Commit graph

2 commits

Author SHA1 Message Date
425a0e5790 feat(live): real live mode via HubClient.submit → outputs
HubRepository.list() now runs the inline reclaim-evaluate flow per
pilot norm via submit() and maps the returned outputs to Evaluation
— the correct path, since the hub audit log stores only hashes, not
outputs. skm/benefit get their structured input as JSON (submit
jsonInputs), so the live heatmap gets real Schaden/Nutzen axes.

- live_inputs.dart: the inline eval flow + the 5 pilot norms' inputs
  (XML asset, cohort_id, skm duties, benefit ratings) plus curated
  overrides the bund adapter can't derive (true Geltungsbereich for
  the EU/Berlin norms, freshness, jurist sign-off).
- assets/norms/*.xml: 5 gesetze-im-internet-style source documents.
- tool/live_smoke.dart: dev smoke test (5/5 evaluate against a live
  hub: Schaden 3M–450M €, Nutzen 2.0–3.8, EU/Berlin scope correct).

Needs econ.skm_score's float-population fix (chain-modules commit
80f6620).

Signed-off-by: flemming-it <sf@flemming.it>
2026-06-19 02:39:15 +02:00
8ee6a294a1 feat: initial F∆I Law-Heatmap Phase 0 scaffold
First end-to-end-buildable cut of the pflicht-/graph-basierte
Wirkungsanalyse von Rechtsnormen, running on Ch∆In. Demo runs
without external dependencies and without a hub connection so
the methodology + UI stack can be reviewed before any Verbands-
Kooperation or Beirats-Validierung kicks in.

Tree layout

  MACHBARKEITSSTUDIE.md       v0.3 — pflicht-basierte Methodik,
                              4-Tier-Evidenz, 8 law*-Module
                              auf Ch∆In statt der alten F∆I-
                              Plattform.
  RUN.md                      one-page how-to-run.
  flows/durchstich-gewo-14.yaml
                              vertical Phase-0 flow: pull →
                              normalize → duties → citations →
                              cohort → skm → frust → benefit →
                              attribution → system.approval.
  app/                        Flutter macOS client, design
                              language pulled from fai_web
                              (ink #08090A, paper #F4F3EF,
                              petrol signal #2E8F9E). Mock
                              repository ships five fixture
                              norms with real source URLs and a
                              permanent T4 demo-banner so no
                              figure can be mistaken for a
                              validated one.

Module repos (separate, see chain-modules*/):

  text-akoma-normalize, text-deontic-extract, graph-citation-
  extract, text-readability-score, stats-cohort-size, graph-
  shapley-attribution, econ-skm-score, law-benefit-score.

What runs today

  - flutter analyze: 0 issues
  - flutter test: 2/2 (landing + nav-to-shell smoke)
  - flutter build macos (via app/build-macos.sh) and ad-hoc
    codesign, app launches and the Dart VM service comes up
  - native cargo test green on every module
  - cargo build --release --target wasm32-wasip2 produces a
    130 KiB artefact for text-akoma-normalize

What is deliberately mock / stub

  - gRPC wire from the client to a chain serve hub (Repository
    abstraction is in place; live impl is the next step)
  - NKR Bürokratiekosten-Datenbank ingestion (for the canonical
    h-Werte that close out the Engpass per Studie §6.7)
  - DESTATIS GENESIS-API adapter for stats.cohort_size

License: Apache-2.0. Author/contact in MACHBARKEITSSTUDIE.md.
Signed-off-by: flemming-it <sf@flemming.it>
2026-06-18 11:26:28 +02:00