KMP-IMPACT

Data Contracts

The Pydantic v2 models that travel between phases, plus the AnalysisConfig dataclass. The same models govern both the in-memory representation and the on-disk JSON files. The authoritative declaration lives in src/kmp_impact_analyzer/contracts.py.

Contracts

Type	Purpose	Page
`AnalysisConfig`	The CLI surface flattened into a configuration dataclass.	Open
`VersionChange`	One detected catalog change — emitted by `detect-version-changes`.	Open
`ShadowManifest`	Phase 1 output: where the BEFORE and AFTER trees live on disk.	Open
`FileImpact`	One impacted Kotlin file (Phase 2).	Open
`ExpectActualPair`	A detected `expect`/`actual` pairing (Phase 2).	Open
`ImpactGraph`	Phase 2 aggregate output.	Open
`UIRegressions`	Phase 3 dynamic output — UTG diff result.	Open
`ConsolidatedResult`	Phase 4 canonical artefact consumed by the HTML report and the PR comment.	Open
`EvaluationResult`	`kmp-impact evaluate` output — Precision / Recall / F1 plus TP/FP/FN lists.	Open

Conventions

All Pydantic contracts use from_attributes=False. Fields default to null or to the empty collection.
Enum values are lowercase strings on disk ("direct", "transitive", "completed", "blocked", "skipped").
AnalysisConfig is a plain @dataclass, not a Pydantic model — it is constructed from YAML or CLI flags and is never serialised to disk in the canonical pipeline.
The HTML renderer computes some derived values (risk label, recommendation prose) that are not persisted in ConsolidatedResult. They are generated at report time.

Working with the JSON on disk

To consume the artefacts from another language, generate the JSON Schema:

python -c "
from kmp_impact_analyzer.contracts import ConsolidatedResult
import json; print(json.dumps(ConsolidatedResult.model_json_schema(), indent=2))
" > consolidated_result.schema.json

Smaller helper types

A few short types are documented inline rather than on their own pages.

`VersionChange` { #versionchange }

{
  "dependency_group": "io.ktor",
  "version_key": "ktor",
  "before": "2.3.8",
  "after": "2.3.11"
}

Returned in batches inside the VersionChangeSet emitted by kmp-impact detect-version-changes.

`ShadowManifest` { #shadowmanifest }

{
  "before_dir": "output/phase1/before",
  "after_dir":  "output/phase1/after",
  "version_change": { /* VersionChange */ },
  "init_script_injected": true,
  "detekt_reports": {},
  "kover_reports": {},
  "compilation_after": {},
  "compile_targets": { ":android": true, ":shared": true }
}

compile_targets records the result of ./gradlew :<module>:assembleDebug for each detected Android target. Failed targets are still listed with false, which is how Phase 3 decides whether to mark itself BLOCKED.

`EvaluationResult` { #evaluationresult }

{
  "scenario": "pokedex_ktor_minor",
  "precision": 0.83, "recall": 0.75, "f1": 0.79,
  "true_positives":  ["..."],
  "false_positives": ["..."],
  "false_negatives": ["..."],
  "screen_precision": 1.0, "screen_recall": 1.0, "screen_f1": 1.0,
  "screen_tp": [], "screen_fp": [], "screen_fn": []
}

Written by kmp-impact evaluate.

This site is open source. Improve this page.