Product
The daily brief
One brief per market, per day. It leads with the standing, states its own limits, and never renders a section it cannot support. The figures below are sanitized for illustration; the structure is real.
Example brief panels
Standing
Status: inconclusive. Flags: thin, concentrated, no calibration record. Confirmed pool 000, decided pool 00.
Momentum
Leader share flat for 0 days, then moved +0.0 over 2 days after the event dated below. Reported as association with the event window, never as cause.
Head to head
Impact-weighted favorability with the five percent cap applied.
Issues that matter
Topics ranked by impact-weighted volume, with mean stance and the top supportive and opposed candidate per topic.
Events timeline
Public events on the market's calendar, overlaid on the 48-hour move that followed each. Association only.
Integrity screen
Standing shown unfiltered and with flagged accounts removed. Flagged accounts 0, flagged impact share 0.0.
Methodology, classifier accuracy record
Cross-model AI agreement, or human-validated agreement, with the reviewed count and date. Labeled distinctly by which produced it.
Every section states its window, platforms, and pool sizes. A section that lacks the pool to speak is marked, not filled.
Operating modes
Modes are reads over the same measured pool and calibration record. Phase 1 is live today. Phase 2 and Phase 3 are roadmap and are not presented as current capability.
Phase 1, live
- standing - the daily headline, three-way support, bootstrap interval, event-annotated timeline, and anchor divergence.
- h2h - head to head over any two candidates, with top supportive and opposed topics.
- event - the political-event registry that annotates the timeline. Association only.
- inauthentic - the per-market coordination screen, flag not remove.
- salience - the topic landscape, ranked by impact-weighted volume.
- casestudy - the client-shareable export, gated on an attested review before the clean version renders.
Phase 2, roadmap
- snr - signal versus noise on a topic frame, by signature not volume.
- breakthrough - the same engine at a higher confidence floor for resource calls.
- discipline - the breakthrough signatures on their own, so a call can be inspected.
- mobilize - posting frequency and engagement by confirmed account, conditional on stance.
Phase 3, roadmap
- crisis - a continuous rolling-window monitor for a hostile frame before it is loud. Requires continuous ingestion.
- transfer - cross-market learning, gated until the calibration record is deep enough to trust.
Roadmap means built when the record and the architecture support it, not sold as present.
The intelligence modes
Five surfaces, each reporting its own pool size and abstaining where the pool cannot carry the question.
Standing. Ranked support across the registered field, impact-weighted with the five percent per-account cap, published with the decided pool size and a stability band on every reading.
Issues. What the conversation is actually about, by share of tagged posts, with per-candidate approval where the pool supports a figure and an explicit abstention where it does not.
Movement. Day-by-day marks with the pool behind each one, so a shift in the line can be separated from a shift in who was talking.
Integrity. Coordination, duplicate-text and burst screening run before weighting, with the screened and unscreened readings shown together.
Publisher. What the campaigns themselves publish and what it earns — a separate data path from the resident pool, never mixed into it.
The record a deployment produces
Where a race certifies, the prediction is locked before the vote — written once, hashed, and stamped into the Bitcoin blockchain — and scored against the certified result afterwards. Scored and pending pairs are published together, and a missed call stays on the record.
A market that does not meet the pre-registered readiness bar does not get a prediction at all. In July 2026 a market prepared for a lock was refused by its own gate: the pool was too concentrated and was shrinking, and the certification recorded that in writing rather than publishing a number the pool could not support.
See the calibration record for every pair, and verify the record for how to check it without trusting us.
Talk to us
Deployments are scoped per race, and a market is stood up before it is sold: we would rather show you a live pool than a slide about one. Write to rrodgers@tsgco.com.
Deployment model
- Four to seven day standup. A market goes from configuration to a first standing dome in under a week.
- Standing dome. The confirmed-account pool persists between runs and grows as new residents are confirmed.
- Twice-daily compounding. A scheduled heartbeat harvests, confirms, and re-measures on a fixed cadence, so the record compounds rather than resetting.