Informivore: when sources disagree
A research prototype for bringing market observations and prediction-market signals into one place, then asking which differences deserve a closer look.
I want a signal system to help allocate attention. An alert should make it easier to inspect an observation, its source, and its context—not simply add another stream of confident notifications.
Status · September 2026: local prototype. The implementation includes polling, event storage, rule-based detectors, and state export. This page does not establish live uptime or predictive value.
- CollectPoll sources and normalize observations into events.
- CompareApply configured rules to recent observations.
- InspectStore findings and export state for an interface.
A detector you can inspect
The VIX/equity test fixture supplies the following synthetic observations. These are software test inputs, not a report of a trading day.
| Input | Level | Change |
|---|---|---|
| VIX | 24.0 | +9.0% |
| SPY | 530 | +1.6% |
With the default thresholds, the detector checks that the absolute VIX move reaches 5% and the equity move reaches 1%. Both pass. Because both move in the same direction, it emits a vix_equity_divergence alert. The severity is medium; the rule’s high-severity cutoff is a 12% VIX move.
That label is the output of a rule. The default −0.75 correlation used in its description is an assumed baseline, not an estimate calculated from these two observations. The alert is a prompt to investigate, not evidence of mispricing or a forecast.
Download the annotated synthetic fixture
The questions after the alert
Do the change percentages cover the same interval? Are the observations fresh? Are the underlying instruments and market hours comparable? Has the relationship changed? An attention tool needs to expose these questions before it earns a place in a daily routine.
The broader implementation also contains rules for bonds, currencies, commodities, crypto, and prediction-market observations. Adding rules is easier than demonstrating that their findings are useful. I would judge the next stage by reviewed alerts: how many led to a meaningful investigation, how many repeated something already known, and how often missing or stale data misled the interface.
From a dashboard to an ambient display
The project includes a local dashboard and experiments in delivering information to e-ink and pixel displays. A display that stays visible has a different job from a terminal: a compact summary, a clear age, and a reason to look more closely.
The dashboard’s adapter reads structured state and distinguishes absent or stale source files. That helps describe the pipeline’s condition; it does not verify the truth of a source observation. The local dashboard remains separate from this public site.
Related: access and trust boundaries, information that can stand on its own, and the project map.