[feat7]

Readers had to infer terms such as source offsets and ID reconciliation before seeing what the collector does. Rewrite the 19 reader-facing pages around one order event: how it enters the queue, reaches PostgreSQL, survives the two crash gaps and informs the measured writer choice. Explain unfamiliar terms where they matter and keep operator and experiment details in their reference pages.

Add source-backed event-journey and worker-replay diagrams with readable phone guides. The article links to the complete experiment plots; captured data, numeric result tables, previous figures and runtime behavior remain unchanged. Fix the first-run examples to reload the generated key after the deployment demo unsets it, and distinguish locally pending outbox rows from queue-acknowledged events awaiting SQL observation.

Validation:

• One fresh reader/fact-check review; both identified issues corrected.
• All 19 pages rendered on desktop and at 390px phone width; no page overflow or broken images, and no clipped labels in the phone guides.
• All 153 local links/images checked. Existing commands retained, with explicit key loading added to four examples and one duplicate configuration block removed.
• Actual shell setup plus the Python client constructor failed in all four examples before the credential fix and passed afterward using generated private settings; no credentials printed.
• Both full study audits passed: each 99 database runs, 33 HTTP windows, 9 recovery runs and 9 limited cases retained. Auditor unit tests passed. Measured result tables and prior figure/index hashes remain unchanged.
• Both new diagrams passed Archify source/artifact/browser gates and were visually inspected. New SVG exports only trim trailing source whitespace; geometry, labels and CSS tokens are unchanged.

Standard policy text and historical design/review records remain historical records. The rewrite follows the requested Humanizer editing guidance without inventing measurements, adoption or production claims.