What the analytics numbers mean — the questions behind each report

Who this guide is for: customer ADMINISTRATORS and team leads who need to trust and compare analytics numbers. The analytics reports guide walks the screens step by step; this companion explains what each kind of report actually counts, which mistakes are refused instead of silently corrected, and who is allowed to ask.

About the evidence. Every attempt row, funnel and template in the run behind this guide was seeded by the test. No live WhatsApp/Meta message contributed to any number here.

Account totals vs campaign totals — two different questions

Practical tip: when two dashboards disagree, check first which of the two questions each one was asking.

Custom date ranges — and the three ways a bad range is refused

A custom range covers whole UTC days. When the range is wrong, the product refuses instead of guessing — three distinct messages:

Message The mistake What to do
RANGE_IN_FUTURE The end day is ahead of now Use an end day that has already happened
RANGE_TOO_WIDE The span is over 366 days Narrow the range
RANGE_INVALID The start is not before the end Swap or correct the days

One phone number's own truth

The number-activity report takes one phone number (international format) and returns that recipient's own counts for the window you choose. An unknown number is a valid empty report, never an error — the honest answer to "we have no history with this number". A malformed number is refused with NUMBER_INVALID.

Per-template delivery — including the honest zero

Per-template rows use that template's own recorded count as the denominator — one template's percentages are never diluted by another's volume. A window with no activity renders explicit zero rows — a real answer of "nothing happened", not a blank and not an error.

Why messages failed — grouped, with an explicit "unknown"

The failure-reasons report groups failures by the provider's documented error codes, with an explicit unknown row for failures carrying no code (seed rows excluded). The unknown row is a feature: it tells you plainly how much of your failure mass is unexplained.

The bot subject

Analytics also covers bots: a bot's run totals and run history load without disturbing its campaign metrics. The bot lifecycle itself is documented in the bots guides.

Who may ask

What this guide does NOT claim (kept honest)


Evidence record (kept for honesty). The FULL3747-O14EV lane executed suite o14-ev3747-s1-analytics.pg.test.mjs — 14/14 cases passed, commit 808f4d3e on the combined candidate (C:/CT-ZC-FULL3747, branch codex/zcode-full3747) — the real compiled application over real loopback HTTP against a throwaway PostgreSQL 16 cluster (app role NOBYPASSRLS, row security forced on every table). Verbatim case receipt: C:/CT-ZC-FULL3747/tools/zcode-integration3723/runtime/FULL3747-O14EV/evidence/O14EV-S1-ANALYTICS-CASES.json (suite S1-analytics, started 2026-10-25T00:02:55.081Z); cases cover matrix rows O01.01–O01.10 and O14.01. Standing: local-candidate, test-harness evidence; Chief acceptance and publication OPEN. The screens-side guide stands on its own Chief-accepted evidence (CT-ANALYTICS3022 / CT-ANALYTICS3044) and is not modified by this guide.