Discovery & Intelligence

One logical worker: evidence-backed product, market, technology, and user intelligence · Discovery recommends. Hermes coordinates. Ari decides.
DCP READ-ONLY: CONNECTED DEVELOPMENT / UI PREVIEW DCP BATCH 5/6 SOAK: PASS GATE PACK A: PASS BACKUP: UNKNOWN RESTORE: UNKNOWN STATIC_CONFIG TELEMETRY: UNKNOWN COMMAND EXECUTION LOCKED

DISCOVERY INBOX

On-demand Discovery · Continuous Intelligence (data model + UI only in V1)
Discovery recommends. Hermes coordinates. Ari decides.
DEVELOPMENT FIXTURES — not real current research · MOCK STATIC All counts and findings below are loaded fixture data, not live research. No REAL findings exist in V1.
New
4
fixture data only
Reviewed
1
fixture data only
Promoted
1
fixture data only
Conflicting Evidence
1
fixture data only
Filters

Simple attribute filters only — no semantic search, no vector DB. Every filter is a plain GET link; no JavaScript.

Project
Allexample-crm-toolexample-inventoryexample-note-app
Type
AllFACTUSER COMPLAINTTRENDASSUMPTIONOPPORTUNITY
Confidence
AllLOWMEDIUMHIGH
Relevance
AllLOWMEDIUMHIGH
Mode
AllON DEMANDCONTINUOUS
Status
AllNEWREVIEWEDPROMOTEDDISMISSED
Finding type legend
FACT Verifiable claim backed by a named source. Without independent support it stays UNVERIFIED.
USER COMPLAINT A reported user pain from a community/forum/review source — real to the reporter, not yet verified as a widespread problem.
TREND Repeated signal across sources. Never escalated on its own; watch for corroboration.
ASSUMPTION An unvalidated internal belief. Stays ASSUMPTION until tested with real user evidence.
OPPORTUNITY An actionable possibility assessed from evidence (LOW/MEDIUM/HIGH) — an assessment, not a score.
CONFLICTING EVIDENCESources disagree — a contradiction is visible and unresolved. Conflicting findings are never auto-resolved and never escalated until a human reconciles them.

Evidence discipline

Every finding visibly distinguishes FACT / USER COMPLAINT / TREND / ASSUMPTION / OPPORTUNITY and shows source, URL (if present), date, confidence and provenance. UNKNOWN or missing evidence never silently becomes HIGH confidence. Contradictions are shown as CONFLICTING EVIDENCE and are never auto-resolved. Opportunity level (LOW/MEDIUM/HIGH) considers: user pain, evidence strength, strategic relevance, implementation burden, differentiation potential. It remains an assessment, not mathematical truth.

A CONTINUOUS finding is escalated to Hermes only when it is relevant, materially useful, evidence-backed, and not trending noise.

Findings (1 of 7 shown · fixture data)

Counts and findings reflect the loaded MOCK/STATIC fixtures only — they are not production intelligence.
DF-007TRENDCONTINUOUSDISMISSEDMOCK

Short-term spike of AI-summary posts (noise)

Example social posts show a short-term spike of interest in AI-generated summaries; no material use identified.

Project
example-note-app
Confidence
LOW
Relevance
LOW
Opportunity
LOW
Source
Social signal sample (static example) https://social.example.com/sample
Source date
2026-10-02
Captured at
2026-10-05
Recommended next action
None — dismissed as trending noise.
Dismissed: fails the continuous-intelligence escalation rule (not materially useful).
PROMOTE · LOCKED (PREVIEW ONLY) DISMISS · LOCKED (PREVIEW ONLY)

Promotion to Product Planning (conceptual handoff)

Discovery Finding→Hermes Review→Product Architect / Product Critic→

UI semantics only. V1 has no write API: promotion/dismissal is LOCKED preview. A promoted finding is reviewed by Hermes before the Product Architect / Product Critic act on it — discovery never bypasses the Planning Cell and never creates build tasks from weak signals.