Digital Policy Alert

Evaluate whether a query lies inside the DPA corpus WITHOUT executing it. Returns the same deterministic `coverage` block that search/count responses carry — per-dimension verdicts on DPA's GRADED envelope (focus-set jurisdictions, priority vs secondary policy areas, bindingness, litigation stage), possibly-engaged exclusions, and data currency — computed from configuration alone. NO API call is made and no records are returned. USE THIS TOOL: - BEFORE a search, when the question might fall outside DPA's scope (pre-2020 dates, non-focus countries, public-sector regulation, standalone strategy documents, lawsuit filings, secondary policy areas). - AFTER a zero-result search, to diagnose whether the zero means genuine absence (`absence_within_scope`) or a coverage boundary. - When orchestrating across data sources, to decide whether DPA is the right corpus before spending a data call. Accepts the same typed filters as `dpa_search_events`. Filters without a dedicated coverage dimension (economic_activities, government_branch, implementation_level, threads) are accepted without error but do not change dimension verdicts. Optionally pass `question` (the user's natural-language question): keyword signals from it feed `exclusions_possibly_engaged`, but verdicts are computed ONLY from the explicit typed arguments. The full corpus envelope (time floor, focus set, wild-card triggers, priority areas, verification regime) is a document of its own: call read_resource with uri=dpa://reference/corpus-manifest first.

dpa_check_scope

BEFORE calling this tool, read each guide whose question shape matches, the first when none does: - an overview, a count or a list, of events or of policies → call read_resource with uri=dpa://guide/query-intent - announced, in force, published, active, draft or a year → call read_resource with uri=dpa://guide/date-fields - a policy area or topic, AI, or a category such as 'data rules' → call read_resource with uri=dpa://guide/policy-areas - a zero result, or an absence to confirm → call read_resource with uri=dpa://reference/database-scope - how complete coverage is, or why a topic looks thin → call read_resource with uri=dpa://guide/reporting-heuristic - soft law, strategies, lawsuits or content moderation → call read_resource with uri=dpa://guide/reduced-tracking-rules - combining DPA with another corpus → call read_resource with uri=dpa://reference/corpus-manifest Count and aggregate digital policy interventions or events by 1–3 dimensions. USE THIS FIRST for overviews and trends: it pre-aggregates server-side in 1–3s. Aggregates ONLY — no titles, descriptions or records; for those, follow up with `dpa_search_events`. NEVER sum group rows into a topline: event-level and many-to-many dimensions (event_type, event_date_year, implementer, policy_area, ...) place one intervention in several rows. Use the reported "Total (distinct ...)" value. The response carries its counting unit and a chartable `{labels, datasets: [{label, values}]}` JSON block; pass it to a charting tool, if the client has one, without re-keying or recomputing the numbers. Default event window: 2020-01-01 → today, like the DPA website; widen it with event_period_start / event_period_end.

dpa_count_data

Discover DPA reports, blogs, digests and AI Rules without reading prose. Resolve a named or described publication to exact ``report_id`` values. If candidates are ambiguous, present them rather than choosing silently. Then pass selected IDs to ``dpa_search_analyses``. Use event/count tools for live structured policy facts and document tools for legal text.

dpa_find_analyses

Fetch complete details for one event or an ordered batch of up to 10 events. Cite the record through the workbench citation system using its single linked `event_id`; keep the semantic identifier visible once. Use this tool to get full information for events identified from search results. A batch uses one upstream request, preserves caller order, and reports missing IDs without discarding successful records. Full-text records consume the monthly DPA record allowance; already-counted records are not charged twice in the same month.

dpa_get_event

Turn a natural-language question into a STRUCTURED QUERY PLAN, without executing it. Use this BEFORE any data tool when the user's question is vague — i.e. when more than one of the DPA's thirteen vectors of ambiguity is unresolved (intervention vs event, implementation level, policy area boundaries, instrument granularity, event type binding/non-binding/enforcement, lifecycle status, government branch, economic activity, date semantics, wild card, cross-area, n.e.s., priority). The plan is JSON with keys: summary, primary_tool, suggested_filters, unit_of_analysis, ambiguities_resolved (list), ambiguities_outstanding (list), scope_note_preview (object), next_step (string), references (list). It tells the user which tool would run, with which filters, at which unit of analysis, and what ambiguities were resolved vs. what remains open. The user confirms or refines the plan; only THEN does the LLM call the data tool. Important: this tool does NOT call the DPA API. It does NOT return data. It returns a plan. Calling it before a data tool reduces silent default-application and the resulting "false zero" / "wrong scope" outcomes.

interpret_query

List the full catalogue of readable DPA documents — every static resource and URI template on this server, with its name and description. Use it to find the reference tables behind the filter values (jurisdictions, policy areas, event types, threads, statuses and the rest), the DPA Activity Tracking Handbook, and the per-jurisdiction and per-activity lookup templates. For building a DPA query you do not need this list: `read_resource`'s own description already names the documents that matter and the question shape that selects each one. Pass any URI listed here to `read_resource` to fetch it.

list_resources

Return the DPA economic-activity table with optional fuzzy NL→activity mapping. The DPA's economic-activity taxonomy is its own working classification because no international standard exists at this granularity. "Big Tech" is not a category; it splits across platform intermediaries, search service providers, software providers, and others. "AI" most often maps to ML and AI development, but AI rules frequently apply cross-cuttingly. Use this tool to: - Browse the full activity table (no `query`). - Resolve a natural-language term to candidate activities (`query="big tech"` → ranked matches). For activity-specific queries, run this tool first and let the user confirm the mapping before calling `dpa_search_events` / `dpa_count_data` with the activity filter. For cross-cutting policies, default to no economic-activity filter at all. Returns JSON with `activities` (full table) and optional `matches` (when `query` was given, ranked by similarity score).

lookup_economic_activities

Return the DPA policy-area tree (areas → instruments) with optional fuzzy mapping. The DPA's natural-language category terms rarely map cleanly onto its taxonomy. "Data rules" can plausibly cover data protection regulation, cross-border data transfer, data localisation, cybersecurity, or DPA governance. "Content rules" splits across content moderation, intellectual property, and consumer protection. "AI rules" splits across design and testing standards, data governance, and IP. Use this tool to: - Browse the full policy-area-and-instrument tree (no `query`, no `policy_area`). - Resolve a natural-language term to candidate areas / instruments (`query="data privacy"` → ranked matches). - Drill into a single area's sub-instruments (`policy_area="data governance"` → that area's instruments only). Returns JSON with `tree` (areas with nested instruments), optional `matches` (when `query` was given), and `priority_areas` for orientation. `tree` is sorted: priority areas first. The DPA priority policy areas (close-to-comprehensive coverage) are data governance, content moderation, competition, and taxation. Other areas (intellectual property, labour law, consumer protection, etc.) receive non-exhaustive secondary coverage: call read_resource with uri=dpa://guide/reporting-heuristic first.

lookup_policy_areas

Read one DPA document by URI, in a single call. Call this tool BEFORE your first `dpa_search_events`, `dpa_search_interventions`, `dpa_count_data` or `dpa_semantic_search` call, with each document whose question shape matches the user's, the first when none does — each line below is a shape, then the URI to pass as `uri`, then what that document settles: - an overview, a count or a list, of events or of policies → dpa://guide/query-intent — event vs intervention as the unit, counts first for overviews, interpret_query when 2+ ambiguities are open - announced, in force, published, active, draft or a year → dpa://guide/date-fields — announcement vs implementation vs publication dates, and the lifecycle-status aliases - a policy area or topic, AI, or a category such as 'data rules' → dpa://guide/policy-areas — policy area vs instrument, AI through the 'Regulating Artificial Intelligence' thread, priority vs secondary areas - a zero result, or an absence to confirm → dpa://reference/database-scope — coverage starts 1 January 2020, the focus-country set, and the false-zero query patterns - how complete coverage is, or why a topic looks thin → dpa://guide/reporting-heuristic — the six priority dimensions and the five wild cards that override normal priority - soft law, strategies, lawsuits or content moderation → dpa://guide/reduced-tracking-rules — outlines only when tied to a specific law, civil lawsuits only from the ruling, platforms but not publishers - combining DPA with another corpus → dpa://reference/corpus-manifest — the JSON coverage envelope carrying the stable dpa.* rule ids that coverage blocks cite A URI carrying a `#section` fragment reads the whole document, so any URI a response cites can be passed through verbatim. Reading the matching document before the first query is cheaper than a misinterpreted result. `list_resources` has the full catalogue.

read_resource

Retrieve grounded passages from one or more published DPA analyses. This returns dated publication findings, not live event totals or legal-text documents. Resolve named publications first, preserve selected IDs, and cite the returned DPA publication URLs and dates. Unknown IDs fail closed.

dpa_search_analyses