Can one namespace clear the savings threshold?
CacheFold starts with in-scope GiB, replicas, pricing, and a reduction range to test whether a $50k/year savings case is plausible.
Redis/Valkey memory assessment
Test whether one repeated-data namespace can support $50k/year in savings.
CacheFold assesses deterministic catalog, config, pricing, localization, or similar reference data for lower memory pressure, delayed capacity expansion, a quantified savings range, and a pilot-ready proof path.
First action: free async fit screen. Paid one-namespace assessment: $7.5k-$10k.
- Designed for deterministic repeated values, not sessions or tokens.
- Assessment checks parity, p99 budget, source of truth, and rollback ownership.
- No pilot recommendation unless the evidence supports both savings and control.
What the first screen decides
Do the economics, data shape, and owner path justify an assessment?
Are the values deterministic and repeated?
Strong candidates look like catalog, config, pricing, localization, or reference data. High-churn user state routes to no-bid or diagnostic only.
Can a pilot be controlled?
A PASS requires a source-of-truth check, p99 budget, validation owner, and rollback owner before any pilot is recommended.
Free async fit screen
Send the minimum data needed for a credible yes/no.
Use ranges if exact data takes time. Unknowns are acceptable, but they route the next step toward a tiny diagnostic before a paid assessment.
Economic case
Translate repeated values into capacity pressure and dollars.
CacheFold does not sell a generic Redis tuning pass. It tests whether repeated, deterministic values can be represented with smaller parameters and exceptions, then reconstructs reads while preserving baseline behavior.
Lower memory pressure
Estimate saved GiB after accounting for DDA structures and exceptions.
Delay capacity expansion
Map saved GiB to nodes, shards, replicas, or serverless GB-hours.
Justify a pilot
Move forward only when savings, p99 budget, and rollback ownership line up.
Buying path
Separate the free screen, paid assessment, pilot, and license decision.
Async fit screen
Free screen using platform, top prefixes/GiB, value shape, read/write ratio, TTL/churn, source of truth, p99 budget, and owners.
$7.5k-$10k assessment
One namespace. Produces a quantified savings range, validation plan, rollback path, and PASS/FAIL recommendation.
$15k-$25k pilot after PASS
Shadow, canary, gated rollout, and rollback drill against the agreed p99 and correctness criteria.
Annual license
max($30k, 10-15% of validated first-year net savings), after the pilot establishes the buyer-specific savings case.
Proof posture
Confident about the artifact path, careful about buyer-specific savings.
CacheFold uses a successful proof path for deterministic repeated data: exact reconstruction, mixed read/write checks, and core Redis method parity before a namespace can move toward pilot.
The public promise stays assessment-gated: your savings range depends on in-scope GiB, replica/headroom factor, pricing, reduction, p99 budget, and rollout authority.
Evidence
Buyer-safe claim
0 mismatches in local OpenAlex-derived proof artifacts
Assessment verifies parity before pilot.
93.68% net Redis reduction on the bounded local proof
Use measured net Redis reduction, not raw key deletion, for savings math.
Mixed read/write and restart consistency checks
Pilot requires runtime safety checks and rollback ownership.
Core Redis drop-in parity checks
Assessment scopes the runtime methods used by the namespace.
Common objections
Answer the hard questions before the assessment.
"We can tune Redis ourselves."
Do that first. CacheFold is for structural repetition left after normal sizing, TTL, eviction, and serialization work.
"Latency is non-negotiable."
Then the p99 budget should be Tier 1 or no-bid. Cost-first latency tradeoffs require explicit buyer acceptance.
"We cannot share production data."
Start with anonymized prefix and memory stats. A proof can be scoped around approved snapshots and source-of-truth checks.