Build intelligence

Build reports

Automatic build metrics, the cache rule, and offline regression analysis.

What a build report is

Every successful native Go/Rust build (build, rebuild, zero-downtime deploys) automatically prints a Build Report and records it with the version metadata - no flags, no configuration, no network. It works fully offline.

Output

→ Build Report  Application   api  Version       v12  Compiler      Go 1.27  Duration      31.2s  Cache         COLD (go-build-cache)  Binary        14.8 MB (linux/amd64)  Commit        8f31c2a  ────────────────────────────────────────────────   Regression Analysis  Compared against 1 previous comparable build.   Binary size    Previous      14.1 MB    Current       14.8 MB    Change        +0.7 MB (+4.9%)   Build duration    Comparison skipped:    cache mode differs from the previous comparable build (COLD → HIT)

What is captured

Application and version
Language and compiler/toolchain version (go version / rustc --version)
Build start/end time and duration
Binary size and target platform (artifact type, size, platform)
Cache status: COLD = real compilation happened, HIT = served from the compiler's build cache
Git commit when available (unavailable outside a git repository)
The build arguments you passed via --build-arg

The cache rule

Binary size remains a valid regression signal across cache states when the artifact, toolchain, platform, and matrix context are comparable - a cache hit does not change what the compiler produces.

Regression analysis

Regression analysis compares the current build against the previous 5 comparable builds. Comparable means the same application, language, toolchain, target platform, and artifact type - with cache state ignored for size comparisons and matched for duration comparisons.

Alert thresholds

Signal	Alert fires when
Binary size	growth ≥ 1 MB or ≥ 5% versus the previous comparable build, plus an N-build average baseline when enough history exists
Build duration	slowdown ≥ 25% and ≥ 2s versus a previous build in the same cache mode (thresholds avoid noisy micro-changes)

Interpreting results

The report is the fastest way to detect unexpected binary growth, compilation regressions, or toolchain changes - the compiler version in the report makes accidental toolchain bumps visible. A comparison that is skipped is information, not a failure: the report would rather say Comparison skipped than show a percentage that compares a cold compile to a cache hit.

Regression analysis is pure observability: missing history never fails a build, and a reporting problem only prints a warning.

phelix build-report

Read-only inspection of stored build reports - never triggers a build, works fully offline:

Terminal

phelix build-report myapp            # 5 most recent reportsphelix build-report myapp --limit 10