Co-Study4Grid 0.8.0

Co-Study4Grid 0.8.0

A feature-rich release that broadens what an operator can do in a study,
consolidates the UI, adds two new ways to run the tool, and quietly
hardens the codebase underneath. Paired with expert_op4grid_recommender
0.2.4 (required for the GST / superposition paths); from this release on,
CI tracks the latest published recommender rather than a pinned version.

Upgrades from 0.7.5. Sessions saved on 0.7.x reload unchanged
(model / compute_overflow_graph / new action-type fields fall back
to their legacy defaults).

🧰 A bigger remedial-action vocabulary

• Redispatching actions — raise/lower a dispatchable generator with an
  editable signed MW delta (±10 MW default), end-to-end through scoring,
  the ActionCard editor, simulation, save/reload and the interaction log.
• GST estimation for injection pairs — the Explore-Pairs estimate now
  covers pairs involving load shedding / curtailment / redispatch (not just
  topology), via the recommender's Generalized Superposition Theorem. The
  "cannot be combined for estimation" caveats are gone.
• Restrict the recommender to chosen action types — a new Settings →
  Recommender control scopes proposals to a subset of families (fixing the
  long-standing "Min = 0 doesn't exclude a family" confusion).

✎ Interactive SLD topology editing

Build a remedial action by clicking breakers on a Single Line Diagram
(mirrors the library's manoeuvre_ihm): a live target-topology preview
(no load flow, stale flows greyed), an interactive maneuver list
(focus / remove / remove-block / reset), one-click Simulate into a
manual action card, and full combined-action support. Six new replay
events; full contract in docs/features/sld-topology-edit.md.

🎛️ Operator UI consolidation

• Light / dark theme — a true token-swap (single source of truth in
  styles/tokens.*), including a legibility pass over the pypowsybl NAD/SLD
  chrome and the overflow viewer; no first-paint flash.
• Shared Action-Filter rings — severity + action-type + Max-loading
  threshold in one sidebar strip that the feed, Action Overview, both
  modals and the overflow iframe all read from (so a card hidden in one
  place is hidden everywhere). Fixes a silent threshold no-op in two tables.
• Readability-first sidebar — the contingency picker and the
  (renamed) Remedial Actions feed swap at trigger; a sticky-banner
  Clear shortcut; an overload info bubble that absorbs the retired
  OverloadPanel; a collapsible shell that hands width to the diagram.
• Notices → one discrete pill (⚠ Notices N) instead of stacked
  banners, operator-dismissed.
• Per-stage execution-time breakdown on the "Suggestions produced
  by …" line (step1 / graph / prediction / assessment / enrichment +
  wall-clock), persisted in sessions — plus a one-load-flow optimization
  that reuses the N-1 contingency observation (~1–3 s saved per analysis).
• Inspect elements by their displayed name (e.g. LESQUIVE 400kV),
  not just the raw id, across every inspect surface.

▶️ Two new ways to run it

• Game Mode (?game=1) — a timed, scored wrapper: remediate each
  study with ≤ 3 actions before the timer, export a game_session.json a
  Codabench competition scores (60·R + 25·R·A + 15·R·T). Additive and
  inert unless enabled.
• Online deployment — a one-container HuggingFace Docker Space
  serving the SPA same-origin with the backend on :7860 (booting straight
  into Game Mode). Binary grids ship via Git LFS, with the large France
  225/400 kV grid transparently decompressed from network.xiidm.zip.

🧱 Under the hood

• Code-quality gate, hardened into a multi-axis guard (new in this
  release) — mypy now gates at 0 (a TYPE_CHECKING shared-state base
  lets the recommender mixins type-check as the one class they compose;
  zero runtime change), test coverage gates both ends (frontend vitest
  thresholds + backend fail_under = 72), and the gate gained
  dependency-free cyclomatic-complexity / nesting / logical-LoC /
  return-annotation metrics plus ratchets for weak casts and lint
  suppressions. The scan now also covers the previously-ungated
  recommenders/ package. 52 backend return annotations added; frontend
  as unknown as casts 19 → 12. Full record in
  docs/architecture/code-quality-analysis.md §§17–22.
• Internal decomposition — the action-variant-patch pipeline split out
  of diagram_mixin.py (1220 → 769 lines) into services/diagram/.
• Shipped first-run defaults (config.default.json) so a fresh
  checkout (and the Space) opens on a working study.

⚙️ Compatibility

• Requires expert_op4grid_recommender >= 0.2.4.
• Saved 0.7.x sessions reload with legacy-default fallbacks; the step-2
  NDJSON contract is additive-only.

Next up: decomposing the files now riding their tightened
size/complexity ceilings — scoped in
docs/proposals/decompose-ceiling-riders.md.

0.7.5 — Pluggable recommendation models`

Feature + polish release headlined by the pluggable recommendation models integration (paired with expert_op4grid_recommender 0.2.2), the new "Combined only" pin filter, and a couple of operator-reported regressions (config-modal stale-write, PyPSA-EUR Mercator layout) that landed on the way.

Highlights

Pluggable recommendation models (PR #145, paired with expert_op4grid_recommender PR #90 / 0.2.2)

The analysis pipeline no longer hardcodes the expert system: it dispatches to any class implementing the RecommenderModel ABC. Three models ship out of the box:

Name	Label	Requires overflow graph	Best for
expert	Expert system	Yes	Default — rule-based discovery + scoring on every action type.
random	Random	No	Sanity-check baseline. Samples uniformly from the action dictionary, augmented with synthetic reconnection / load-shedding / curtailment actions.
random_overflow	Random (post overflow analysis)	Yes	"Is the overflow analysis useful?" baseline. Samples uniformly inside the expert-reduced action space (rule filter + overflow paths + network existence).

• Selecting a model is a one-dropdown gesture in Settings → Recommender. The parameter inputs render dynamically from each model's params_spec() so the UI hides knobs the active model doesn't consume.
• The Compute Overflow Graph (step 1) toggle is locked-on for models that require it (requires_overflow_graph=True) and editable for the others — letting the operator opt in to the overflow analysis even with the Random model.
• The active model is persisted in session.json under analysis.active_model (backend ground truth) and configuration.model (operator intent at save time); same split for compute_overflow_graph.
• Third-party models plug in with three lines of code — implement the ABC, decorate with @register, ensure import. The frontend picks them up automatically via GET /api/models. See the new Plug Your Own Recommendation Model section in the README.

"Combined only" pin filter

New pin-scoped filter on both the Action Overview tab and the Overflow Analysis iframe sidebar. When enabled, the pin layers render combined pairs plus their two constituents (dimmed for context) and drop every other unitary / un-simulated pin — the Action Feed cards remain unfiltered. Round-tripped through the existing cs4g:filters postMessage envelope so both surfaces stay in lock-step.

Config-modal stale-write fix

Switching the config-file path and clicking Apply now actually sends the freshly loaded config to /api/config. The previous behaviour silently sent the previous render's closure values, which the auto-save effect then persisted back into the freshly loaded file — undoing the operator's selection.

PyPSA-EUR grid layouts use raw Mercator metres

The previous 8 000-unit rescaling default forced pypowsybl VL circles to overlap in dense regions like Paris. Default behaviour is now raw projected metres (~1.4 M span for the French grid); pass --target-width N to reproduce the legacy rescaled output.

Added

• Recommender model registry (expert_backend/recommenders/): registry.py, random_basic.py, random_overflow.py, synthetic_actions.py, overflow_path_filter.py, network_existence.py, _service_integration.py.
• GET /api/models endpoint — returns the full list of registered recommenders with their params_spec(), label and capability flags.
• ConfigRequest.model / ConfigRequest.compute_overflow_graph fields, echoed as active_model / compute_overflow_graph in the final result event of the step-2 NDJSON stream.
• Saved session model echo — session.analysis.active_model (backend ground truth) + session.configuration.model (operator intent), same split for compute_overflow_graph.
• Frontend types and hooks — ModelDescriptor / ModelParamSpec, recommenderModel / computeOverflowGraph / availableModels state in useSettings, dynamic dropdown + locked-vs-optional toggle states in SettingsModal, action-card VL chip reads action_topology.voltage_level_id (so OPEN / CLOSE coupling cards double-click-zoom to the correct VL).
• "Combined only" pin filter on the Action Overview tab and the Overflow Analysis iframe.
• docs/backend/ subfolder — README.md (broader backend overview) and recommender_models.md (relocated from docs/recommender_models.md with app-side integration + filter chain + step-by-step guide).
• "Plug Your Own Recommendation Model" section in the root README.md.
• Backend tests — tests/test_recommenders_registry.py, test_random_recommenders.py, test_model_selection_mixin.py, test_service_integration.py, test_models_api.py, test_network_existence.py, test_overflow_path_filter.py, test_action_enrichment.py.

Changed

• extract_action_topology robustness (expert_backend/services/analysis/action_enrichment.py): backfills empty lines_or_bus / lines_ex_bus / gens_bus / loads_bus from dict_action[id].content.set_bus, surfaces voltage_level_id, and tolerates numpy arrays via a new _is_meaningful_dict truthy-check. Fixes pins stacking on max_rho_line when running the Random model.
• build_recommender_inputs propagation: the expert-rule filter result (context["filtered_candidate_actions"]) is now forwarded to the DTO so sampling models actually see the filtered pool — caught a silent bypass where RandomOverflowRecommender ran against the full action dictionary while the filter was running upstream.
• overview_filter_changed interaction-log event now carries a combined_only discriminator.

Fixed

• Settings modal stale-write on config-file switch: changeConfigFilePath now returns the resolved UserConfig, and configRequestFromUserConfig derives the request from it. Regression test in frontend/src/App.configUpload.test.tsx.
• PyPSA-EUR grid-layout rescaling — default is now raw Mercator metres. data/pypsa_eur_fr225_400/grid_layout.json and data/pypsa_eur_fr400/grid_layout.json regenerated.
• Action overview pin localisation for non-disconnection actions (Random / Random Overflow runs): pins are now anchored on the action's voltage level rather than the contingency max_rho_line.
• numpy.str_ comparison crash in _resolve_node_to_name on legacy distribution graphs.

Documentation

• docs/features/save-results.md — model / active_model / compute_overflow_graph field tables + new "Recommender model persistence" section.
• docs/features/interaction-logging.md — model / compute_overflow_graph added to config_loaded / settings_applied; active_model / compute_overflow_graph added to analysis_step2_completed; new "Pluggable recommender model" section.
• docs/backend/README.md — NEW: backend overview (architecture, mixins, data flow, conventions, endpoints, tests).
• docs/backend/recommender_models.md — NEW (relocated): app-side integration + filter chain + step-by-step guide.
• README.md — NEW "Plug Your Own Recommendation Model" section.

Compatibility

• model and compute_overflow_graph default to "expert" and true at every entry point that lacks them (older session dumps, missing form values, third-party callers that didn't update their request shape) — byte-for-byte the same behaviour as 0.7.0.
• Frontend dynamic UI from params_spec() — adding a model requires zero UI code; the dropdown and parameter inputs refresh automatically.
• Step-2 NDJSON contract unchanged — active_model and compute_overflow_graph are additive fields on the existing result event.
• Requires expert_op4grid_recommender>=0.2.2 (for the RecommenderModel ABC, the RecommenderInputs / RecommenderOutput DTOs, the reusable reassessment phase and the idempotent _run_expert_action_filter helper). Older versions raise an ImportError from expert_op4grid_recommender.models.base on backend startup.

Install

pip install "co_study4grid==0.7.5" "expert_op4grid_recommender>=0.2.2"

See CHANGELOG.md § 0.7.5 for the full entry.

Full changelog: marota/Co-Study4Grid@0.7.0...0.7.5

---

Generated by Claude Code

0.7.0 — interactive overflow analysis tab + PyPSA-EUR + design tokens

Major feature release headlined by the interactive overflow analysis tab, the PyPSA-EUR European-wide grid pipeline, and a full design-token migration of the frontend.

Re-tagged 2026-05-05 at commit a6bbe40 to fold in the post-initial-tag pin-overlay fixes (PR #129). The 0.7.0 release artefact now points at the merge that ships the working pin overlay; the original 519ad61 tag is gone.

Highlights

Interactive overflow analysis tab (PRs #116, #122–#127, #128, #129)

The static overflow PDF is replaced by a same-origin HTML viewer produced by upstream alphaDeesp/core/interactive_html.py:

• Layer-toggle sidebar grouped into three sections — Structural Paths (Constrained path, Red-loop), Individual entities properties (Overloads, Low-margin lines, Hubs, Reconnectable, Non-reconnectable, Swapped flow, Production nodes, Consumption nodes) and Flow redispatch values (Positive / Negative / Null).
• Hierarchical ↔ geographic layout toggle backed by a per-study cache.
• Action-pin overlay synced with the Action Overview filters — single-click pins open the same ActionCardPopover, double-click drills into the SLD overlay. Pins land on the actual visual mid-arc of the rendered Bézier edge (getPointAtLength(len/2)), spread on a small circle when colocated, and combined-action pairs render a dashed Bézier connector between the two unitary halves with a + badge on the midpoint glyph.
• Filter-driven dimming, mirroring ActionOverviewDiagram's three-pass pipeline: a unitary action that fails the active filter is rendered with dimmedByFilter opacity 0.35 (instead of being dropped) when a passing combined-action pin references it as a constituent.
• Sidebar filter chips match the React CategoryToggle visual: no blue solid fill on selected, dim the unselected ones, swatch-coloured border on the active chip.
• Auto-installer for the graphviz dot binary on package install (PR #126).
• Six new replay-log event types: overflow_layer_toggled, overflow_select_all_layers, overflow_node_double_clicked, overflow_pin_clicked, overflow_pin_double_clicked, overflow_pins_toggled, overflow_layout_mode_toggled.

PyPSA-EUR European-wide grid (PRs #112, #117)

Full pipeline in scripts/pypsa_eur/ for generating XIIDM grids from PyPSA-EUR data, with calibrated thermal limits, an fr225_400 dataset, and pytest coverage.

Design-token migration (PR #120, three phases)

frontend/src/styles/tokens.{css,ts} is the single source of truth for colour, spacing, typography and radius. Code-quality gate enforces FRONTEND_HEX_LITERAL_MAX = 0 outside the token files.

UI polish

• Tiered warning system + diagram legend (PR #122): structured notice tiers and an in-place legend on the Visualization panel.
• Progressive-disclosure ActionCard (PR #121): severity icons drive a glanceable summary; details expand on demand.
• Voltage-level names toggle (PR #118): 🏷 VL chip with a native <title> tooltip fallback.

Merged PRs since 0.6.5

PR	Title
#112	feat: PyPSA-EUR pipeline consolidation, fr225_400 dataset, full test coverage
#113	Remove unnecessary f-string prefixes from static strings
#114	Surface target max rho on user-selected overloads in pair estimation
#115	docs: Update CLAUDE.md, README.md, and backend/frontend guides for 0.6.5
#116	feat(overflow): interactive HTML viewer + Hierarchical/Geo toggle + N-1 disconnect fix
#117	feat(pypsa-eur): support European-wide grid generation
#118	Add voltage-level names toggle with native tooltip fallback
#119	docs: add development-cycle retrospective (4 phases)
#120	feat(frontend): design-token migration (recommendation #1)
#121	Redesign ActionCard with progressive disclosure and severity icons
#122	Implement tiered warning system and diagram legend (UI critique #4-5)
#123	Fix NoticesPanel popover clipping by rendering via portal
#124	feat(overflow): interactive overflow analysis tab
#125	Document Overflow Analysis tab interactions and ephemeral UI state
#126	Add Graphviz auto-installer and pydot dependency for graph rendering
#127	fix(build): restore quality extras under optional-dependencies
#128	Mid-arc anchoring + Action-Overview-style pin layout (initial pin overlay)
#129	Pin overlay fixes: duplicate const, dimmedByFilter, protectedIds, chip restyle

See CHANGELOG.md § 0.7.0 for the full entry.

Co-Study4Grid 0.6.5

Follow-up release to 0.6.0 consolidating four strands of post-0.6.0 work merged across PRs [#111](https://github.com/marota/Co-Study4Grid/pull/103) → [#111](https://github.com/marota/Co-Study4Grid/pull/111):

⚡ Performance — SVG DOM recycling on N-1 / Action tabs ([PR #108](https://github.com/marota/Co-Study4Grid/pull/108))

Stop re-fetching and re-parsing the full pypowsybl NAD SVG (12–28 MB) on every N-1 / Action tab change. The new /api/n1-diagram-patch and /api/action-variant-diagram-patch endpoints return only the per-branch delta; the frontend clones the already-mounted N-state SVGSVGElement and patches it in place — dashed contingency lines, absolute flow labels, concentric rings for coupling / node-merging / node-splitting actions.

Measured on the French grid benchmark (bare_env_20240828T0100Z, ~10 k branches, contingency ARGIAL71CANTE, warm median of 3):

Endpoint	Cold	Warm	Payload
/api/n1-diagram (full)	3.01 s	2.39 s	27.1 MB
/api/n1-diagram-patch (new)	0.49 s	0.50 s	5.5 MB
Δ	−83.8 %	−79.1 %	20.3 % of full

Graceful full-NAD fallback on every unsupported edge case (session reload, mount race, extraction failure). Docs: [docs/performance/history/svg-dom-recycling.md](https://claude.ai/code/docs/performance/history/svg-dom-recycling.md).

🎯 Action Overview — filtering & unsimulated pins ([PR #105](https://github.com/marota/Co-Study4Grid/pull/105), [#107](https://github.com/marota/Co-Study4Grid/pull/107))

• Severity category toggles (green / orange / red / grey), threshold slider, action-type chip filter and a new unsimulated pin layer for scored-but-not-yet-simulated actions (dimmed, dashed, double-click to simulate).
• Filter state is shared between the overview pins and the sidebar action feed, so both views stay in lock-step. New actionPassesOverviewFilter() predicate + classifyActionType module unify the filtering logic across three UI surfaces.
• Protected constituent pins: when a combined action passes the filter, its unitary parts stay visible (dimmed) even if they individually fail, preserving context.

🧪 Code-quality gate + decomposition sweep ([PR #104](https://github.com/marota/Co-Study4Grid/pull/104), [#106](https://github.com/marota/Co-Study4Grid/pull/106))

New continuous quality reporter (scripts/code_quality_report.py) and CI gate (scripts/check_code_quality.py) guarding LoC ceilings + smell counts. Metrics drive five behaviour-preserving decomposition passes:

Target	Before	After
simulate_manual_action	599 LoC	146
compute_superposition	285 LoC	108
frontend/src/utils/svgUtils.ts	1807 LoC	60 barrel + 8 focused modules
analysis_mixin.py	1116 LoC	509 + 4 modules
diagram_mixin.py	974 LoC	469 + 7 modules

Public API unchanged; @patch compatibility preserved via dependency injection; 230+ new unit tests across the extracted helpers. CI publishes the Markdown metrics to $GITHUB_STEP_SUMMARY.

🧹 Frontend refactor — App.tsx hook extraction ([PR #109](https://github.com/marota/Co-Study4Grid/pull/109))

• New hooks/useN1Fetch.ts (svgPatch fast-path + /api/n1-diagram fallback + contingency-change confirm routing).
• New hooks/useDiagramHighlights.ts (per-tab Flow/Impacts view-mode + applyHighlightsForTab DOM pass).
• Sidebar extracted into components/AppSidebar.tsx, SidebarSummary.tsx, StatusToasts.tsx.
• Unified ActionTypeFilterChips component reused by Manual Selection, Explore Pairs, Action Feed and Action Overview.
• App.tsx: 1575 → ~1150 lines.

📚 Docs reorganisation ([PR #103](https://github.com/marota/Co-Study4Grid/pull/103))

Flat docs/ restructured into docs/{features,performance/{,history/},architecture,proposals,data}/ with per-folder README.md indexes. Three overlapping rendering-LoD proposals consolidated into [docs/proposals/rendering-lod-strategies.md](https://claude.ai/code/docs/proposals/rendering-lod-strategies.md). All cross-references in source, tests, benchmarks, scripts and CI workflows updated.

🐛 Fixes worth calling out

• Dynamic reco_ reconnection actions: reconnection actions generated by expert_op4grid_recommender but absent from the loaded action dictionary (e.g. reco_CAZARL72MARSI from the Explore Pairs tab) now simulate correctly, via the same dynamic-creation path already used for curtail_ / load_shedding_ / pst_.
• Combined disco + coupling halo restored on the disco line in the combined action (PR #108).
• Blank flash + stale-response guard on svgPatch for large grids (PR #108).
• Node-merging classifier: "Ouverture … dans le poste" is now DISCO, not OPEN coupling (PR #105).
• Popover viewport + combined-pin constituent protection (PR #105).

See [CHANGELOG.md](https://claude.ai/code/CHANGELOG.md) for the full, sectioned list.

Full changelog: marota/Co-Study4Grid@0.6.0...0.6.5

# Co-Study4Grid 0.6.0 — 2026-04-20

Released 2026-04-20

Follow-up release to 0.5.0 consolidating the standalone-parity
effort, the Action Overview diagram, inactive-tab SVG perf work,
and the docs reorganisation.

Highlights

• Auto-generated single-file standalone (npm run build:standalone)
  replaces the hand-maintained standalone_interface.html. The React
  source in frontend/src/ is now the single source of truth — no
  manual mirroring required when adding a component, setting, API
  call, or gesture.
• Layer-4 user-observable invariants — runtime Vitest twin
  (userObservableInvariants.test.ts) paired with
  scripts/check_invariants.py, guarding the six classes of
  regression that had previously shipped past layers 1–3.
• Action Overview diagram — map-pin overlay on the N-1 NAD
  showing every prioritised action with severity colouring and
  pan/zoom-aware pin rescaling.
• Perf: display:none on inactive SVG tabs cuts live DOM nodes
  from ~600 k to ~200 k on the French grid.
• Docs reorganised into features/, performance/ (+
  history/), architecture/, proposals/, data/ with a
  consolidated rendering-LoD proposal.

Added

• Auto-generated standalone bundle (PR #101): React + CSS
  inlined into frontend/dist-standalone/standalone.html via
  vite-plugin-singlefile. Canonical distribution artifact.
• Layer-4 parity guard: scripts/check_invariants.py for the
  standalone and frontend/src/utils/userObservableInvariants.test.ts
  for the React side.
• Layer-3 Playwright E2E suite (scripts/parity_e2e/e2e_parity.spec.ts)
  and a gesture-sequence static proxy (scripts/check_gesture_sequence.py).
• Action Overview diagram: severity threshold parameterised by
  monitoringFactor, topology-first pin anchoring, combined-pair
  dashed curves, Overview backdrop dim, auto-switch to Action tab on
  "Display Prioritized". See docs/features/action-overview-diagram.md.
• Detached + tied visualization tabs brought to full Layer-1
  parity (save-only rho arrays, tied viewBox sync,
  detach-in-Overview-mode support).
• /api/simulate-and-variant-diagram streaming endpoint —
  NDJSON {type:"metrics"} then {type:"diagram"} so the sidebar
  updates ahead of the SVG.
• frontend/PARITY_AUDIT.md — working record of the parity
  effort, split out of the root CLAUDE.md.

Changed

• Docs reorganised into docs/{features,performance/{,history},architecture,proposals,data}/
  with per-folder README indexes. Three overlapping rendering-LoD
  proposals merged into docs/proposals/rendering-lod-strategies.md.
  All in-repo references updated across CLAUDE.md files,
  README.md, benchmarks, scripts, tests and source comments.
• expert_backend/CLAUDE.md / frontend/CLAUDE.md refreshed
  with the post-decommission wording (single source of truth in
  frontend/src/).

Performance

• display:none on inactive SVG tabs (PRs #99, #102): the
  off-screen N / N-1 / Action SVGs drop from the browser paint tree,
  cutting live DOM nodes from ~600 k to ~200 k on the French grid.
  See docs/performance/history/svg-tab-unmount.md.

Fixed

• SLD highlight for LS / curtailment / PST targets on the Action
  tab, with a text-search fallback when the metadata index misses an
  equipment ID.
• Overload halo suppression on "Solved — low margin" actions in
  both NAD and SLD.
• Session reload now refreshes the N-1 diagram and preserves
  action-bucket / combined-pair state without re-simulation;
  /api/restore-analysis-context wired in.
• Manual re-simulate refreshes the SLD overlay; modal content
  word-wraps correctly.
• Pin severity uses monitoringFactor − 0.05 instead of a
  hardcoded 0.9 / 1.0 cutoff, fixing the MF = 0.85 misclassification.
• Pin anchor uses the topology target (action's disconnected
  line) rather than max_rho_line.
• Rendering fidelity for Overview, detached tabs, overflow tab,
  and action auto-zoom.
• ESLint errors in userObservableInvariants.test.ts
  (@typescript-eslint/no-explicit-any, no-unused-vars).

Removed

• expert_backend/tests/test_ui_regressions.py — its assertions
  targeted strings in the decommissioned standalone_interface.html;
  equivalent coverage now lives in the four parity scripts
  (scripts/check_*.py) and in the Vitest suite.

Documentation

• New consolidated doc: docs/proposals/rendering-lod-strategies.md.
• New index files: docs/README.md, docs/performance/history/README.md.
• docs/features/action-overview-diagram.md for the Action Overview
  diagram feature.
• Root, frontend, and expert_backend CLAUDE.md files reflect the
  auto-generated standalone workflow and the new docs tree.

Upgrade notes

• No migration required for existing sessions — session.json
  format is unchanged.
• If you were editing standalone_interface.html by hand, stop:
  edit the React source in frontend/src/ and run
  npm run build:standalone to regenerate
  frontend/dist-standalone/standalone.html. The legacy file is
  preserved as standalone_interface_legacy.html for reference.

Full changelog: https://github.com/marota/Co-Study4Grid/blob/0.6.0/CHANGELOG.md
Compare with 0.5.0: marota/Co-Study4Grid@0.5.0...0.6.0

Co-Study4Grid 0.5.0 — Release Notes

Released 2026-04-14

First tagged release under the Co-Study4Grid name (formerly ExpertAssist). This release consolidates the work merged in PRs #57 → #91 and ships a stable, production-ready contingency-analysis assistant for large grids.

---

TL;DR

• ~4× faster manual-action simulation on the full French grid (~10k branches)
• ~1,100× faster overload detection, 47× faster delta computation
• Two-step N-1 workflow (detect → select → resolve) as the primary path
• Full remedial-action catalog: topology, PST tap, renewable curtailment, load shedding — individually, manually, or as pairs
• Detachable visualization tabs for dual-monitor studies
• Session save/reload + replay-ready interaction log
• Rebrand ExpertAssist → Co-Study4Grid, MPL-2.0 banners across all sources

---

Highlights

Performance — Big Grids Now Fly

Measured on the full French grid (~10k branches) with scripts/profile_diagram_perf.py:

Metric	Before	After	Speed-up
care_mask / overload detection	12.17 s	0.01 s	~1,100×
Branch flow extraction	0.82 s	0.06 s	~13×
Flow delta computation	0.47 s	0.01 s	~47×
get_obs() overhead	0.65 s	0.01 s	~65×
Total manual-action simulation	~16.5 s	~4.0 s	~4×
Base diagram rendering	~7.2 s	~3.5 s	~2×
N-1 contingency analysis	~19.8 s	~12.9 s	~1.5×

How: NumPy vectorization of the care_mask loop, flow extraction and delta pipelines; observation caching in the manual-action loop; pre-built SimulationEnvironment and dict_action reused across steps; lxml-based NaN stripping + gzip compression for the 13 MB NAD SVG payload; frontend throttling and zoom-tier level-of-detail. Details in docs/PR_PERF_OPTIMIZATION.md and docs/performance_profiling.md.

Workflow — Two-Step N-1 Analysis

The primary user path is now a two-step flow:

1. /api/run-analysis-step1 — detect overloads under the selected contingency.
2. Operator picks which overloads to resolve.
3. /api/run-analysis-step2 — stream prioritized remedial actions (NDJSON).

The legacy single-shot /api/run-analysis endpoint remains for backward compatibility. AC/DC fallback is transparent — AC load flow first, DC on non-convergence.

Full Remedial-Action Catalog

• Topology: switches, bus reconfiguration, line disconnection / reconnection, close / open coupling.
• Phase-Shifting Transformers (PST) (PR #78): tap start / target columns, re-simulation from the score table, target-tap sync, superposition fallback for PST pairs.
• Renewable curtailment (PR #72) and load shedding (PR #61, #73) using the new set_gen_p / set_load_p power-reduction format — elements stay electrically connected, preserving voltage and reactive support. Configurable MW reduction.
• Combined actions: evaluate pairs via a fast superposition (beta-coefficient) estimation or an exact full simulation through the Computed Pairs / Explore Pairs modal. LS + RC combinations are restricted in the UI.

See docs/curtailment-loadshedding-pst-actions.md and docs/combined-actions.md.

Visualization — Made for Dual Screens

• Detachable tabs (PR #84/#86/#87/#90): pop any of the four visualization tabs — Network N, Contingency N-1, Remedial Action, Overflow Analysis — into a second browser window. Tie/untie, bidirectional controls, per-window pan/zoom preservation, stable-portal DOM move (no unmount/remount cascades). See docs/detachable-viz-tabs.md.
• SLD impacted-asset highlights (PR #63): clone-behind halos for switches and coupling breakers, robust across pan/zoom and N-1 / action state changes.
• Focused sub-diagrams with configurable depth for inspecting specific voltage-level neighborhoods on large grids.
• Zoom-tier level-of-detail (PR #76): labels, nodes and flow arrows are boosted proportional to sqrt(diagramSize / referenceSize) so large grids stay legible at any zoom.
• Auto-zoom on contingency, newly overloaded line, or action target; sticky compact feed summary; overload-click jumps to the N-1 tab (PR #88).

Sessions & Replay

• Save Results / Reload Session (PR #62 family): export the complete analysis state — config, contingency, actions with status tags, combined pairs, overflow PDF, loading ratios — to a timestamped session folder (costudy4grid_session_<contingency>_<timestamp>/). Reload restores the UI exactly, without re-simulating. Documented in docs/save-results.md.
• Replay-ready interaction log (PR #64): every UI interaction is written to interaction_log.json as a self-contained, timestamped event with correlation IDs for async completions — suitable for deterministic browser-automation replay. See docs/interaction-logging.md.
• Persistent user configuration (PR #59) stored outside the repo, with a configurable config-file path.
• Confirmation dialogs (PR #83) guard destructive state resets (switching network while a study is loaded, applying settings on an active study).

Frontend Engineering

• App.tsx — Phase 1 refactor (PR #74): from ~2100 → ~650 lines. App.tsx is now a pure state-orchestration hub; UI extracted into presentational components under components/ and components/modals/. useSettings hook exposes a single SettingsState object passed wholesale to SettingsModal.
• State management — Phase 2 (PR #75): memoized cross-hook wrappers with useCallback, centralized reset logic, React.memo on the three heaviest components (VisualizationPanel, ActionFeed, OverloadPanel). Eliminates unnecessary re-renders.
• Oversized components split (PR #81) into focused subcomponents with dedicated test suites.
• React ErrorBoundary wrapping the app root (PR #82).
• Standalone single-file UI (standalone_interface.html) kept in feature lockstep with the React app — zero-install demos.

---

Notable Fixes

• Second-contingency crash, auto-zoom loss on contingency switch, overload persistence across successive studies (PR #80).
• min_renewable_curtailment_actions missing from saved config (PR #80).
• Auto-zoom double-injection on contingency switch — MemoizedSvgContainer now kept always mounted (PR #81).
• N-1 variant clone made from the clean N state, not the working variant (PR #81).
• Contingency highlight z-ordering: sibling insertion with solid yellow halo, surviving pan/zoom and SLD overlay (PR #71).
• Overload highlight thresholds and loading display values (PR #71).
• Monitoring-factor scaling restored for suggested actions (PR #71).
• Superposition monitoring aligned with simulate_manual_action; overloaded lines force-included in eligible_mask (PR #79).
• PST re-simulation preserves _dict_action structure; additive superposition fallback for PST no-op (PR #78, #79).
• Combined-action estimation filtering: heavily loaded N-state lines are no longer incorrectly filtered out (PR #72).
• Backend regressions on gen_p, observation sequence, legacy keys, and islanding reporting after the manual-action enrichment refactor (PR #72, #73).
• Re-simulate double-click bug on action cards (PR #73).
• SLD rendering regressions: blank screen, missing N-1 highlight, ReferenceError crash in the overlay (PR #76).
• Grid layout functionality restored with regression tests (PR #69).

---

API Changes

New endpoints added in this release:

Analysis

• POST /api/run-analysis-step1 — two-step flow, step 1
• POST /api/run-analysis-step2 — two-step flow, step 2 (streaming NDJSON)
• POST /api/simulate-manual-action
• POST /api/compute-superposition

Diagrams

• POST /api/action-variant-diagram
• POST /api/focused-diagram
• POST /api/action-variant-focused-diagram
• POST /api/n-sld, POST /api/n1-sld, POST /api/action-variant-sld

Sessions & config

• GET /api/user-config, POST /api/user-config
• GET /api/config-file-path, POST /api/c...