Riftbound Knowledgebase

Official-first local knowledgebase for Riftbound rules, strategy analysis, competitive records, and VOD reviews.

Policy

• canon/ is the only place that answers rules questions.
• analysis/ stores attributed strategy guidance, articles, and VOD reviews.
• data/ stores normalized records, indexes, taxonomies, and local card snapshots.
• assets/ stores imported official and supporting visual assets.
• No derived source is promoted into canon/ without explicit verification against local official documents.

Source Inputs

• Raw-source vault: /Users/eric/claude-workspace/riftbound
• Derived site repo: /Users/eric/hextech-analytics
• Duplicate repo intentionally ignored: /Users/eric/Documents/hextech-analytics-current

Commands

scripts/import/import_official_sources
scripts/import/import_hextech_docs
scripts/import/import_competitive_data
scripts/import/import_local_cards
scripts/import/import_legacy_knowledge
scripts/import/import_vod https://youtu.be/m2IjrsUzXgU
scripts/query/kb ask "What happens when time is called?"
scripts/query/kb card "Ava Achiever"
scripts/query/kb meta --legend Draven
scripts/query/kb prep --legend Draven --opponent "Kai'Sa"
scripts/query/kb prep --legend Draven --opponent "Kai'Sa" --dry-run
scripts/query/kb report quality
scripts/query/kb publish --target github prep_brief.draven-vs-kaisa
scripts/ops/kb_ops daily
scripts/ops/kb_ops meta-update --date 2026-03-27
scripts/ops/kb_ops vod-review m2IjrsUzXgU
scripts/ops/kb_ops expand-prep --limit 5
scripts/build/rebuild_indexes
scripts/build/rebuild_graph_indexes
scripts/validate/validate_repo
scripts/validate/run_regression_tests

Layout

• canon/rules/
• canon/errata/
• canon/tournament/
• analysis/videos/
• analysis/articles/
• analysis/guides/
• analysis/creators/
• analysis/community/
• analysis/intel/
• analysis/reference/
• analysis/matchups/
• analysis/archetypes/
• data/events/
• data/decks/
• data/cards/
• data/sets/
• data/rulings/
• data/tournaments/
• data/prices/
• data/keywords/
• data/meta-tiers/
• data/schedule/
• data/players/
• data/meta/
• data/taxonomy/
• data/indexes/
• data/ops/
• assets/official/
• scripts/import/
• scripts/query/
• scripts/ops/
• scripts/build/
• scripts/validate/

Query Model

• scripts/query/kb is the main local interface.
• kb ask returns grounded local hits split by canon, analysis, and data.
• kb rule searches official rules first and keeps derived conflicts separate.
• kb card joins the full card corpus with local rulings and mentions.
• kb prep writes machine-generated local prep artifacts under analysis/ and data/indexes/prep_briefs/, and --dry-run prints the full brief without mutating the repo.
• kb report quality surfaces draft objects, derived-unverified buckets, and conflict topics for review.
• kb publish writes downstream outbox payloads for GitHub, Notion, or Linear without changing the repo's source-of-truth policy.

Trust Policy

• Local trust overrides live in data/taxonomy/trust_policy.json.
• Promoted summaries can move from derived_unverified to derived_verified only through this local policy layer.
• The first verified guide pack is policy-driven, not file-edited, and currently covers fundamentals plus a small high-value Draven/Kai'Sa/meta set.
• trust_policy.json also carries a verification_manifest so each promoted guide records why it was eligible for derived_verified.
• Verified guides now use a strict-proof, split-claims model: only structured summary claims may drive retrieval snippets, while weaker topics stay visible as unsupported or attributed context.
• Noisy legacy surfaces such as bookmark intel, archival meta-tier snapshots, and unresolved player records are tagged and downranked rather than deleted.

Agent Ops

• scripts/ops/kb_ops daily refreshes generated local artifacts and emits an ops summary.
• scripts/ops/kb_ops meta-update writes dated meta snapshots to data/meta/ and analysis/meta/.
• scripts/ops/kb_ops vod-review rebuilds local VOD review artifacts from stored captions without needing YouTube access again.
• scripts/ops/kb_ops expand-prep generates a small batch of high-value matchup prep briefs from the current local meta snapshot.

Verification

• scripts/validate/validate_repo checks required artifacts, graph/index integrity, conflict detection, duplicate-repo exclusion, and prep-brief coverage.
• scripts/validate/run_regression_tests runs the CLI regression suite in tests/.

Current Gaps

• Full local card canon is now imported from the legacy local knowledge.db, but it still needs provenance/version refresh if you want a stricter official-source chain.
• The legacy DB includes rich derived datasets, but some fields are curator-generated or scraped summaries rather than direct official exports.
• rb_competitive_players has at least some suspect field mapping, so those player records are preserved locally but marked as less trustworthy than cards, sets, rulings, or event bundles.
• YouTube intake currently depends on yt-dlp plus browser cookies.
• Taxonomy is intentionally seeded, not complete.
• kb publish currently writes connector-ready outbox payloads locally; actual connector-side posting still happens through Codex tools rather than from the script itself.