Data engineering & infrastructure
Bytescrap
Repeatable extraction runs that stay stable when a source page changes
Bytescrap extracts structured data from public web pages for research and monitoring workflows. Grafiesto built it around the failure mode that kills most scraping projects: a source page changes shape, extraction silently returns partial data, and nobody notices until a decision has already been made on it.
Bytescrap
Data engineering & infrastructure
Inside the project
Context before craft. Decisions before decoration. The story behind the finished surface.
Not a reveal. A reason.
Bytescrap extracts structured data from public web pages for research and monitoring workflows. Grafiesto built it around the failure mode that kills most scraping projects: a source page changes shape, extraction silently returns partial data, and nobody notices until a decision has already been made on it.
The starting point
What needed to change
Extraction is easy to demo and hard to operate. The demo runs against a page whose markup is known. Six weeks later the source has been redesigned, three selectors now match nothing, and the pipeline keeps succeeding — returning rows with empty fields that look like real observations. Every downstream analysis inherits the gap without ever being told about it.
The work
How we found the shape
Runs are structured around explicit output contracts rather than selector lists. A field that does not resolve is a run-level signal, not a null quietly written to a row. Sources are organised so a change can be attributed to the page that changed, and outputs are shaped for direct consumption by an analysis pipeline instead of requiring a cleanup pass before use.
What shifted
The difference it made
Bytescrap runs in production as a Grafiesto-operated product. Supported sources, usage limits, and product-specific safeguards are documented on the Bytescrap website.
The best work starts with the honest version of the problem.
Tell us yoursKeep going
01 / 03
Grafiesto
An audit engine that reports unknowns instead of inventing scores
Enter the next storyGrafiesto
Product engineering & measurement
Bring us the messy version
Let's make it mean something.
A useful first conversation needs no polished brief—just the real problem, the people it affects, and the change you want to see.