Sources

TDI — Total Daily Insulin (Glooko web-stats)

Refresh re-pulls the Glooko web-stats feed via the glooko_rest poller backfill (202 queued — the sync-pollers cron is the effective trigger).

No Glooko web-stats snapshots captured yet.

Source Health

HP's data model is two-layer: raw events per source (the cards below — actual live HP sources, e.g. dexcom_share, android_agent, glooko, glooko_rest, nightscout, gluroo_ggc, op5_accessibility) → canonical production_events (the Production Layer panel below).

No source events yet.

Unconfirmed real-time feed (Tier 3)

UNRELIABLE — freshness only. NS/Pixel/Gluroo capture insulin events without a user-vs-algorithm flag (NS captures ~58 % of canonical TDI). Never used in analytics or recommendations.

No tier-3 freshness data.

Production Layer

Canonical production_events counts (active rows).

Production status unavailable.

Recompute

Non-destructive supersede + rebuild over a window. Rows are never deleted.

Microbolus Audit

OP5 SmartAdjust canonical-vs-raw coverage. ~288 rows/day expected.

Data Rules

Priority order per data kind: highest on the left. The production layer uses the highest-priority source that has data in each match window; lower-priority sources contribute as alternatives only. Drag chips within a row to reorder. Changes take effect at the next recompute (≤ 2 h via cron, or run a Recompute in the Production Layer above).

Data-rules config unavailable.
Transformation Rules 0 rules click to expand
No transformation-rules endpoint in HP — see the editable Data Rules section above (the HP analogue of per-kind source priority).

Source vs Production Comparison