Errors / issues
Errors / issues
Read this reference only when the user has explicitly asked about errors, bugs, JavaScript crashes, HTTP failures, broken pages, or revenue-loss projections — or after analytics has pointed at an error hypothesis and the user has confirmed they want to investigate.
Do NOT include these tools as a routine step in a cohort analysis, "user behaviour analysis", or "UX improvement" plan. For "what to fix" / "top opportunities" / open-ended improvement questions, lead with noibu_search_sessions and noibu_get_page_visits instead.
Error tools
(In the Noibu console these are labelled "Issues" — we call them errors at the API layer.)
noibu_list_priority_errors — Quick view of the top priority errors ranked by occurrence. Only use when the user explicitly asks about errors or bugs. Applies the importance filters below server-side — prefer it over noibu_search_errors for priority questions.
noibu_search_errors — Advanced error search with filtering and sorting. Use when the conversation turns technical: specific errors, bugs, error types, stack traces, or when investigating root causes behind an analytics pattern.
noibu_get_error — Full detail for a single error including stack trace and error info. Use after ErrorsSearch or GetPriorityErrors to drill into a specific error.
noibu_get_error_diagnosis — AI-generated explanations for errors. Pass error UUIDs from ErrorsSearch to get plain-language explanations of what the error is.
noibu_get_error_trends — Error occurrence trends over time. Use to chart how errors are trending (getting worse or better).
How "Important" Works — and ARL is NOT It
When the user asks anything shaped like "important", "priority", "what to fix", "what to focus on", "what's affecting my users", or "what matters most" — apply this filter and sort. Do NOT rank by ARL.
Filter (apply all four):
| User says | API |
|---|---|
| State: Unset, Open, In progress | STATE ∈ [new, open, in-progress] |
| Conversion impact: Verified, Unknown, Likely | MANUALLY_VERIFIED ∈ [IMPACT, UNKNOWN, LIKELY] |
| Insight is any: Caused by click, Lost sessions, Frequently broken image | INSIGHTS ∈ [ALL_USERS_CLICKED, LOST_SESSIONS, FREQUENT_IMAGE_ERROR] |
| Insight is not: Invalid coupon, Script error, HTTP undefined error, Dormant | INSIGHTS ∉ [INVALID_COUPON, SCRIPT_ERROR, HTTP_ZERO, DORMANT] |
If the user uses a left-column label, translate to the right.
Sort: MANUALLY_VERIFIED DESC, then REV_LOST DESC (actual — NOT REV_LOST_ANNUALIZED).
Columns to surface: ID, Title, Conversion impact, Revenue lost (revLost), Occurrences. ARL is NOT a column for importance answers.
noibu_list_priority_errors applies these filters server-side — prefer it.
Verification terminology: "verified errors" / "verified issues" = MANUALLY_VERIFIED = "IMPACT" (human-confirmed). Do NOT use "LIKELY" — that's an AI prediction, not verified.
ARL is Directional, NOT Factual
revLostAnnualized (ARL) and revLost are Noibu projections computed from conversion-rate variance × cart value × leads, extrapolated to a year. Correlation-based, assumption-dependent, artifact-prone.
Always caution when you surface revLost or ARL. Footer-style note under the data, e.g.:
Note: revLostAnnualized is a Noibu projection — directional, not measured causal loss. Fixing the issue may recover more, less, or roughly the projected amount.
Phrasing rules:
- Hedge: "projected", "estimated", "associated with ~$X". NOT "is costing $X", "is bleeding $X", "you're losing $X" — those imply measurement.
- No recovery promises ("you'll recover $X if you fix this").
- Sums across issues can double-count overlapping affected sessions.
Do not use ARL as a proxy for importance. Importance = the filter+sort above, not dollar ranking. Even within IMPACT-verified results, sort by REV_LOST (actual), NOT REV_LOST_ANNUALIZED.
Do NOT surface revLost or ARL when:
- The user didn't ask about revenue/money/financial impact. Lead with verification, symptoms (RageClick, BrokenButton, lost_sessions), and affected-session counts.
- Hard-zero categories (CSP, script, resource, page-check, fetch-API). $0 ARL here is a measurement gap, not "no impact".
- Brand-new issues (< ~7 days). ARL hasn't converged.
- Sitewide issues. Variance-based ARL collapses to ~$0 — no comparison group.