Result Interpretation

The workbench produces rich JSON outputs.

This is intentional. The raw outputs are meant to be:

inspectable by humans
parseable by AI agents
suitable for reproducible pipelines
traceable back to model inputs

Result Envelope

All current result documents use the same top-level envelope:

result_type
inputs
content
diagnostics

inputs identifies what was analyzed.

content contains the operation-specific analytical result.

diagnostics records non-fatal notes, limitations, or processing details.

Reading Assessment Results

An assessment_result should be read as a model instrument panel.

The main sections are:

metrics
maintainability findings
module summaries
trace index

The metrics show model size, explicitness, constraint readiness, and interoperability signals.

Maintainability findings should be inspected as candidate issues, not automatically as defects.

Reading Coverage Results

A coverage_result answers whether a declared use case is representable from the composed SHACL model.

Important fields are:

overall_coverage_class
required_item_findings
required_join_findings
supporting_trace

supporting_trace is important because it shows why the workbench classified an item or join in a certain way.

Reading Comparison Results

A comparison_result compares exactly two assessment results.

Important sections are:

metric_deltas
ranked_observations
optional alignment

ranked_observations are sorted to emphasize normalized analytical signals before raw size differences.

When an alignment file is provided, inspect:

alignment_coverage_ratio
evaluated_pairs
ranked_alignment_observations

Alignment results are declaration-based. The workbench does not infer semantic equivalence automatically.

Reading Prioritization Results

A prioritization_result is a ranked follow-up list.

Each target includes:

target_id
priority_rank
message
rule_trace

rule_trace explains why the target was ranked.

Prioritization is not a final quality judgment. It is an interpretable queue for follow-up analysis.

Lightweight Interpretation Layer

The tool includes a preliminary simplified interpretation operation:

summarize

It produces a compact summary_result from one or more existing result documents.

This component does not replace the raw outputs. It sits on top of them.

It is deterministic and rule-based.

The summary does not use AI interpretation. It selects signals using fixed rules:

coverage findings are ordered by fixed coverage severity
comparison signals use the already ranked comparison observations
alignment gaps use the already ranked alignment observations
prioritization signals use the already ranked target list
only a small fixed number of strongest signals is promoted

The CLI can also render the same deterministic summary as Markdown:

dpawb summarize --result comparison.json --format markdown

Example output shape:

{
  "result_type": "summary_result",
  "inputs": {
    "result_count": 2
  },
  "content": {
    "headline": "Both compared profiles satisfy the selected use case; main differences are constraint explicitness and modelling compactness.",
    "key_points": [
      "Coverage is representable on both sides.",
      "BatteryPass is more compact on the reduced slice.",
      "CE-RISE represents the same concepts through a composed model set."
    ],
    "follow_up_questions": [
      "Should compactness or modular extensibility matter more for this use case?"
    ]
  },
  "diagnostics": []
}

This would help humans and AI agents interpret rich outputs consistently while preserving deterministic analysis underneath.

Keyboard shortcuts