Skip to main content

Opportunity Radar — bounded deterministic consumer pilot

Scope: Exactly three reviewed producer–consumer opportunities; report generation is an offline Projects consumer and not an MCP change, dynamic crawler, runtime integration attestation, scoring engine, or authority to publish/execute.

Run​

python3 scripts/build_opportunity_radar.py --as-of 2026-10-18
python3 -m unittest discover -s tests -p 'test_opportunity_radar.py' -v
make opportunity-radar-check

Inputs live in estate/intelligence/opportunity-radar-pilot.json and its pinned estate/intelligence/opportunity-radar/*.json receipts, plus the accepted generated/project-agenda-index.json (29 Agendas at d3b2b88) and Capability Atlas manifest/reuse proposals. The output is artifacts/opportunity-radar/pilot/{report.json,REPORT.md,manifest.json}. Generator date is an explicit Argentine calendar date; outputs contain no execution timestamp. It does not fetch HTTP, query GitHub, read a different repository, execute a source command, or resolve private data during generation.

Pinning provenance and limits​

The receipts are one-time observed inputs, gathered by a reviewer on 2026-10-18 ART / 2026-10-19 UTC. They are not fresh every time this report is rebuilt. Public HTTP receipt bodies are represented by facts and upstream response hashes; Site Factory HTML itself is intentionally not replicated. Media Monitor public health JSON is captured with its health fields. The geography receipt embeds the owner manifest which explicitly says fixture-v1 is synthetic and unsuitable for official analysis. These observations are not authenticated signed attestations and do not replace producer acceptance receipts or runtime verification.

The baseline Atlas snapshot originates from 2026-09-13 and contains 137 capabilities, 28 proposals and only 2 previously classified integrations. Other unmerged Atlas work may contain more integration records; the report intentionally uses committed accepted state rather than merging those drafts silently. Atlas MEDIUM confidence is heuristic, never direct working integration proof. The Site Factory Atlas seam is a potential reuse of publication/CLI conventions with capture-site; public catalogue availability does not prove a functioning request or admin-review pipeline. The Media Monitor health check does not clear owner live-summary gate #141. The synthetic geography fixture is explicitly blocked for real territorial/electoral use.

Qualification states​

  • READY_FOR_BOUNDED_TEST: public landing is reachable, but only a staging-only and nonsensitive producer acceptance test is proposed. Does not authorize a production action.
  • VERIFY_FIRST: a healthy digest snapshot does not prove live summaries or a real consumer handoff. Owner acceptance and one genuine hash-pinned selected evidence use case are prerequisites.
  • HOLD: an owner-declared synthetic fixture cannot be promoted to a scientific, geographic or electoral crosswalk; official source vintages/licenses/coverage/methodology approval are missing.

Cases deliberately have no estate-wide rank, automatic score, or command execution. Missing Agenda pages, index SHA disagreement, Atlas proposal mismatch, mutated pinned receipts, path escapes, and changed synthetic-release disclaimers fail closed. review_due_on is shown for review scheduling, not used to infer that producer truth is false.

Updating a case​

First inspect accepted owner and Projects evidence, then update the reviewed pilot input and, if justified, pin a new producer receipt with its observed-at time, source URL or producer artifact and SHA-256. Do not overwrite history without a reviewer-visible diff. Rerun the generator and checks, inspect REPORT.md, and make a small reviewable PR. Do not add a general discovery/scoring framework until repeated operator sessions prove the three-case consumer inadequate. Do not change any private/public exposure or source authority by inference.