Evidence-graphics design fixture Representative benchmark, compatibility and runtime evidence for visual review. Values are deterministic fixtures, not current Hara performance or support claims. hara-v2-data-visualisation-fixture-2026-08 · 9a88bddd7a539d7a

Issue 91 · evidence before decoration

Show what was measured, what is uncertain, and what cannot be compared.

The guide records accessible patterns for benchmark evidence, package compatibility, and runtime observations. Scales, units, baselines, confidence, source revision, and methodology remain visible while each source system retains authority over its facts.

Compare benchmark evidence Inspect compatibility states Read a runtime sequence

Reference composition 01 · Benchmarks

Lead with the metric, then expose the evidence.

Bars communicate ordered magnitude. Confidence ranges, sample count, baseline, environment, methodology and exact evidence revision remain adjacent rather than hidden behind a tooltip.

Median wall time · ms

Cold-start executable form

lower is better · baseline 0 ms · domain 0–16 ms

bench-fixture:7b2ac941

Hara native30 samples · run:hara-native:7b2ac941

4.8 ms

95% CI 4.5–5.2 ms

Babashka30 samples · run:babashka:3c88fe20

6.7 ms

95% CI 6.2–7.3 ms

Clojure JVM30 samples · run:clojure-jvm:d67e091b

12.4 ms

95% CI 11.7–13.2 ms

Hara Wasmtarget lacks an equivalent process-start boundary

≠IncomparableNo equivalent process-start boundary

Samples30 per measured series

Confidence95% bootstrap interval

EnvironmentLinux x86_64 · 8 vCPU · 16 GiB · isolated process

Generated2026-08-20T19:20:00Z

Accessible data table and methodology

Thirty cold processes per implementation; median shown with a 95% bootstrap interval.

Cold-start executable form — Median wall time; ms; lower is better.
Implementation	Median	95% interval	Samples	State	Evidence
Hara native	4.8 ms	4.5–5.2 ms	30	Measured	`run:hara-native:7b2ac941`
Babashka	6.7 ms	6.2–7.3 ms	30	Measured	`run:babashka:3c88fe20`
Clojure JVM	12.4 ms	11.7–13.2 ms	30	Measured	`run:clojure-jvm:d67e091b`
Hara Wasm	Not comparable	Not available	0	incomparable	`target lacks an equivalent process-start boundary`

Parse 10k forms

1ce20d9a

Hara18.2 ms

Clojure31.6 ms

Babashka22.9 ms

Map 1m integers

830c1be2

Hara34.6 ms

Clojure62.1 ms

Babashka54.8 ms

Dispatch 100k calls

721d829e

Hara13.7 ms

Clojure24.4 ms

Babashka— Missing

Distribution, not only a point estimate

Hara native cold-start distribution

distribution:7b2ac941

34.2–4.5

94.5–4.8

124.8–5.1

55.1–5.4

15.4–5.7

Histogram counts for Hara native cold-start distribution
Range (ms)	Count
4.2–4.5	3
4.5–4.8	9
4.8–5.1	12
5.1–5.4	5
5.4–5.7	1

Reference composition 02 · Packages

A compatibility matrix is a claim ledger, not coloured wallpaper.

Every cell names its state, target, release revision and supporting evidence. Missing, unsupported and incomparable remain different facts; none is rendered as zero or silent absence.

Package/runtime compatibility

Exact release evidence across four target families

Cell symbol + state word + evidence text · fixtures only

compatibility-fixture@2026-08

Representative package compatibility. These deterministic fixtures are not current registry claims.
Package and release revision	JVM 21	Rust native	Wasm32	JavaScript host
hara/std.typed@0.9.2`release:typed:b37d9b71`	supported 0.9.2 61/61 conformance	partial 0.9.2 57/61; recursive aliases pending	supported 0.9.2 40/40 portable subset	unsupported No version claim No direct implementation
hara/std.work@0.3.0-experimental`release:work:d4502d3d`	supported 0.3.0 Reference executor	partial 0.3.0 Checkpoint store only	unsupported No version claim Durability provider unavailable	incomparable No version claim Host adapter has no durable-step contract
hara/std.native@2.0.0`release:native:845cf281`	supported 2.0.0 Portable + JVM capability inventory	supported 2.0.0 Portable + native capability inventory	partial 2.0.0 Browser-safe capability subset	missing No version claim Evidence index not received
hara/hara-wasm-core@1.4.0`release:wasm:71cc80f4`	incomparable No version claim Browser session product, not a JVM package	incomparable No version claim Compiled through Rust but not installed as a native runtime	supported 1.4.0 Browser session contract 40/40	partial 1.4.0 JavaScript host bridge; evaluator remains Wasm

✓Supported

An exact implementation and evidence set supports the declared release.

—Missing

The expected evidence value was not received. Keep the empty fact and source boundary visible.

◐Partial

Only a named subset is available. State the numerator, denominator or missing dimensions.

×Unsupported

The implementation or target cannot produce this measure. Do not render it as zero.

≠Incomparable

The measurements do not share an equivalent boundary, unit or methodology.

hara/std.typed@0.9.2release:typed:b37d9b71

JVM 21: ✓ supported
61/61 conformance
Rust native: ◐ partial
57/61; recursive aliases pending
Wasm32: ✓ supported
40/40 portable subset
JavaScript host: × unsupported
No direct implementation

hara/std.work@0.3.0-experimentalrelease:work:d4502d3d

JVM 21: ✓ supported
Reference executor
Rust native: ◐ partial
Checkpoint store only
Wasm32: × unsupported
Durability provider unavailable
JavaScript host: ≠ incomparable
Host adapter has no durable-step contract

hara/std.native@2.0.0release:native:845cf281

JVM 21: ✓ supported
Portable + JVM capability inventory
Rust native: ✓ supported
Portable + native capability inventory
Wasm32: ◐ partial
Browser-safe capability subset
JavaScript host: — missing
Evidence index not received

hara/hara-wasm-core@1.4.0release:wasm:71cc80f4

JVM 21: ≠ incomparable
Browser session product, not a JVM package
Rust native: ≠ incomparable
Compiled through Rust but not installed as a native runtime
Wasm32: ✓ supported
Browser session contract 40/40
JavaScript host: ◐ partial
JavaScript host bridge; evaluator remains Wasm

Reference composition 03 · Runtime

Telemetry reads as a monotonic event record.

Sequence, generation, source revision, backend and capability fence stay visible while dense observations scroll locally. High-frequency telemetry is not announced as a continuous live region.

Session telemetry

pg-7a91 · generation 4

interpreter · sequence 140–148 · 2026-08-20T19:22:04.100Z/2026-08-20T19:22:05.012Z

source:9f3c2ab7

Evaluation 83 ms ready

Queue wait 12 ms ready

Heap estimate 18.4 MiB partial

3D capability × unsupported

140
session.ready+0 ms · success
generation 4 · interpreter
141
eval.queued+12 ms · queued
source 9f3c2ab7
142
eval.started+21 ms · active
capabilities eval · observations
143
namespace.loaded+46 ms · success
tutorial.profile
144
observation+73 ms · partial
heap estimate sampled
145
eval.completed+95 ms · success
83 ms
146
canvas.unavailable+108 ms · unsupported
sample declares no canvas capability
147
heartbeat+514 ms · stale
interval exceeded 400 ms review threshold
148
session.observed+912 ms · success
monotonic sequence retained

Accessible event table and session fence

Session: pg-7a91
Generation: 4
Source revision: source:9f3c2ab7
Backend: interpreter
Capability revision: capabilities:1f72aa90
Sequence range: 140–148

Monotonic event sequence for pg-7a91, generation 4
Sequence	Offset	Event	State	Detail
140	+0 ms	`session.ready`	success	generation 4 · interpreter
141	+12 ms	`eval.queued`	queued	source 9f3c2ab7
142	+21 ms	`eval.started`	active	capabilities eval · observations
143	+46 ms	`namespace.loaded`	success	tutorial.profile
144	+73 ms	`observation`	partial	heap estimate sampled
145	+95 ms	`eval.completed`	success	83 ms
146	+108 ms	`canvas.unavailable`	unsupported	sample declares no canvas capability
147	+514 ms	`heartbeat`	stale	interval exceeded 400 ms review threshold
148	+912 ms	`session.observed`	success	monotonic sequence retained

Evidence states and ownership

Absence has more than one meaning.

Shared presentation grammar keeps these states legible. Benchmarks, package registries and runtimes still own the underlying values, freshness policy, methodology and compatibility decisions.

—

Missing

The expected evidence value was not received. Keep the empty fact and source boundary visible.

△

Stale

The value exists but is older than the declared freshness window. Show its timestamp and threshold.

◐

Partial

Only a named subset is available. State the numerator, denominator or missing dimensions.

Unsupported

The implementation or target cannot produce this measure. Do not render it as zero.

≠

Incomparable

The measurements do not share an equivalent boundary, unit or methodology.

Shared visual grammar8 rules

Name metric, unit, direction and baseline next to the graphic.
Retain exact source, methodology, environment and revision references.
Use position and length before area, volume or decorative perspective.
Pair colour with words, symbols, patterns or structural placement.
Keep confidence, ranges and sample counts visible when a point estimate is shown.
Render missing, stale, partial, unsupported and incomparable as different facts.
Provide an accessible table or ordered textual alternative for every evidence graphic.
Let products own data and methodology; the package owns presentation grammar only.

Downstream adoption

Adopt the grammar without copying the fixtures.

The additive stylesheet supplies evidence-graphic composition classes. Products bind those classes to their own exact data, methodology, states and accessible descriptions.

Benchmarks

Use comparison, confidence, distribution and table patterns while retaining workload and methodology authority in `hara-benchmarks`.

Packages

Use the matrix and state grammar while retaining exact release and target evidence in the package registry.

Playground and tools

Use event-sequence and telemetry patterns while runtime systems remain authoritative for observations and receipts.

Docs and WWW

Use simplified evidence projections with direct links to the exact data, method and current revision.