docs: add ariadne planning notes

docs/00 INBOX/ai/ariadne-project-goals.md

@@ -0,0 +1,68 @@
+# Ariadne project goals
+
+Ariadne analyzes a local Ren'Py game install.
+It extracts source, builds a source graph, and plans routes through target nodes.
+
+## Privacy constraint
+
+Do not write the target game's title in the monorepo.
+Use `target game` or `Ren'Py game` in committed notes, code, and pull requests.
+Keep title-specific paths and names in ignored local data only.
+
+## Goal order
+
+1. Unlock all Steam achievements.
+2. Maximize Season 3 dialogue coverage.
+3. Extend the same coverage work to Seasons 1 and 2.
+
+## Steam achievements
+
+Achievement unlocks appear in source as reachable calls.
+The first planning target is every achievement grant node.
+
+Ariadne needs to find each grant call and compute route conditions that reach it.
+This is source-graph coverage over sparse semantic targets.
+
+## Season 3 dialogue coverage
+
+Season 3 exposes an in-game percentage for seen dialogue.
+The stat computes Ren'Py dialogue-block coverage from persistent seen data.
+
+For this game, dialogue blocks map mainly to unique `Say` statements.
+Ariadne should target Season 3 `Say` nodes as coverage targets.
+
+The in-game percentage gives a validation signal for Ariadne's coverage model.
+Season 3 also rewards high coverage inside the game.
+
+## Seasons 1 and 2 coverage
+
+Seasons 1 and 2 lack the same in-game percentage display.
+Ariadne can still compute source coverage for their dialogue nodes.
+
+Validation for these seasons will depend on Ariadne's own model.
+Persistent seen data can still help later.
+
+## State model phases
+
+The first phase ignores player state.
+It asks which routes cover all targets from source alone.
+
+A later phase reads existing player state.
+It uses save files or persistent data to remove completed targets.
+
+That phase produces a personalized plan for remaining achievements and dialogue coverage.
+
+## Graph model
+
+The source graph needs typed nodes for labels, jumps, conditions, menus, visual statements, dialogue, and achievement grants.
+Menus and conditions require branch edges and merge edges.
+
+Dialogue and visual nodes support route explanation.
+Achievement and dialogue nodes become coverage targets.
+
+## Near-term implementation direction
+
+Keep extraction and decompilation generic.
+Parse source into typed graph artifacts.
+Add achievement grant nodes before broad route planning.
+Add Season 3 dialogue nodes as the next coverage target set.