# CacheFold Assessment Criteria

Use this page to decide whether a paid CacheFold assessment is likely to produce
measurable value for one Redis or Valkey namespace.

## What CacheFold Is Designed For

CacheFold is designed for deterministic Redis/Valkey namespaces where repeated
values create material memory pressure.

Good candidates can include:

- catalogs and product reference data
- configuration and feature-flag matrices
- pricing, shipping, tax, and rules tables
- localization/reference datasets
- published data cached from a clear source of truth

## What Usually Does Not Fit

These patterns usually should not enter a paid assessment:

- sessions, tokens, or per-user state
- high-entropy blobs with little structural repetition
- high-churn domains without controlled TTL behavior
- workloads where baseline parity cannot be measured before pilot
- namespaces without validation and rollback owners

## Economic Screen

The default qualification target is a plausible in-scope baseline of at least
90 GiB and a path to at least $50k/year in annual infrastructure savings.

The screen estimates annual savings from:

- in-scope baseline GiB
- expected reduction range
- Redis/Valkey price per GiB-month
- replica/headroom factor

This estimate is directional. The paid assessment determines the measured
reduction range and whether the savings case supports a pilot.

## Assessment And Pilot Standards

Core assessment standards:

- correctness parity against baseline behavior
- mixed read/write safety checks, including restart consistency
- measured latency impact against an agreed p99 budget
- net Redis memory reduction on the in-scope namespace
- explicit rollback path before pilot

Performance acceptance is agreed before pilot:

- Tier 1: up to about 1.0 ms p99 regression
- Tier 2: up to about 5.0 ms p99 regression when explicitly accepted for a large validated savings case

## Engagement Flow

1. Free async fit screen: platform, prefixes/GiB, value shape, read/write ratio,
   TTL/churn, source-of-truth check, p99 budget, validation owner, rollback owner.
2. Paid assessment: $7.5k-$10k for one namespace, ending in PASS/FAIL.
3. Pilot after PASS: $15k-$25k with shadow, canary, gated rollout, and rollback drill.
4. Annual license: max($30k, 10-15% of validated first-year net savings).

## What To Send For A Fast Fit Verdict

1. Redis/Valkey platform and deployment shape
2. Rough in-scope memory footprint in GiB, including replicas or headroom
3. Top prefixes by memory
4. Value shape and source-of-truth path
5. Read/write mix, TTL/churn, and p99 budget
6. Validation owner and rollback owner