INFOSTOA / PRACTICE
A practical reading path through AI evaluation, web security, image performance, and data consistency, with official sources and small verification tasks.
Start with one problem you need to solve. Each reading pairs a reason to open the original documentation with an output you can produce yourself.
Write success criteria and example cases first. Compare prompt changes on the same inputs, and distinguish a fluent answer from a correct one.
Your output: an evaluation sheet with typical inputs, edge cases, and expected results
Audit external scripts and image requests. Start with report-only security policies and compare image changes at real display sizes on a slow connection.
Your output: a policy violation review and an image transfer comparison
Identify requests that need fresh data before enabling replicas. Test reads after writes and repeated reads separately to clarify consistency requirements.
Your output: freshness requirements by request and a session handling plan
Turn subjective output review into repeatable, task-specific tests with human-calibrated scoring.
Create 20 representative test cases, define a pass criterion, and compare two prompt versions using the same rubric.
Use the methodology. OpenAI’s Evals dashboard and API are scheduled to become read-only on 2026-10-31 and shut down on 2026-11-30; do not build a new workflow around it.
Read the official source developers.openai.comDefine success before revising prompts, and recognize when model selection deserves attention instead.
Choose one failing task, write measurable success criteria, and change one prompt element before retesting.
Understand browser-enforced resource restrictions, nonce/hash policies, and safe deployment testing.
On a staging page, test a report-only policy, inspect violations, and repair one incompatible script pattern.
Report-only policies require an HTTP response header; they cannot be supplied through a meta element.
Read the official source developer.mozilla.orgConnect image format, responsive sizing, and loading choices to actual transfer cost.
Add three width variants and accurate sizes to one image; compare selected files and transferred bytes at mobile and desktop widths.
Understand the relationship between nearby database reads, replica lag, and session consistency.
In a test database, preserve a session bookmark across a write/read sequence and inspect serving-region metadata.
Replication needs the Sessions API. Writes still reach the primary; an unconstrained first read does not guarantee the latest database version.
Read the official source developers.cloudflare.com