Give your agent
the evidence that matters.
Reific finds the relevant changes in your time-series data, then passes the readings, context and sources to your agent.
Follow the evidenceWhat the agent receives
1,008 readings.
Three that answer the question.
The highest three-reading total in this week of building data, together with the context needed to use it.
2016-01-16 18:501,080 WhOriginal CSV row 733
Open source dataEnough context to answer.
Three observations, their timestamps, the source rows, and the calculation. The original history stays available for inspection.
Enough context to stop.
This is the maximum in the selected week. It does not establish a fault or identify which device caused it.
16,838tokens in the two-column source
529tokens in the complete evidence packet
Measured input payloads. No model calls; total agent token burn and resolution are unmeasured.
The landscape illustrates finding relevant evidence. The numbers above come from the attributed public dataset.
Benchmark results
Better resolution rate
More questions answered successfully.
Less token burn
Fewer tokens across the agent run.
What do those percentages mean?
This worked example interprets “50% better” as a relative increase. These starting values illustrate the arithmetic; they are not the benchmark’s measured counts.
Resolution 40% → 60%Baseline × 1.50. A 40% baseline becomes 60%, an increase of 20 percentage points.
Tokens per task 20,000 → 12,000Baseline × 0.60. Count all model calls, outputs, retries and tool-result context.
Inspect the work.
Two studies on public data. Open the source, follow the calculation, and see exactly what goes into the context.
How we count tokens