Chronote Product Analytics

Use PostHog for product behavior and Langfuse for model telemetry. Keep website visitors, authenticated product users, guilds, and AI generations distinct.

Product metrics

• Confirm the active PostHog project is Chronote before querying.
• Read the live event schema before relying on event or property names.
• Unless the user chooses another definition, count a product-active user as a unique person who emitted meeting_started, meeting_completed, or ask_question_asked. Report anonymous website visitors separately.
• Use bounded UTC ranges. State whether the current day is partial.
• Report meeting starts and completions separately. Equal totals are evidence of aggregate completion, not a proven one-to-one pairing.
• Use the langfuse-metrics skill for aggregate model volume, cost, and latency. Do not treat healthy telemetry as output-quality evidence.
• For the daily operating view, run its cost-report command over the same UTC window as PostHog. Supply completed meetings and transcribed minutes as denominators, and report fully-unpriced model groups alongside attributed cost. This is operational cost accounting, not an OpenAI invoice reconciliation.
• Do not retrieve questions, answers, notes, transcripts, or identities when aggregate evidence answers the request.

Acquisition attribution

Prefer an explicit server_install_attributed event when it exists. Report only the sanitized source, medium, campaign, landing path, referrer domain, and CTA location; never expose a browser/session id, Discord user id, full URL or query string, raw user agent, or precise geography.

Older server_installed events are keyed to a guild while website activity is keyed to a browser or signed-in user, so they cannot be joined deterministically. For those installs, a bounded timestamp correlation can produce an inference:

1. Read the guild's server_installed timestamp.
2. In a short lookback before that timestamp, group $pageview and add_to_discord_clicked by landing path, referrer domain, UTM source, and CTA location.
3. Call the source a strong inference only when there is one coherent path, no competing click path, and the final install click is within a few minutes of the install. Otherwise report acquisition as UNKNOWN.
4. Use guild-scoped onboarding events only to validate subsequent activation; do not retrieve or report the acting person's identity.

Always label timestamp correlation as inferred rather than deterministic. A tight sequence such as one referral landing, one CTA path, an installation seconds later, completed onboarding, and immediate meeting usage is useful product evidence, but it is not a cryptographic join.

Resolve active guild names

PostHog intentionally stores guild_id, not the mutable guild name. Group the requested product events by guild_id first, including only the counts and time bounds needed for the question.

Resolve names in this order:

1. Match IDs visible in the signed-in Chronote server picker when the operator already has access.

2. For remaining IDs, run scripts/resolve-production-guilds.mjs with authorized production AWS access:

   node .codex/skills/chronote-product-analytics/scripts/resolve-production-guilds.mjs --guild-id <id> --guild-id <id>
   

   The helper reads meeting-notes-prod/discord-bot-token from Secrets Manager in us-east-1, verifies that Discord identifies it as the Chronote bot, and prints only the requested IDs and names. Override --secret-id, --region, or --expected-bot-id only when the target environment is explicitly different.

3. If AWS authentication is unavailable, stop and request the appropriate read-only login. Do not substitute an unrelated local bot token, dispatch a workflow, expose a secret, or persist guild names into analytics.

Treat an ID absent from the bot guild list as unresolved; it may represent a removed installation or the wrong environment. Report that distinction rather than guessing a server name.

This skill is read-only. It does not authorize production configuration, analytics instrumentation, billing, deployment, or Discord mutations.