We publish and maintain benchmarks for the claims we make about community intelligence: what the system reads, how accurately it scores it, and whether the evidence can be checked. Every number traces to a published results file, and the full evidence pack is available for technical review.
Want to check our work? Request the benchmark pack - test data, scoring code, raw outputs, and methodology notes.
Turns consultation evidence into reports reviewers can trace, test, and defend.
See the full benchmark
Report citations resolved
Runs without a decision flip
Minority viewpoint preserved
Every issue in every response, with its own sentiment, filed under your reporting topics.
See the full benchmark
Issues found in long written submissions
Issues implied but never named
Issues correctly filed under your reporting topics
Measures what people want, not just how they sound.
See the full benchmark
Frozen test stance accuracy
Tone-divergent feedback read as stance
Meaning-preserving wording changes
Finds the reasons behind community positions, with source evidence attached.
See the full benchmark
Reasons found and correctly filed
Implied reasons residents never name
Quoted arguments rejected by the resident
Find coordinated campaigns without silencing genuine residents who share the same concern.
See the full benchmark
Campaign relationships found
Organic pairs correctly kept separate
Unique-voice count accuracy
How we benchmark
Benchmarks are only useful if you can check them. Ours are built to be checked - and to be rerun when the products change.
Every corpus is written for testing - spanning typos, sarcasm, voice transcripts, low-literacy writing, implied issues, and ten community languages.
Each benchmark states what was tested, what counted as a pass, and which baseline or comparison helps explain the result.
Every number traces to a published results file, and the full evaluation suite re-runs end to end - including a non-live path that re-scores cached outputs.
Test data, scoring code, raw model outputs, charts, and methodology notes are available for technical review by your IT and AI governance teams.
Bring one real, de-identified feedback export to a 30-minute walkthrough and see the evidence trail behind the analysis.
Product notes, practical field guides, and evidence-led thinking for teams working under public scrutiny.
