Publishing and reading are separate
Scheduled jobs publish artifacts for the reader pages. The tradeoff is a snapshot with a source date, rather than a new analysis on demand.
Follow the data path →An engineering project by Eric Caskey
The engineering behind a market research system.
Conviction Read publishes the output of a market research engine I built. Explore how the pipeline runs, inspect its dated record, and see what the measurements can and cannot establish.
The publishing pipeline
From model output to a record you can inspect
- Model outputs
The engine records scores and dated verdicts.
- Scheduled publication
Exporters prepare the public data artifacts.
- Reader pages
The site displays those artifacts and their source dates.
Reader pages consume scheduled artifacts. Opening a page does not generate a new research report.
Read the design and its tradeoffs →Published outputs
The system at work
The board and record below contain experimental model output. Research and education, not investment advice.
On the record: 9,095 dated verdicts across 70 verdict dates · latest · saved snapshot; live totals not confirmed · see the receipts
Experimental Saved read · live board not confirmed · as of
Design decisions
What the interface needs to tell you
A fresh export can contain old inputs
Source dates and publication times appear separately. An old model output keeps its age when the exporter runs again.
See how dates are handled →