agents/time-of-day: auto-infer quiet hours, peak hours, timezone drift #112

New Issue

alvis · 2026-05-05T09:26:01Z

alvis commented

2026-05-05 09:26:01 +00:00

Goal

Auto-infer per-user quiet hours, peak hours, and timezone drift from observed activity, and write them into agent:time-of-day preferences via the inference framework (#111).

Signals

tip_views.served_at — when the user is reachable.
tip_feedback.created_at + action — when they actually engage and what they do.
integration_tokens.connectedAt and recent integration activity — secondary cadence hints.

Inferred parameters

Key	Shape	Default
`quietStart`, `quietEnd`	HH:MM in user TZ	22:00 / 07:00
`peakHours`	int[] (0–23) sorted by completion-rate density	[9, 14, 20]
`tz`	IANA string inferred from activity histogram	from auth provider, fallback UTC

Method (sketch)

Histogram engagement events by local hour, last 30d.
Quiet window = longest contiguous run of below-baseline engagement.
Peak hours = top quartile of completion rate.
TZ drift = mode shift between weekly histograms.

Done when

Inference function plugged into the framework with TTL = 24h.
Cold-start uses defaults; flips to inferred values once ≥ 50 engagement events.
Unit tests with synthetic histograms covering night-owl, early-bird, and shift-worker cases.
Snippet output for TimeOfDayAgent reads inferred values.

## Goal Auto-infer per-user quiet hours, peak hours, and timezone drift from observed activity, and write them into `agent:time-of-day` preferences via the inference framework (#111). ## Signals - `tip_views.served_at` — when the user is reachable. - `tip_feedback.created_at` + `action` — when they actually engage and what they do. - `integration_tokens.connectedAt` and recent integration activity — secondary cadence hints. ## Inferred parameters | Key | Shape | Default | |---|---|---| | `quietStart`, `quietEnd` | HH:MM in user TZ | 22:00 / 07:00 | | `peakHours` | int[] (0–23) sorted by completion-rate density | [9, 14, 20] | | `tz` | IANA string inferred from activity histogram | from auth provider, fallback UTC | ## Method (sketch) - Histogram engagement events by local hour, last 30d. - Quiet window = longest contiguous run of below-baseline engagement. - Peak hours = top quartile of completion rate. - TZ drift = mode shift between weekly histograms. ## Done when - Inference function plugged into the framework with TTL = 24h. - Cold-start uses defaults; flips to inferred values once ≥ 50 engagement events. - Unit tests with synthetic histograms covering night-owl, early-bird, and shift-worker cases. - Snippet output for `TimeOfDayAgent` reads inferred values.

alvis added this to the M2 — AI tips + multi-source signals milestone 2026-05-05 09:26:01 +00:00

alvis referenced this issue from a commit

2026-05-05 11:14:30 +00:00

feat(profile): /api/profile + eligibility filter + inference framework (ADR-0014 steps 4-6)

alvis closed this issue

2026-05-06 06:15:34 +00:00

alvis referenced this issue from a commit

2026-05-06 06:55:01 +00:00

feat(agents): quiet window + peak hours + tz prefs for time-of-day agent (#112)

Sign in to join this conversation.