# A weekly review

The daily pass answers 'what needs me now'. The weekly review answers 'is this still set up correctly' — overdue commitments, what went out unsupervised, whether anything has quietly stopped working, whether allowances are being hit, and whether what Connect knows is still true. Half an hour, and each finding should produce a change rather than a note.

- **Status:** Available
- **Audience:** both
- **In the app:** #/follow-ups, #/activity, #/mailboxes, #/billing, #/knowledge
- **Last verified:** 2026-09-10
- **Canonical:** https://connectbyjbrh.com/docs/getting-started/weekly-review/

## The five

| Check | Where | What it tells you |
|---|---|---|
| Overdue follow-ups | `#/follow-ups` | Whether commitments are being kept, and whether `task` items are being done by anybody at all |
| What went out unsupervised | `#/activity` and the decision log | Whether autonomous channels are writing things you would have written |
| Mailbox and line health | `#/mailboxes`, `#/calls` | Quiet failures — a mailbox that stopped fetching looks identical to a quiet week |
| Usage against your plan | `#/billing` | Whether a spent allowance is silently holding sends near the end of a period |
| Knowledge and memory drift | `#/knowledge`, the memory viewer | Whether Connect is grounding answers in things that are no longer true |

## The two that repay the most attention

**Overdue follow-ups** are the honest measure of whether the arrangement is working. A short list means commitments are being met; a growing one means either the autonomy is too tight for the volume, or the `task` items nobody can automate are piling up on a person who has not noticed. Both are fixable, and both are invisible on any single day.

**Knowledge drift** is slower and more damaging. Prices change, a product is withdrawn, a policy is rewritten, and the entry that grounds every answer still says what it said in March. Because grounding is budgeted and selected, a stale entry does not merely sit there unused — it competes with the correct one. Removing an outdated source is as valuable as adding a new one.

> **Note** Memory drifts too, and at the contact tier it drifts fastest: a preference recorded from one conversation may not have been a preference at all. Everything Connect remembers can be read at every tier and forgotten from the same place, so a review pass over a handful of your most important contacts is a reasonable weekly habit.

## Turning a finding into a change

**Autonomous replies you would not have sent** — Narrow the scope — pull that endpoint or contact back to `ask_before_send` rather than reversing the whole channel.
**A mailbox that stopped fetching** — Fix the connection, then check what arrived while it was down; the gap is real conversations, not a blank period.
**Allowances hit before the period ends** — Either the plan is the wrong size or something is being attempted that should not be. Read the refusals in the decision log before assuming it is the first.
**Follow-ups nobody can do** — They belong to a person by name, not to the queue. If nobody owns them, they are not commitments — they are hopes.
**A duplicate person appearing twice** — Merge the records. One person with two records means the relationship view is wrong in both directions.

## What does not belong in a weekly review

- Reading every message. Sampling is the point; if a channel needs full reading it should not be autonomous.
- Re-deciding approvals you already made. The decision log exists to answer a question, not to invite regret.
- Rewriting rules from single incidents. A rule written from one odd Tuesday tends to contradict the one written last month.
- Tuning the voice by ear from memory. Batch voice changes, then test them deliberately — [Review a call properly](/docs/how-to/review-a-call/) is the method.

## Questions

### Is weekly the right interval?

It suits most workspaces because it matches how quickly business facts change. A workspace handling high volume on autonomous channels benefits from a mid-week sample as well; a quiet one can review fortnightly without much risk, provided health and allowances are still glanced at.

### Who should do it?

Somebody who knows what the business currently promises — the review is mostly a judgement about whether Connect's picture of the business is still accurate. That is a different skill from working the queue, and in a larger team they are often different people.

### What is the single most valuable check if I only do one?

Overdue follow-ups. It is the one that shows up as customer damage soonest, and it is a good proxy for the rest: workspaces with a healthy follow-up list usually turn out to have healthy mailboxes and sensible autonomy too.

## Related

- [A working day with Connect](https://connectbyjbrh.com/docs/getting-started/daily-routine/)
- [Measure whether Connect is working](https://connectbyjbrh.com/docs/how-to/measure-whether-it-is-working/)
- [Clean up duplicate records](https://connectbyjbrh.com/docs/how-to/clean-up-duplicates/)
- [Reduce what Connect costs you](https://connectbyjbrh.com/docs/how-to/reduce-your-costs/)
- [Review a call properly](https://connectbyjbrh.com/docs/how-to/review-a-call/)
- [Create your first follow-up](https://connectbyjbrh.com/docs/getting-started/your-first-follow-up/)

## What this page is based on

- docs-source/sources/GENERAL.md §5, §6 — the audit trail and memory tiers
- docs-source/sources/GENERAL.md §7 — follow-ups
- Connect capability registry (docs-source/facts.py) — billing_usage, mailbox_health