NeuroCVguard — research report

{{ check.status | replace('_',' ') }}{{ check.severity }}Evidence: {{ check.evidence_kind }}

{% if check.scope %}

{% for key, value in check.scope.items() %}{{ key | replace('_',' ') }}: {{ value }}{% if not loop.last %} · {% endif %}{% endfor %}

{% endif %}

Next action: {{ check.recommendation }}

{% set table = check.evidence.get('display_table') %} {% if table %}

{% for level in table.target_levels %}{% endfor %}{% for row in table.counts %}{% set i = loop.index0 %}{% for count in row %}{% endfor %}{% endfor %}

Descriptive participant counts · local category aliases
Acquisition category	{{ level }}
{{ table.row_levels[i] }}	{{ count }}

{{ table.estimator }} · Value: {{ table.statistic.value if table.statistic.value is not none else 'undefined' }}{% if table.statistic.reason %} ({{ table.statistic.reason }}){% endif %}. Descriptive association is not proof of causality or model shortcut use.

{% elif check.evidence.get('details_omitted') %}

Table details omitted. Linked cells, totals and statistics are not included in this projection. This may reflect privacy suppression or unavailable detail.

{% endif %} {% if sensitive and check.evidence %}

{{ check.evidence | pretty_json }}

{% elif check.evidence.get('statistic') %}

Statistic: {{ check.evidence.statistic.value }} · Reason: {{ check.evidence.statistic.reason }}

{% endif %}

1. Scope and objective

{% if comparison_only %}

Comparison of precomputed design summaries. Objectives are stated separately for each design. Cohort inventory and check coverage were not supplied.

{% else %}

Objective: {{ report.objective | replace('_',' ') }} · Result: {{ report.result_type }} · Execution: {{ report.execution_status }}

{% endif %}

Version {{ version }} · Schema {{ report.get('schema_version', '1.0 (comparison-summary)') }}. No audit, model fit or statistic is recomputed during rendering.

2. Execution and coverage

{% if incomplete %}

Incomplete assessment coverage

Unassessable checks or missing assessment context remain. No overall validity verdict is provided.

{% endif %} {% for status in ['pass','fail','not_assessable','not_applicable'] %}{% endfor %}

Coverage of supplied checks
Pass	Fail	Not assessable	Not applicable
{{ counts[status] }}

{% if not checks %}

No check inventory supplied.

{% endif %} {% for check in checks if check.status == 'not_assessable' %}{{ finding(check) }}{% endfor %}

3. Actionable findings

{% for check in checks if check.status == 'fail' %}{{ finding(check) }}{% else %}

No triggered findings in the supplied record. Review incomplete coverage before drawing conclusions.

{% endfor %}

4. Cohort and missingness

{% if report.get('input_summary') %}

{{ report.input_summary.n_observations }} observations · {{ report.input_summary.n_participants }} participants

Supplied roles: {{ report.input_summary.supplied_roles | join(', ') or 'None recorded' }}

Missing/incomplete roles: {{ report.input_summary.missing_roles | join(', ') or 'None recorded' }}

{% else %}

No cohort inventory supplied.

{% endif %} {% for check in checks if check.rule_id.startswith('NCG-COHORT-') %}{{ finding(check) }}{% endfor %}

5. Partition checks

{% for check in checks if check.rule_id.startswith('NCG-SPLIT-') or check.rule_id.startswith('NCG-PLAN-') %}{{ finding(check) }}{% else %}

No partition checks supplied.

{% endfor %}

6. Acquisition and target associations

{% for check in checks if check.rule_id.startswith('NCG-ASSOC-') %}{{ finding(check) }}{% else %}

No association diagnostics supplied.

{% endfor %}

7. Evaluation and comparison

{% if report.get('evaluation_summary') %}

Participant-level precomputed evaluation. Null metrics and reasons are retained; failed folds are not removed.

{{ report.evaluation_summary | pretty_json }}

{% elif comparison %}

{% for metric in ['accuracy','balanced_accuracy','macro_f1','roc_auc'] %}{% endfor %}{% for design in comparison.designs %} {% for metric in ['accuracy','balanced_accuracy','macro_f1','roc_auc'] %}{% endfor %}{% endfor %}

Precomputed participant metrics · 0–1 scale
Design	Objective	Diagnostic only	{{ metric \| replace('_',' ') }}
{{ design.name }}	{{ design.objective }}	{{ design.diagnostic_only }}	{% if design.metrics %}{{ design.metrics[metric].value if design.metrics[metric].value is not none else 'Undefined' }}{% if design.metrics[metric].reason %} ({{ design.metrics[metric].reason }}){% endif %}{% else %}Unavailable or privacy-suppressed{% endif %}

{% for design in comparison.designs %}

{{ design.name }} · recorded context

{% set context = design.get('context') %} {% if context %}

Execution: {{ context.execution_status }} · {{ context.n_completed_folds }}/{{ context.n_folds }} folds completed · {{ context.n_participants }} participants · {{ context.n_observations }} observations

Training participants per fold: {{ context.training_participants_min }}–{{ context.training_participants_max }} · {{ context.n_features }} selected features · metric unit: {{ context.metric_unit }}

Cohort identity: {{ context.cohort_reference }} · Feature identity: {{ context.feature_reference }}. These are local aliases for equal recorded digests, not authenticated identities.

Classes: {{ context.class_order | join(', ') }} · Positive class: {{ context.positive_class if context.positive_class is not none else 'Unspecified' }}

Prescribed baseline: median imputation, standard scaling, logistic regression. Recorded C values: {{ context.recorded_C_values | join(', ') or 'None recorded' }} · Tuning recorded: {{ context.tuning_recorded }}. Consult the original configuration for complete model settings.

{% else %}

Legacy summary: design context was not recorded; no missing settings were inferred.

{% endif %} {% if design.metrics %}{% for item in design.metrics.per_class %}{% endfor %}

Pooled participant class support
Class	Support	Recall
{{ item.class_label }}	{{ item.support }}	{{ item.recall if item.recall is not none else 'Undefined' }}

{% else %}

Complete pooled class support is unavailable or privacy-suppressed.

{% endif %}

Upstream preprocessing remains unverified. Review the original evaluation's scoped findings and limitations.

{% endfor %}

{% for item in comparison.differences %}{% endfor %}

Signed score_difference · A minus B in metric units
A	B	Metric	Difference	Unavailable reason
{{ item.design_a }}	{{ item.design_b }}	{{ item.metric }}	{{ item.difference if item.difference is not none else 'Null' }}	{{ item.reasons \| join(', ') }}

{% else %}

No model evaluation or comparison supplied.

{% endif %}

8. Provenance

Upstream preprocessing was not verified. A Pipeline cannot repair earlier global fitting.

{% if report.get('provenance') %}

{{ report.provenance | pretty_json }}

{% endif %} {% for check in checks if check.rule_id.startswith('NCG-PROV-') %}{{ finding(check) }}{% endfor %}

9. Limitations

No clinical advice, certification or universal leakage-free verdict.
Privacy projection is not formal anonymization; aggregates may still be sensitive.
{{ item }}

10. Recommended next actions

Review unassessable checks and supply missing context.
Freeze the evaluation plan and document later amendments.
Keep identity-bearing split plans and private evaluation records separate from shareable reports.
{{ check.rule_id }}: {{ check.recommendation }}

11. Rule and source references

Normative source: NeuroCVguard Foundation 1.0.0, scientific contract, report privacy and cross-contract clarifications. Association method: SciPy contingency.association (R08); adjacent inferential work: mlconfound (R09). No inferential equivalence is claimed.

{% for rule in references %}{% endfor %}

Rules represented in this report
Rule	Meaning
{{ rule.id }} · {{ rule.name }}	{{ rule.meaning }}

{{ check.rule_id }}

1. Scope and objective

2. Execution and coverage

3. Actionable findings

4. Cohort and missingness

5. Partition checks

6. Acquisition and target associations

7. Evaluation and comparison

{{ design.name }} · recorded context

8. Provenance

9. Limitations

10. Recommended next actions

11. Rule and source references