v0.3.0 — Policy IR, Search Engine, Deployment Scaffolds

quant-cache v0.3.0

Economic cache control plane — Phase B-D release.

New: Policy Intermediate Representation (Phase B)

Declarative cache policy configuration as JSON/TOML:

{
  "backend": "sieve",
  "capacity_bytes": 50000000,
  "admission_rule": { "type": "score_threshold", "threshold": 0.5 },
  "bypass_rule": { "type": "size_limit", "max_bytes": 10485760 },
  "prewarm_set": ["/images/hero.jpg"],
  "ttl_class_rules": [
    { "content_type_pattern": "image/", "ttl_seconds": 7200 }
  ]
}

qc policy-eval --input trace.csv --policy config.json --preset ecommerce

New: Policy Search Engine (Phase C)

Automatic search over backend × admission × bypass × prewarm × TTL class rules:

qc policy-search --input trace.csv --capacity 50M --preset ecommerce

New: Deployment Scaffold Generators (Phase D)

• qc compile --target cloudflare — Cache Rules + Workers script
• qc compile --target cloudfront — CacheBehaviors + CloudFront Functions
• qc compile --scores policy.json — populate admission scores from optimizer

New Tests

11 new tests for PolicyIR (serde round-trip, equivalence, admission, bypass, prewarm).

CLI Commands (10 total)

import, generate, optimize, simulate, compare, calibrate,
policy-eval, policy-search, compile, help

Full Changelog

v0.2.0...v0.3.0

v0.2.0 — QUBO Solver, Belady, Calibration, CloudFront Import

quant-cache v0.2.0

Major feature release completing V1.1 through V2.0 milestones.

New Features

V1.1: CDN Log Import

• qc import --provider cloudfront — convert CloudFront access logs to canonical trace CSV
• OriginCostConfig with fallback chain (rule → content-type → latency → global default)
• ProviderLogParser trait for future provider extensions

V1.5: Belady Oracle & Calibration

• BeladyPolicy: optimal eviction oracle using pre-indexed future access map
• qc compare --include-belady — hit-rate ceiling comparison
• qc calibrate: automatic coefficient tuning (coordinate descent + random restarts)
  • Optimizes replay estimated_cost_savings on validation set
  • Outputs calibrated TOML config + sensitivity report

V1.6: Reuse Distance

• reuse_distance::compute_reuse_distances() — stack distance distribution per object
• ObjectFeatures extended with mean_reuse_distance, p50, p95

V2.0: QUBO with Simulated Annealing

• qc optimize --solver sa — simulated annealing for quadratic optimization
• QuadraticProblem with sparse PairwiseInteraction list
• co_access::extract_co_access() — time-window co-occurrence counting
• --co-access-window-ms and --co-access-top-k flags
• SA with co-access bonus outperforms greedy on heterogeneous traces

CLI Commands (7 total)

qc import      Import CDN provider logs
qc generate    Generate synthetic trace data
qc optimize    Optimize cache policy (greedy/ilp/sa)
qc simulate    Replay trace against policy
qc compare     Compare against baselines (LRU/GDSF/Belady/ILP)
qc calibrate   Auto-tune economic parameters
qc help        Help

Full Changelog

v0.1.0...v0.2.0

v0.1.0 — Initial Release

quant-cache v0.1.0

Economic CDN cache optimization engine.

What it does

Formulates CDN cache admission as a 0-1 knapsack problem maximizing expected economic benefit (latency savings + origin cost reduction) under capacity and freshness constraints.

Highlights

• Scoring: per-object expected benefit with Poisson-based stale probability model
• Solvers: GreedySolver (O(n log n)) + ExactIlpSolver (HiGHS, n < 10k)
• Evaluation: trace replay with LRU/GDSF baselines, stale detection (TTL + version mismatch)
• CLI: qc generate, qc optimize, qc simulate, qc compare
• Presets: ecommerce, media, api — with configurable stale penalty overrides

Key Results

Metric	Result
Optimality gap (median, n=1000)	0.01%
Optimality gap (p95, n=1000)	0.72%
Greedy solve 10k objects	0.6ms
Trace replay 1M events	825ms
75 tests + 5 acceptance guards	All passing

Quick Start

cargo build --workspace --release

qc generate --num-objects 10000 --num-requests 1000000 --output trace.csv
qc optimize --input trace.csv --output policy.json --capacity 50000000 --preset ecommerce
qc simulate --input trace.csv --policy policy.json
qc compare --input trace.csv --capacity 50000000 --preset ecommerce

What's Next

• V1.1: CloudFront/Cloudflare log import + real trace validation
• V1.5: Belady baseline, offline coefficient calibration
• V2.0: QUBO with quadratic terms (co-access, purge-group, origin-group)

See docs/roadmap.md for the full plan.