tracking-and-funnels
Skill13 starsChanged 5 months agoMITView on GitHub
Designs what to track and how to read funnels — event schema derived from open hypotheses, funnel/cohort analysis for validation and briefs. Use for "воронка", "funnel analysis", "drop-off", "where are users dropping", "что трекать", "event schema", "tracking plan", "cohort analysis", "трекинг событий".
In this repository23 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ambiguity-resolverSkill
- challengeSkill
- deep-thinkSkill
- design-critique-templateSkill
- experiment-designSkill
- ingestSkill
- interview-analysisSkill
- multi-source-signal-synthesiserSkill
- next-advisorSkill
- pipeline-stepsSkill
- post-launch-reviewSkill
- problem-structuringSkill
- product-requirements-docSkill
- setup-initiativeSkill
- solution-scoringSkill
- strategic-narrative-generatorSkill
- system-design-docSkill
- technical-spec-documentSkill
- ui-pattern-librarySkill
- user-persona-builderSkill
- user-story-generatorSkill
- user-testingSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Every event: user id, timestamp, platform, session id.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Tracking & Funnels
One skill for both directions of the same loop: what to instrument (so
hypotheses become measurable) and how to read the funnel (so
measurements become verdicts). Serves /brief (steps 4–5), /validate
(step 6) and experiment-design (step 14/16).
Design tracking FROM the hypotheses
For every testing/draft hypothesis in output/hypotheses.json, name the
event/metric that would confirm or refute it — that mapping IS the
acceptance test of the schema. An event no hypothesis needs is noise; a
hypothesis no event can test is a research gap → into the analytics brief.
Schema conventions:
object_actionnaming (checkout_started, notclick_btn_3), snake_case- Properties over event proliferation: one
order_completedwithpayment_methodbeats three payment events - Every event: user id, timestamp, platform, session id; funnel events additionally carry the entry surface/source
- Define each metric once, in writing, with its window ("CTR" without a definition is how gate presentations die — see /challenge)
Read the funnel
- Sequential windows: a funnel is ordered events within a window per
user — state the window and whether re-entry counts; unstated windows
make numbers incomparable across sources (→
data_inconsistency). - Segment before averaging: an aggregate drop-off hides opposite behaviors; split by the CONTEXT.md segments and by entry surface first.
- Cohorts for time effects: compare users by start week when the product changes underneath them.
- Skeleton (adapt to the PM's warehouse; pseudo-SQL is fine in briefs):
SELECT step, COUNT(DISTINCT user_id) AS users,
ROUND(100.0 * COUNT(DISTINCT user_id) /
FIRST_VALUE(COUNT(DISTINCT user_id)) OVER (ORDER BY step), 1) AS pct
FROM funnel_events -- events pre-mapped to ordered steps
GROUP BY step ORDER BY step;
-- splitting by segment/platform? add PARTITION BY to the window fn,
-- or pct will be computed against the wrong baseQuality checks before trusting numbers
Totals reconcile across sources (same metric, same definition, same window)? Bot/internal traffic excluded? Event fired once per action (dedup)? A funnel built on unverified events produces confident nonsense — flag first, analyze second.
Output to the registry
A drop-off finding is evidence: hypotheses.py set <id> --type REAL
--confidence <C> --add-source "research/analytics-data.md::<funnel, step>".
A drop-off matching no hypothesis → candidate hypotheses.py add (ask the
PM). Contradicting sources → --flag data_inconsistency, never silent
averaging.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

