Performance

Current comparison

Assura CLI vs native LS-Lint CLI

Generated stable trees. Equivalent rules. Case definitions live on the test-case page.

Faster tool: Assura
Runtime ratio: 1.62x
2x gate: Not complete
2x cases: 2 / 8
Assura faster: 8 / 8

Dataset: generated-fixtures-v1 + generated-fixtures-v2
Build profile: release; linux / x86_64
Hardware: AMD EPYC-Milan Processor; 16 logical CPUs
Closest case: 1.27x on Web app
Floor-limited misses: 0 / 6
Warm 2x gate: Complete; 8 / 8 cases
Evidence source: Commit unknown; measured worktree clean
Command: assura performance-report --output benches/history/current.json --history benches/history/ls-lint-comparison-history.jsonl --website-dir website/public/data/performance --iterations 5 --suite ls-lint
Captured: 2026-07-05 15:13:56 UTC
Versions: Assura 0.3.0; ls-lint v2.3.0 native binary

Headline Case Results

One row per measured headline case. Open a case to inspect the repository shape and the policy surface behind the comparison.

Test case	Project shape	Runtime	Result
Small library `simple_library`	Rust/package library with docs, tests, and build output. 5 checked, 2 ignored, 4 dirs; 6 LS-Lint rules Low-end case where startup and config loading are visible.	Assura CLI: 1.75 ms Native LS-Lint: 2.30 ms	Assura: 1.32x 24.0% lower runtime 2x target missed (1.15 ms)
Web app `web_app`	Frontend app with components, tests, CSS modules, assets, and dist output. 5 checked, 2 ignored, 4 dirs; 7 LS-Lint rules Common app layout with generated output pruning.	Assura CLI: 2.01 ms Native LS-Lint: 2.55 ms	Assura: 1.27x 21.1% lower runtime 2x target missed (1.27 ms)
Package monorepo `monorepo_packages`	Two packages with package-local source, tests, docs, and dist output. 7 checked, 3 ignored, 10 dirs; 10 LS-Lint rules Nested package scopes without becoming a synthetic stress test.	Assura CLI: 1.87 ms Native LS-Lint: 2.52 ms	Assura: 1.35x 25.7% lower runtime 2x target missed (1.26 ms)
Monorepo policy `monorepo_policy`	Strict TypeScript/Next-style workspace with app, package, docs, scripts, infra, and CI scopes. 39 checked, 10 ignored, 38 dirs; 38 LS-Lint rules Policy-shaped rules: root whitelisting, config exceptions, source bans, and generated-output pruning.	Assura CLI: 2.73 ms Native LS-Lint: 4.05 ms	Assura: 1.48x 32.6% lower runtime 2x target missed (2.02 ms)
Rule-heavy repo `rule_heavy_repo`	Repo-shaped tree with many extension-specific file rules. 192 checked, 2 ignored, 9 dirs; 38 LS-Lint rules Broad LS-Lint-compatible naming surface.	Assura CLI: 2.38 ms Native LS-Lint: 3.70 ms	Assura: 1.55x 35.6% lower runtime 2x target missed (1.85 ms)
Generated-heavy repo `ignored_generated_heavy_repo`	Project with large generated and coverage folders inside the tree. 2 checked, 769 ignored, 52 dirs; 2 LS-Lint rules Ignored-path pruning before validation work is spent.	Assura CLI: 1.50 ms Native LS-Lint: 5.65 ms	Assura: 3.77x 73.5% lower runtime 2x target met (2.83 ms)

What this does not prove yet: this checked report uses generated fixtures. Run the external-fixture command on the test-case page to produce the real-repo headline evidence.

Diagnostic Coverage

The report also keeps synthetic stress cases for regression detection. They are intentionally excluded from the headline comparison.

simple_small 51 checked, 1 ignored, 6 dirs 2 rules; directory naming plus TypeScript file naming

simple_medium 1001 checked, 1 ignored, 21 dirs 2 rules; directory naming plus TypeScript file naming

monorepo_large 5001 checked, 1 ignored, 51 dirs 2 rules; directory naming plus TypeScript file naming

rule_heavy 2001 checked, 1 ignored, 41 dirs 31 rules; directory naming plus 30 extension-specific TypeScript naming rules

ignored_generated_heavy 2 checked, 1001 ignored, 53 dirs 2 rules; directory naming, TypeScript file naming, and generated-path pruning

Assura-Native Performance

Native rows measure Assura-only capabilities that LS-Lint does not provide: content collections, graph queries, references, context packs, Markdown safe-fix dry runs, agent-query paths, session-style requests, and daemon status. These rows are diagnostic product evidence, not substitutes for the LS-Lint-equivalent cold CLI gate above.

Native report: native-current.json
History: native-history.jsonl
Rows: 150 passing rows across 6 fixtures
Evidence source: Commit 4eb5d51; source lane dirty when materialized
Captured: 2026-07-05 02:08:48 UTC
Command: assura performance-report --suite native --output benches/history/native-current.json --history benches/history/native-history.jsonl --website-dir website/public/data/performance --iterations 5

Native fixture	Scale	Rows	Slowest measured row
native small `native_small`	66 checked, 1 ignored, 7 dirs native content authoring matrix with 25 goals, 25 specs, 5 ADRs, Markdown links, relations, context packs, and search	25 native diagnostic rows Gate: `cargo xtask native-performance-no-regression`	48.0 ms `native:daemon-status-cli`
native medium `native_medium`	651 checked, 1 ignored, 7 dirs native content authoring matrix with 250 goals, 250 specs, 50 ADRs, Markdown links, relations, context packs, and search	25 native diagnostic rows Gate: `cargo xtask native-performance-no-regression`	149.4 ms `native:daemon-status-cli`
native large `native_large`	6501 checked, 1 ignored, 7 dirs native content authoring matrix with 2500 goals, 2500 specs, 500 ADRs, Markdown links, relations, context packs, and search	25 native diagnostic rows Gate: `cargo xtask native-performance-no-regression`	1635.6 ms `native:daemon-status-cli`
native reference heavy `native_reference_heavy`	651 checked, 1 ignored, 7 dirs native reference-heavy matrix with many-to-many relations, missing targets, Markdown links, and diagnostic classification	25 native diagnostic rows Gate: `cargo xtask native-performance-no-regression`	152.0 ms `native:daemon-status-cli`
native adapter mix `native_adapter_mix`	652 checked, 1 ignored, 8 dirs native adapter-mix matrix with Markdown frontmatter, JSON records, code-symbol references, relations, and context packs	25 native diagnostic rows Gate: `cargo xtask native-performance-no-regression`	141.8 ms `native:daemon-status-cli`
native real project `native_real_project`	326 checked, 1 ignored, 8 dirs native real-project shaped content authoring matrix with goals, specs, ADRs, notes, broken references, and agent-query rows	25 native diagnostic rows Gate: `cargo xtask native-performance-no-regression`	104.2 ms `native:daemon-status-cli`

How To Read This Page

Use the table above for the scoped product comparison behind the public assura check path: one Assura run versus one native LS-Lint binary run on the same repository checkout and equivalent LS-Lint-compatible ruleset. The benchmark row is named assura-cli because the measured command is the primary installable assura launcher, which routes supported one-shot assura check invocations through the lightweight checker before dispatching less common commands to assura-full. The timed command uses quiet successful validation so the comparison measures rule checking and traversal rather than JSON report serialization. The LS-Lint binary is resolved from the pinned @ls-lint/ls-lint package before timing so the measured loop does not include npm package resolution or the package’s Node wrapper.

The current checked data uses the realistic-equivalent report cohort: eight in-repo fixtures that exercise LS-Lint-compatible naming, ignore, scope, and extension behavior without requiring an external checkout cache. Pinned real-repository rows are optional extended evidence when a report is generated with the external fixture flag; they are not part of the checked report rendered on this page today.

Rows marked fixture_acceptance=accepted-ls-lint-equivalent are the enforced fixture floor. CI runs cargo xtask performance-no-slower against the checked report and fails if any accepted assura-cli fixture row is slower than the matching native ls-lint-cli row, even when aggregate numbers still look good.

The warm gate is a separate editor-session measurement. It uses a persistent assura-check-session CLI process connected to assura-checkd, so it is not a replacement for the cold one-process assura check headline path.

Assura-native rows are a second checked report, not a replacement for the LS-Lint comparison. They measure product surfaces LS-Lint does not provide: content collections, document graph queries, references, context packs, Markdown safe-fix dry runs, agent-query paths, session-style requests, and daemon status. Native rows use fixture_acceptance=assura-native-diagnostic and are protected by cargo xtask native-performance-no-regression. That gate is baseline-relative, not LS-Lint-relative: each native row carries native_regression_* metadata calibrated from the checked native baseline so reviewers can see whether the current row remains inside the expected envelope.

The post-beta LS-Lint reassessment keeps those claims separate: current accepted cold rows must be no slower than native LS-Lint row by row, while remaining 2x misses are explained through process floor, Rust CLI floor, config-load, checker-init, walk-and-validate, and report-sort phase rows.

Recent Product Gains

The native suite has moved on product-path work, not just harness/reporting:

native_large native:phase:fact-ingest-load: 2993.6 ms -> 137.0 ms (-95.4%)
native_large native:phase:incremental-replace-generation: 2733.0 ms -> 118.3 ms (-95.7%)
native_large native:phase:object-load-validate: 236.3 ms -> 188.7 ms (-20.1%)
native_large native:phase:repository-validate-total: 257.7 ms -> 209.8 ms (-18.6%)

Those deltas come from the checked native history under /data/performance/native-history.jsonl, not from ad hoc local smoke runs. The remaining LS-Lint-equivalent cold miss is much narrower: the checked many_configured_scopes_regression row is already faster than native LS-Lint (14.95 ms versus 19.08 ms), but it is still not at the separate 2x target. The current cold cost on that row is concentrated in assura:phase:config-load (4.15 ms) and assura:phase:walk-and-validate (5.12 ms), so the next headline work is still product-path runtime rather than more report plumbing.

Beta Performance Posture

For beta, the cold LS-Lint release guardrail is row-by-row no-slower parity, not universal cold 2x. The current checked report has every accepted cold LS-Lint-equivalent fixture faster than native LS-Lint, while warm/session rows meet the 2x target across the accepted fixture set. Remaining cold 2x misses are tracked as diagnostics so they can guide future work without forcing entrypoint-specific branches or sub-millisecond tuning into the product.

Future cold optimizations must improve the ordinary public command. A change is not enough if it only helps a phase row, an in-process row, assura-check-cli, or a traversal experiment while assura check --quiet is flat or slower. The working stop/keep matrix is maintained in docs/analysis/2026-07-05-performance-decision-matrix.md.

More Detail

Performance test cases defines the current checked cohort, optional pinned real repositories, generated support cases, and diagnostic stress cases.
Performance implementation notes explains traversal rows, why walkdir is the default non-fail-fast traversal path, and why implementation-detail rows are not headline product evidence.
Post-beta LS-Lint performance reassessment records the current no-slower result, cold/warm claim split, and phase-level CLI-floor attribution.
Current comparison report is the exact machine-readable report rendered on this page.
Chart-ready JSONL history is a rolling audit trail capped to the most recent 1,000 rows, not a clean release-over-release chart yet.
Current native report is the checked machine-readable Assura-native suite rendered on this page.
Native JSONL history is the native suite audit trail used by regression review.

Updating The Baseline

Baseline updates are explicit source changes:

cargo build --release --bin assura --no-default-features --features json-output,yaml-config
cargo build --release --bin assura-full
cargo build --release -p assura-check-cli
target/release/assura performance-report \
  --output benches/history/current.json \
  --history benches/history/ls-lint-comparison-history.jsonl \
  --website-dir website/public/data/performance \
  --iterations 5

Refresh the native Assura report with:

cargo build --release --bin assura --no-default-features --features json-output,yaml-config
cargo build --release --bin assura-full
cargo build --release -p assura-check-cli
target/release/assura performance-report \
  --suite native \
  --output benches/history/native-current.json \
  --history benches/history/native-history.jsonl \
  --website-dir website/public/data/performance \
  --iterations 5
cargo xtask native-performance-no-regression benches/history/native-current.json

Review the generated diff in benches/history/ and website/public/data/performance/ before merging.

When a checked refresh is produced from a temporary clean snapshot of a dirty source lane, also export the source provenance before running the command so the checked artifact preserves the original branch/commit/patch tuple and the fact that the source lane was dirty:

export ASSURA_SOURCE_COMMIT_SHA="<source-lane commit>"
export ASSURA_SOURCE_BRANCH="<source-lane branch>"
export ASSURA_SOURCE_PATCH_ID="<stable patch id for the materialized diff>"
export ASSURA_SOURCE_WORKTREE_DIRTY=true