Clarity, with a point of view.
Luci names the mechanism, makes the framing visible, and connects political choices to the people who live with them.
What they said and did
Attributed words, specific proposals, and recorded votes. Dates and original sources sit underneath the story.
What Luci sees
Direct, compassionate interpretation. No manufactured alarm, false balance, or demand for perfect information before offering a useful reading.
Why this matters
A clear account of who is affected, what changes, and what remains unresolved. The goal is understanding.
Three questions, not one score.
Policy direction concerns what someone proposes. Proposal specificity concerns how much of the mechanism they explain. Rhetorical intensity concerns how strongly the language expresses urgency or emotion. None alone measures truth, sincerity, passion, or a person's entire ideology. Issue-specific anchors are more useful than an unexplained left/right number.
Google's Natural Language documentation distinguishes sentiment score from magnitude; document magnitude is not length-normalized. GDELT's GCAM describes text through multiple content-analysis dictionaries. These are useful references, not interchangeable measures of a speaker's political position. Article-level news tone is not automatically a guest's tone.
Google Natural Language concepts ↗
What LucidCivics currently does.
The map uses official membership snapshots retrieved November 11, 2026. The House feed is dated November 3. A selected official roll call in each chamber demonstrates the voting layer. It is not a complete vote history. The incoming House and Senate views combine dated reported election results, continuing terms and explicit pending transitions. These are not certified or sworn membership records. Current committee and subcommittee records are sourced from official chamber directories; caucus coverage is selected and dated. Future committee assignments are not inherited. Portraits are attributed in each profile, with initials if an external image is unavailable.
The Voices view starts with Pete Buttigieg and Leigh McGowan (PoliticsGirl). Profile grouping is editorial organization, not a measured influence network or a claim of coordination. The Questions view adds a small sourced collection: eight legislators, Pete Buttigieg and seven official source pages, dated September 2024 through June 2026. The original eight-person, four-source collection remains available through the graph’s evidence selector. Luci-style interpretations and manually coded features describe those dated proposals, not each person’s complete or current position. The separate signal-card demonstration still uses explicitly fictional examples.
New-evidence sparks mark unread entries in a manually curated update feed. Source publication dates and collection dates are shown separately. The feed distinguishes an added proposal in selected sources, repetition, first collected evidence and analysis pending. A new entry is not automatically a new position. Each comparison retains its earlier source and explains its effect on the graph. Read-state and the spark preference are stored only in this browser. No background monitoring, email or push delivery runs. Collection status lists the checked channels, dates and gaps—including PoliticsGirl’s uncollected housing statements and the untranscribed Psaki interview.
Group the reasoning, not the party.
Select a question to compare proposed solutions and stated rationale. The first algorithm uses 65% overlap in proposed mechanisms and 35% overlap in reasoning, measured with Jaccard set similarity. Average-linkage clustering merges groups at a similarity of 0.48 or higher. Party is never an input. Question-specific filters determine which features are relevant.
These are transparent prototype choices, not a trained or validated political scale. Graphic distances are schematic; group membership is computed. Multiple proposals remain in a person’s bundle. Missing evidence stays unassessed. A joint statement counts as one source even if it contains several speakers. The current collection does not establish current bill status or who first introduced a talking point.
Inspect the collection and feature coding ↗
Follow the evidence across views.
Seat profiles show collected issue bundles and link directly into the question graph. Side-by-side comparisons use those same bundles to expose shared proposed actions, shared reasoning and ideas appearing in only one selected record. An idea absent from a selected record is not evidence of opposition. A joint release counts once in a comparison’s source count.
Useful interpretations can be corrected.
An interpretation should identify its evidence, meaningful uncertainty, and what would change the reading. Earliest observed in our collection is different from first-ever. A repeated phrase can show a pattern without proving coordination. An official statement is evidence of what was said, not automatic proof that it is true.
Luci's editorial lens is human dignity. That value is stated openly. Readers should be able to inspect, disagree with, and understand the reasoning. The method is adapted from the archived Luci voice and assessment prompts; numerical scoring needs calibration before being presented as a validated measurement.