FAULT FINDER · MEMORY EVIDENCE · SCHEMA {{ evidence.schema_version }}

{{ evidence.created_at }} · {{ evidence.backend }} execution · {{ evidence.id }}

{% if evidence.assessment.growth_per_operation is not none %}

{{ '%.3f'|format(evidence.assessment.growth_per_operation) }} {{ evidence.assessment.metric }} per operation over the tested range.

{% endif %}

Measured retention

Each point is one fresh-process trial. Blue: subject. Green: control. Metric: {{ metric }}. Negative byte deltas are plotted at zero; exact values remain below.

Trials

{% for t in evidence.trials %}{% if t.error %}{% endif %}{% endfor %}

Variant	Workload	Repeat	OK	Before GC	After GC	After owner release	Traced bytes
{{ t.variant }}	{{ t.completed }}/{{ t.size }}	{{ t.repeat+1 }}	{{ t.ok }}	{{ t.natural_survivors if t.natural_survivors is not none else '—' }}	{{ t.survivors if t.survivors is not none else '—' }}	{{ t.after_release if t.after_release is not none else '—' }}	{{ t.retained_bytes if t.retained_bytes is not none else '—' }}
{{ t.error }}

Scope and limitations

Measurements describe this workload. They are not production memory savings or proof that an application is leak-free.
{{ item }}

{% if evidence.agent_review %}

Agent review

{% endif %}

Reproduce

faultfinder replay <this-bundle-directory>

Local experiments require an explicit --trusted-local flag. Generated code requires Docker. Source and harness hashes are validated before replay.

Provenance

Commit	{{ evidence.project.get('commit', '') or 'Not available' }}
Source digest	{{ evidence.project.get('source_sha256', '') }}
Dirty source	{{ evidence.project.get('dirty', false) }}
Policy	{{ evidence.plan.policy }}