Auto-tracking everywhere — every family-wide metric is captured automatically

decision Sat Jun 20 2026 00:00:00 GMT+0000 (Coordinated Universal Time) decisionsarchitecturetrackingobservabilitylifestreamautometrics

Auto-tracking everywhere

Decision

Every metric tracked anywhere in the chirag127/oriz* family is captured automatically. No manual entry anywhere in the metric pipeline.

This decision file locks the family-wide principle. The companion rules/interaction/auto-only-tracking.md spells out the rule that every future tracking pick must pass.

For the oriz-me lifestream specifically, this decision pre-locks the auto-event sources that feed canonical JSONL per lifestream-jsonl-canonical:

User direction (verbatim, 2026-06-20): "I want everything to be automatically tracked no manual tracking."

Why

• Manual tracking decays. See rules/interaction/auto-only-tracking.md §"Why" for the full argument.
• Dishonest data is worse than no data. A manually-fudged metric produces wrong decisions; an untracked gap is at least honest about being a gap.
• Auto scales. Adding a hook / SDK / webhook costs human attention once; manual tracking costs human attention forever.
• Everything else in the stack is already auto. Errors (Sentry), logs (CF Workers Tail
  • Better Stack Logs), uptime (Better Stack + Instatus), analytics (CF Web Analytics
  • GA4 + PostHog + Clarity), CI status (GitHub Actions). Time tracking was the only manual outlier — Toggl walk-back closes that gap. See time-tracking-wakatime-only.

Implications

For the lifestream pipeline (concrete auto sources)

Each event source has a designated auto-capture path. None require the user to do anything beyond their normal work.

• GitHub commits ? JSONL line per commit. Webhook configured on the chirag127/oriz* org ? Hookdeck for retry-protected ingress (per hookdeck-for-webhook-reliability) ? CF Worker at api.oriz.in/lifestream/ingest/github ? appends to oriz-me-data/lifestream/YYYY-MM-DD.jsonl. Idempotent on commit SHA.
• npm publishes ? JSONL line per publish. A postpublish script in each package's CI (or a GH Actions step in npm-publish.yml) POSTs to api.oriz.in/lifestream/ingest/npm after a successful npm publish. Idempotent on (package, version).
• VS Code coding sessions ? daily summary line. A GitHub Actions cron runs daily at 02:00 UTC, calls wakatime.com/api/v1/users/current/summaries?start=yesterday&end=yesterday, maps each language / project bucket to a JSONL row, appends to yesterday's file. Idempotent on (date, project, language) via dedup key. The Wakatime IDE plugin is the only piece the user installs once; from there it's automatic.
• Site visits ? already auto via CF Web Analytics. No additional ingest path needed for the per-visit signal — CF Web Analytics' dashboard already aggregates server-side. A daily cron pulls the previous day's aggregates from CF's GraphQL Analytics API (free tier) and appends a single daily-rollup JSONL row per site (count, top paths, top countries) — this gives the lifestream visibility into "the site got visits" without burning JSONL volume on per-pageview rows. Idempotent on (date, site).
• Builds ? JSONL line per workflow run. GitHub Actions workflow_run webhook ? Hookdeck ? CF Worker at api.oriz.in/lifestream/ingest/build ? appends. Captures (repo, workflow, run_id, conclusion, duration_s). Idempotent on run_id.

All five sources push to the same canonical JSONL store per lifestream-jsonl-canonical. Turso libSQL warm cache rebuilds from JSONL nightly.

For non-lifestream metrics

The rule still applies. If a future feature wants to track something the family doesn't already auto-capture, the acceptance gate is: show the auto-capture path before adopting. If there isn't one, the metric isn't tracked. No manual fallback.

What this excludes

• Journal / thoughts entries on oriz-journal-site are CONTENT, not metrics. Manual writing is intentional and expected. See rules/interaction/auto-only-tracking.md §"When NOT to apply this rule".
• Markdown blog posts, design briefs, knowledge bundle entries — also content. Not subject to this decision.
• Code itself is content.

Walk-backs locked alongside this decision

• Toggl Track REJECTED — manual timer violates this principle. Walked back same day it was adopted. See time-tracking-wakatime-only.
• No future "manual fallback for non-coding time" tool. If a non-coding-time signal becomes valuable, the answer is to find an auto source for it (e.g. Google Calendar API ingest), not to add a manual tracker.

Failure modes covered

• Hookdeck quota (50K events/mo free, per distribution-and-queues-locked) — at family scale, commits + builds + npm publishes total well under 1K events/mo, ~50× headroom.
• CF Worker 100K req/day — every ingest endpoint sits behind the CF Worker quota mitigation playbook; ingest is write-shaped (no caching) but well under the cap.
• Wakatime API rate limit — daily cron pulls a single summary endpoint per day; no concern.
• CF Web Analytics GraphQL — also free; daily pull only.
• Webhook delivery failures — Hookdeck retries with exponential backoff + manual replay surface; same posture as Razorpay webhook flow.

Cross-refs

• Auto-only-tracking rule — the rule version of this principle
• Time-tracking — Wakatime only — Toggl walk-back driven by this decision
• Lifestream JSONL canonical — the durable store the auto sources feed
• oriz-me added to family — the lifestream-bearing site
• Hookdeck for webhook reliability — webhook ingress layer
• CF Worker quota mitigation playbook
• Wakatime — sole time-tracking pick
• Cloudflare Web Analytics
• Sentry — auto error capture
• GitHub Actions cron schedule
• Future overrides past rule