The events your agent logs to debug your system are the same events a product person wants charted. A checkout that ends in captured is a path to the evaluator and a conversion to a dashboard; a user_id on it is evidence for one and an active user for the other. So Logika keeps the numbers as the events pass through, and analytics come with every tier: no second SDK, no second bill, and nothing to re-scan.
What's counted on its own
Every event the ledger declares is counted every 5m, by environment and app version, with no setup:
- how often it happens, how often it fails, and how slow it is (p50 to p99);
- for every numeric param: totals, averages and the shape of its values (revenue, items, bytes);
- whether a new version made any of it worse, side by side with the old.
These are kept as small rollups beside the data: 5m rows for about 48h, hourly rows for about 90d, daily rows for as long as the project lives. So the charts outlive the 30 days of raw data, and none of it counts toward your storage.
What you declare
Some questions can't be guessed, because they're about meaning: which param is a person, which is worth grouping by, which events form a journey. Those are declared, in the ledger:
identityon a param (user_id,account_id): exact daily, weekly and monthly actives, and whether they come back (retention), with no sampling and no estimates.dimensionon a param with few values (provider,plan,reason): which values dominate.- a funnel: of those who did A, how many reached B, then C, within an hour.
- a milestone: how long it takes from one event to another, such as from sign-up to first post.
"funnels": [{"name": "purchase", "within": "1h",
"steps": ["landing.viewed.user_id", "cart.added.user_id", "checkout.captured.user_id"]}],
"milestones": [{"name": "first_post", "from": "signup.completed.user_id", "to": "post.created.user_id"}]
Declaring isn't a cliff: each declaration is filled in from the raw data already stored, within minutes of the sync that adds it, so a funnel you declare today shows the past month at once (3d on the free demo).
Panels and views
The console's Analytics page shows the views your ledger declares, each a list of panels. A panel names its question and, optionally, a title in plain words and the sentence it opens with, filled in from the data:
"views": [{"name": "overview", "panels": [
{"event": "checkout", "show": ["volume", "error_rate", "p95"]},
{"event": "checkout.declined", "by": "reason"},
{"funnel": "purchase", "title": "Who buys", "answer": "{conversion} of visitors bought, {range}"},
{"identity": "user_id", "show": "actives", "title": "People using the app", "answer": "{latest} people a day · {change}"}]}]
The page then reads like an answer, not a dashboard: "People using the app: 1,204 people a day", with how that changed and the chart beneath it. Without views, the console words each panel from the event's description.
What changed
Every 5m, Logika also asks of each event's numbers: how surprising is this hour, given the past? Failures and conversions are judged as proportions, volume against the same hour of the week and of the day, latency by the share of requests slower than last month's p95, versions and environments side by side. Four guards keep it quiet: a minimum volume, a minimum effect, a strict threshold, and a change seen in two of the last three passes.
What it finds splits in two:
- A change a release is to blame for (failures or latency worse on the new version, or since a deploy) is a regression: an issue, worked by your agent like any other (the loop).
- Anything else is an insight, for you to read: "checkout: failure rate rose from 1.2% to 3.4%", "traffic rose across 4 events", "purchase: conversion fell from 4.1% to 3.2%". Insights are listed under What changed, at the top of Analytics, week by week, and in Activity's Worth knowing, and ride along on your next new-issues email as Also changed.
Each insight has a Why? button: your own agent investigates with queries, changes no code, and answers in a sentence, shown where you asked. If an insight isn't worth knowing, saying so (not a problem) widens that series' threshold, so Logika gets quieter exactly where it was wrong.
Funnel conversion and the spread of numeric values are judged once a day, on the closed day, against the four weeks before it.
For your agent
Everything on the page is a command away, in the dense form agents read:
logika query series '{"event": "checkout", "env": "prod", "show": ["volume", "error_rate", "p95"], "bucket": "day", "from": "30d"}'
logika query funnel '{"name": "purchase", "env": "prod", "from": "30d"}'
breakdown, actives, retention and milestone work the same way, and logika query sql runs SQL over the same rollups, and over the raw events, when a canned question isn't enough. This is how your agent answers a Why?, checks whether its fix moved a number, or tells you what last week looked like.
Written in reflection
Which params identify people, what's worth breaking down, which funnels and milestones matter and what each panel should say are questions for you, not your agent. They're settled in a reflection round: the agent reads the captures for the paths people actually take and the params that actually vary, proposes each declaration with what it would show, and writes what you agree on into the ledger. When the data lacks something a question needs, the round says which event or param to add, rather than guessing.