Graphs & Data
Every sensor you connect — a feed, a channel, a social search, a repository — feeds the same graphs, at /client/graphs. Nothing here is aggregated on the server; every chart is computed in your browser from the raw scores, so switching a window or a dimension costs no round-trip.
Retention
How far back a sensor's graphs reach depends on your plan: 14 days on Volcanic (free), 45 days on Tectonic and Seismologist. On-prem and Integrator arrangements can extend this further — see Pricing.
Picking a sensor and a window
The picker at the top of the page lists every sensor you've connected — RSS feeds, chat channels, social feeds, and GitHub repositories together. Choose a window (All / 24h / 7d / 30d) to scope every chart on the page to that range; nothing about the underlying data changes, only what's currently drawn.
!!!SCREEN SHOT: The Graphs page with a sensor selected, showing the stat row (observations, dimensions, shunts, last scored) and the signal-over-time chart populated.
Each graph, explained
- Signal over time — every shunt plus your five strongest base dimensions, plotted across the window, with a dashed line marking the detection gate (50%). This is the chart to watch for a slow trend — tension or hedging creeping upward over weeks reads here long before it would show up anywhere else.
- Shunt detections — mean confidence per shunt across the window, as a bar chart. Hovering a bar shows its peak confidence and how often it actually crossed the trip threshold, not just how high it got on average.
- Distribution — a histogram of where one dimension's scores land, for whichever dimension you pick from the selector. Useful for telling "consistently moderate" apart from "usually calm with occasional spikes" — two very different situations that can share an average.
- Spread — quartiles and outliers per dimension, side by side. This is the chart for "how steady is this reading," independent of where its center sits.
For GitHub sensors specifically, a verdicts panel sits above the shunt chart: how many events tripped a shunt, how many were actually acted on, and how many were only logged as would have (the Envoy's telemetry-only forecast). Polling sensors don't carry this — only repository events do.
!!!SCREEN SHOT: The shunt-detections chart with the GitHub verdicts panel above it, showing tripped/acted-on/would-have counts.
When to look closer
A single elevated reading is rarely the point — Foreshock is built to catch the slope, not just the level. Worth a closer look:
- A base dimension (tension, anger, fear) trending upward across several windows, not just one spike.
- A shunt's mean confidence rising even if it hasn't crossed the trip threshold yet — it's getting closer.
- A joy or gratitude dimension trending down on its own, with nothing else obviously rising to replace it.
- A partisan-framing dimension drifting in one direction on a source you expected to be steady.
Exporting
Download CSV exports exactly what's currently windowed — the same rows the charts are drawn from, nothing aggregated away. More export formats are planned; CSV is available today on every plan.
Where to go next
Every feature page — Manual Spectrometer, RSS Feed Garden, Chat Room Safety & Vibe, and the GitHub Interactions Envoy — writes into these same graphs. If you're reading the raw numbers instead of the charts, the API reference documents exactly what each field means.