Steady call path
1.00× baselineThe selected integer call fixture matches the Rust-full baseline after warm-up.
Measurements · Methods · Limitations
A benchmark result is meaningful only within its workload, method, and environment.
How do selected Hara runtime paths compare with the named baselines under this method and environment?
benchmarks.report
sha256:bench-2026-08-19-b19e 5 workload fixtures - Report revision
sha256:bench-2026-08-19-b19e- Methodology
- methodology:runtime-shootout/v3
- Evidence bundle
- evidence:runtime-shootout/2026-08-19
- Comparability review
- review-fixture
01 · Result summary
Begin with the measured observations and their confidence.
Each summary states the comparison, confidence, interpretation, and evidence reference. Missing or incomparable workloads remain visible and are not converted into a ranking.
Persistent collection path
1.05–1.31×Map and parser workloads remain close in this fixture, while allocation-sensitive paths require separate interpretation.
Cold start
Not comparableWasm download, JVM startup, and native process launch answer different questions and are not combined.
02 · Workload matrix
Each value resolves to its exact comparison context.
Filters produce a shareable state. Cells preserve selected, baseline, comparable, missing, incomparable, and low-confidence meanings without relying on colour alone.
| Workload | Category | Hara native | Rust full | Julia | Node |
|---|---|---|---|---|---|
| ns/op | Dispatch | ||||
| ns/op | Collections | ||||
| ns/op | Collections | ||||
| μs/op | Parsing | ||||
| ns/op | Interop |
Dispatch
integer-call
- Hara native
- 18.4selected
- Rust full
- 18.4baseline
- Julia
- 22.9comparable
- Node
- 38.2comparable
Collections
vector-map
- Hara native
- 41.7comparable
- Rust full
- 39.8baseline
- Julia
- 49.2comparable
- Node
- 78.4comparable
Collections
persistent-map
- Hara native
- 113comparable
- Rust full
- 108baseline
- Julia
- Different structureincomparable
- Node
- 172comparable
Parsing
parser-small
- Hara native
- 6.2comparable
- Rust full
- 5.9baseline
- Julia
- 7.8comparable
- Node
- Missingmissing
Interop
ffi-roundtrip
- Hara native
- 84.1low confidence
- Rust full
- 31.4baseline
- Julia
- 72.8low confidence
- Node
- 91.6low confidence
03 · Selected result
Samples, baseline, environment, revisions, and method remain adjacent.
A selected matrix cell resolves to a complete measurement record with source and harness revisions, raw path, confidence, environment, and methodology.
Selected workload
integer-call
Hara native vs Rust full
Hara18.4 ns/op
Baseline18.4 ns/op
Ratio1.00×
Samples30
Warmups12
RSD1.7%
- Hara source
hara@9b6c10e- Harness
bench@b19e4a2- Evidence digest
sha256:b19e4a2f- Raw samples
evidence/runtime-shootout/2026-08-19/integer-call.json
04 · Revision history
A result belongs to a sequence of source and harness revisions.
History separates measured movement from its proposed explanation. Each note remains tied to a commit and a method-compatible evidence bundle.
18.4 19 Aug
19 12 Aug
19.2 05 Aug
19.1 29 Jul
9b6c10e18.4 ns/op−3.2%Dispatch cache avoids repeated symbol resolution.
740a3c819 ns/op−1.0%Lowering emits direct integer path.
3d1120a19.2 ns/op+0.5%Namespace registry safety check added.
0f6c41b19.1 ns/opbaselineFirst comparable harness revision.
05 · Missing, incomparable, and low-confidence results
Incomplete evidence remains visible without producing a conclusion.
The matrix names why a ratio or comparison is withheld instead of assigning a placeholder value or removing the workload.
Missing result
Node parser sample did not publish
The matrix shows a missing cell and keeps the workload visible instead of substituting zero or hiding the row.
- Harness exited before receipt
- No ratio calculated
- Rerun request attached
Incomparable
Persistent collection semantics differ
The selected baseline uses mutable storage, so the interface withholds a winner and explains the semantic mismatch.
- State: partial
- Reviewer: benchmark registry
- Reason is shareable
Low confidence
FFI variance exceeds the review threshold
Values remain inspectable, but summary and ranking surfaces do not promote the result.
- RSD: 12.8%
- Threshold: 5%
- Affinity investigation open
- selected
- 1
- baseline
- 5
- comparable
- 7
- comparable
- 2
- incomparable
- 1
- missing
- 1
- low confidence
- 3
06 · Desktop, tablet, and mobile layouts
Narrow layouts preserve the accessible comparison table.
The desktop matrix remains the canonical table. Narrow screens add workload cards for scanning while the table remains available to assistive technology and horizontal inspection.
07 · Source and evidence table
The complete record follows the readable summary and selected result.
This table provides exact source, Hara revision, baseline version, methodology, evidence digest, and review state for every workload.
| Workload | Source | Hara revision | Baseline | Method | Evidence | State |
|---|---|---|---|---|---|---|
| integer-call | bench/int-call.hal | 9b6c10e | rust@1.89.0 | runtime-shootout/v3 | b19e4a2f | Comparable |
| vector-map | bench/vector-map.hal | 9b6c10e | rust@1.89.0 | runtime-shootout/v3 | 8ca9117d | Comparable |
| persistent-map | bench/persistent-map.hal | 9b6c10e | rust@1.89.0 | collections/v2 | a0dc8271 | Partial |
| parser-small | bench/parser-small.hal | 9b6c10e | rust@1.89.0 | parser/v4 | 3e72be10 | Missing Node |
| ffi-roundtrip | bench/ffi-roundtrip.hal | 9b6c10e | rust@1.89.0 | interop/v1 | f88c020a | Low confidence |