In the admin panel
Ingest sources, inspect what was extracted, explore the graph, and ask questions.
The Clarity area of the admin panel is where your team turns source material into structured research, inspects how it is used, and evaluates the customer-facing output.
Ingest
Bring text, markdown, or article feeds into the graph from Data → Feeds
Explore
Navigate entities, their neighborhoods, timelines, and the links between them
Analyze
Ask Clarity questions and see the evidence behind each answer
Evaluate
Run production cards and analyst surfaces from Settings → Clarity → Evaluation
Maintain evidence
Search and selectively ingest historical Perigon evidence from Settings → Clarity → Evidence backfill
Ingest sources
Clarity can ingest text and markdown files, and article feeds. Each source runs through the ingestion pipeline, and the entities, claims, and relationships extracted from it are committed to the graph automatically. Quality issues are recorded as validation flags rather than blocked by a human approval step.
Sources can be scoped to your organization or, for Motif-managed content, shared globally.
Explore
The explorer is the day-to-day working surface for graph-backed analysis. From it your team can:
Search and inspect entities and their neighborhoods.
Trace causal chains: what drove an event, and what it drives in turn.
Compare the graph's state over time.
Inspect the assets an entity is linked to, including the recorded reason each crypto token was linked (for example "wrapped", "staked", or "underlying asset").
The explorer's starting view (minimum confidence, traversal depth, and edges per node) is set per organization on the Settings → Clarity page.
Analyze
The Analyze tab is a chat sidebar backed by Clarity. It answers only from ingested evidence, with inline citations, and degrades gracefully when coverage is incomplete:
Full coverage: answers directly from the graph, with citations.
Partial coverage: when only one side of a cause/effect question is covered, it describes what it has and names the gap instead of inventing a link.
Adjacent coverage: when nothing matches directly, it surfaces related topics and shows a banner indicating the answer is not a direct hit.
No coverage: it declines and suggests rephrasing.
Partial and adjacent answers are expected. They show where the graph has gaps. If a question consistently returns partial coverage where you expected a full answer, ingest more material connecting the two sides instead of tuning the chat.
The chat's own filters (minimum confidence and traversal depth) gate the evidence the agent sees. Raising confidence or lowering depth shrinks the available context, so the answer narrows accordingly.
Evaluate production output
Settings → Clarity → Evaluation is the operational test surface for Clarity. Its default Portfolio card mode runs the same layered generator used by the portfolio app against the live Clarity graph and a selected client account. If no account is selected, it uses the demo portfolio. The run force-refreshes the card, captures the exact rendered global context, portfolio impact, personal implication, and options, then grades that output with the standing Clarity judge.
The result includes the section-by-section output, citations, retrieval and generation trace, model calls, token use, cost, and the quality report. This makes it possible to verify both the answer and how it was produced before a retrieval, prompt, or card change is released.
Other modes cover the surrounding layers:
Analyst surfaces test the live stock read, individualized position analysis, portfolio analyst, and investment-advisor outputs.
Synthetic pipeline runs a fixed news corpus through ingestion and the real analyst so retrieval and synthesis changes can be compared repeatably.
Extraction, temporal, and live-window modes isolate evidence capture, cutoff behavior, and changing-news quality.
Runs are saved, so the team can reopen the full trace and compare the latest result with earlier baselines.
Backfill historical evidence
Settings → Clarity → Evidence backfill is a Motif-only maintenance surface. An operator chooses a start date, searches Perigon using the saved ingestion policy, reviews the returned articles, and selects exactly which ones to send through the normal quality gate and evidence pipeline. Searching does not ingest anything by itself.
This tool is for deliberate historical catch-up. Normal source activity remains under Feeds, while evaluation and maintenance live with Clarity's settings.
Review client analyses
Clarity drafts an individualized analysis for every active client each day — a portfolio-level card and a separate analysis for each holding. The Review tab on a client's detail page lists those drafts so an advisor can vet them before they carry your firm's endorsement.
Each draft shows its status — pending review or approved — alongside the AI-written text in editable fields.
Edit to amend. Adjust the wording, then Approve. Your edits become what the client sees and are preserved when the next day's draft is generated.
Approve as-is to clear the pending marker without changes; the next day's draft returns for review so the client always sees the latest analysis.
Revert to AI draft discards an amendment and drops back to the generated text.
Clients see drafts immediately, marked pending advisor review, so nothing is withheld while it waits. Each review action is recorded in the audit log.
Asset domains
The Settings → Clarity page exposes a Graph domains setting, Equities and Crypto tokens, controlling which asset domains Clarity may use. This is enforced at retrieval time: turning a domain off removes its nodes and links from cards, chat answers, and agent context for that organization. At least one domain must stay enabled. See Data management for the underlying asset catalogs.
Where Clarity fits
Clarity is the evidence system behind the rest of Motif. Every product built on it (insights, proposals, and monitoring) uses it as its research layer, Foresight uses it as a scenario substrate, and your team can always inspect exactly which sources and relationships support a conclusion.
Last updated