group across any of them, even when the user describes them as “the same kind of bug”:

• Frontend and backend variants of the same exception type. A TypeError from a browser bundle and a TypeError from a Node service share a name and often a message word, but the stack, the runtime, and the fix all differ.
• Different SDKs / platforms. posthog-js vs posthog-python vs posthog-android are different call sites.
• Same type, different file or function on top of the stack. A NullPointerException thrown from OrderService.cancel is not the same bug as one thrown from PaymentService.refund, even if both messages say “user was null”.
• Caught vs uncaught. Two issues that differ only in $exception_handled are usually a code path that swallows the error in one place and lets it propagate in another — keeping them separate makes that visible.
• Conceptually-similar bugs that happen to share a phrase. “Cannot read property of undefined” appears in many independent bugs. Without matching stack frames, message similarity alone is not enough.

Step 2 — Pick the target issue

Pick the issue that should absorb the others:

• Most occurrences — keeps the dominant issue so dashboards stay continuous
• Best name and description — if the user has annotated one, prefer it
• Earliest first_seen — preserves the original timeline

Note the target’s ID. The other candidates become ids to merge in.

Step 3 — Merge existing duplicates

posthog:error-tracking-issues-merge-create
{
  "id": "<target_issue_id>",
  "ids": ["<duplicate_id_1>", "<duplicate_id_2>", "..."]
}

Merge is destructive (annotation destructive: true) — once issues are merged into a target, the source issues are gone from the active list. Confirm the target with the user before calling. Cap each merge call at ~50 source IDs to keep failures localized; for larger sprawl, batch.

Merged changes may not appear in the issue list immediately — re-listing right after the call can still show the source issues for a short window. If a follow-up error-tracking-issues-list call looks unchanged, wait a few seconds and re-query rather than re-issuing the merge.

If after the merge the target’s metadata looks wrong (a duplicate had a better name), use error-tracking-issues-partial-update to fix the name or description on the target rather than re-merging.

Step 4 — Decide if a grouping rule is warranted

A grouping rule is worth creating when both are true:

• The pattern keeps producing new fingerprints (you have seen new duplicates appear since the last merge)
• You can describe the pattern with property filters that won’t accidentally swallow unrelated errors

The canonical exception properties ($exception_types, $exception_values for messages, $exception_sources for file paths, $exception_functions for function names) are arrays at capture time. The property filter compiler special-cases them (opens in a new tab) — it parses the JSON-materialized column and wraps the filter in arrayExists(v -> ..., JSONExtract(...)), so all the standard operators (exact, is_not, icontains, not_icontains, regex, not_regex) work against individual elements with the bare value: exact "TypeError", not exact '["TypeError"]' or regex '"TypeError"'.

The singular forms ($exception_type, $exception_message) and $exception_stack_trace_raw are emitted on a fraction of a percent of events; filtering on them produces a rule that silently never matches.

If the volatility is in the message (e.g., TypeError at /static/main.<hash>.js), a regex filter on $exception_values works. If the volatility is in line numbers within a known file, icontains on $exception_sources does. $exception_handled is also a useful narrowing dimension — separate handled vs unhandled rather than mixing them.

Skip the grouping rule when:

• The duplicates are historical (one-off backfill, no new occurrences) — merge is enough
• You can’t write a filter narrow enough to be safe — broaden the merge cadence instead and revisit later

Step 5 — Create the grouping rule

Translate the step 1 “same error” checklist into rule filters. A rule that matches more loosely than the checklist will silently merge unrelated bugs forever — the rule is more dangerous than the merge because it runs against every future event. At a minimum, scope by SDK and exception type, and add a third dimension (file path via $exception_sources, or a specific message phrase via $exception_values) to pin the call site:

posthog:error-tracking-grouping-rules-create
{
  "filters": {
    "type": "AND",
    "values": [
      {
        "type": "event",
        "key": "$lib",
        "operator": "exact",
        "value": "posthog-js"

Rules are evaluated in order. List existing rules first (posthog:error-tracking-grouping-rules-list) — if a rule already partially covers the pattern, prefer adjusting its filter over stacking a near-duplicate.

The optional assignee field auto-assigns issues created by the rule. Skip it unless the user explicitly wants ownership baked into the rule.

Step 6 — Verify and consolidate

Sample the merged issue’s recent events to confirm the merge succeeded. Watch for the rule’s custom-rule:<rule_id> fingerprint to start matching events — the first match creates a new issue (or routes to whatever was already bound to that fingerprint). To keep events under your historical target rather than scattered across the new custom-rule issue, run a second merge folding the custom-rule issue into the target.

If new (non-rule) fingerprints continue appearing despite the rule, its filter is too narrow — widen it.

Tips

• The user often confuses grouping rules with assignment rules. Grouping rules decide which issue an event lands in. Assignment rules decide who owns the resulting issue.
• Don’t merge issues that “look similar” without inspecting events. Two TypeErrors in different files are different bugs.
• Stack frames are the canonical grouping signal — ingestion already fingerprints on the stack, so a stable stack groups itself. A grouping rule is for cases where the natural fingerprint sprays (volatile filenames, hashed function names, dynamic line numbers) and you need to override it.
• Disabling or tightening a grouping rule does not retroactively un-group existing events; future events route correctly, past events stay where they are. Use error-tracking-issues-split-create if you need to surgically separate fingerprints back out of a merged issue.
• Grouping rules are visible in the UI under Project settings → Error tracking → Grouping rules; mention this when the user asks where rules live.