# First-Week Memory Maintenance Schedule

## Purpose

This schedule is the first live operating cadence after the memory pipes are connected.
It is designed to move the system from installation into observation, assessment, and tuning.

## Weekly objective

By the end of week one, we should know:

- whether search and fetch work reliably
- where retrieval is noisy or weak
- which memories are being used
- which items need pruning, retagging, merging, splitting, or reindexing
- whether the current cadence is too aggressive or too light

## Day 0 — install day

### Goal
- Load the approved Open Brain batch with provenance.
- Verify the install completed cleanly.

### Checks
- counts match the approved batch
- source-only items remain out of durable memory
- Hermes native memory remains empty for this batch

### Output
- install confirmation
- initial smoke-test notes

## Day 1 — baseline retrieval

### Goal
- Test basic search and fetch.

### Checks
- search by project/domain
- search by voice rule
- search by offer/campaign
- fetch source evidence by ID

### Output
- baseline retrieval notes
- first assessment of precision/recall

## Day 2 — boundary test

### Goal
- Confirm the line between source-only archive and approved memory.

### Checks
- can the system distinguish source-only from installed memory?
- are source excerpts easy to recover without promoting them?
- are approved memories too broad or too narrow?

### Output
- boundary audit
- candidate retag/prune suggestions if needed

## Day 3 — dedup / noise check

### Goal
- Look for redundancy, overlap, stale material, and over-promoted items.

### Checks
- duplicate or near-duplicate memories
- stale or low-value items
- repeated query misses
- noisy retrieval clusters

### Output
- dedup/pruning recommendations
- any split/merge suggestions

## Day 4 — surface comparison

### Goal
- Compare behavior across surfaces.

### Checks
- WebUI retrieval
- Telegram retrieval
- any future channels if present

### Output
- cross-surface consistency notes
- any per-surface tuning suggestions

## Day 5 — assessment and tuning day

### Goal
- Convert the first 4 days into a decision set.

### Checks
- what is working well?
- what is producing noise?
- what should change now?
- what should wait for more data?

### Output
- assessment summary
- suggestion list
- candidate changes to schema or cadence

## Day 6 — repeatability check

### Goal
- Re-run the best queries and confirm the system behaves consistently.

### Checks
- same query, same general result?
- has tuning improved or worsened retrieval?
- are metrics trending in the right direction?

### Output
- repeatability notes
- trend table update

## Day 7 — weekly review

### Goal
- Produce the first weekly maintenance report.

### Checks
- metrics summary
- assessment
- suggestions
- decisions
- next cadence adjustment

### Output
- completed weekly report
- next week’s maintenance plan

## Suggested daily rhythm

- 1 short retrieval smoke test
- 1 source-only vs approved-memory check
- 1 tuning observation
- 1 report row in the maintenance table

## Cadence adjustment rule

If the system is noisy or unstable, inspect more often.
If the system is stable and predictable, reduce the cadence.
The cadence should be based on evidence, not habit.