Hint fatigue can be read from ignore and dismiss rates

What it is

Whether a channel has been spent by dose is not a team feeling. It is ignore rate and dismiss rate: a hint present that never earns readable dwell, and a hint closed at once with no subsequent use of the pointed-at feature. These ratios are an indirect measure of fatigue, used to throttle a shared budget—not as the only A/B score for one sentence. They answer whether the channel still opens, not whether this hint deserves to exist.

Why it happens

Fatigue is habituation on a population time series: later hints on the same channel are read less and closed faster. Ignore (present, unread) captures people who no longer treat the layer as information; dismiss captures people who will still move a hand, but whose goal is clearing the board. Clicks alone score dismiss as success. Plotting both ratios against the user’s hint index shows the collapse point—where the budget should clamp. Ratios must be sliced by channel, not by team: team A’s dismiss looks fine and team B’s looks bad because B always ships third that day. Team-level dashboards hide channel collapse; everyone thinks their hint is “fine.” Indirect means there is no in-skull fatigue sensor, only behavioral proxies. Proxies mix “this sentence was useless” with “the channel is dead,” so a low-dose control week on the same copy is needed to split copy quality from channel quality.

Studying it

Number automatic hints per user; record dwell, whether it was dismissed inside readable duration, and whether the pointed-at feature was used shortly after.

Independent variables: how many hints that user already saw that day or week, whether skins match. Dependent variables: ignore rate, dismiss rate, post-dismiss adoption, index of collapse.

A one-shot lab task cannot show serial collapse. Do not treat impressions as successful reach. A control week at half dose across the product is what splits channel effects from copy effects.

Where it stops holding

High dismiss on error copy may mean the error vanished, not fatigue. Forced layers have no lawful close, so the metric distorts; keep them out of teaching-channel fatigue. First-session dismiss for new users is often rushing to work; set thresholds apart from returning users. For assistive-tech users, “dismiss” may be escaping a focus trap, and the ratio means something else.

Applying it

• Define ignore (below readable dwell) and dismiss (closed inside readable duration with no adoption) for the shared teaching channel; report weekly by user index, not as team scoreboards.
• Pin the budget cap before the channel-level collapse point, not to a local “one per team per week.”
• Run a low-dose control on a schedule to separate bad copy from a dead channel.
• Verify: when ignore plus dismiss rise, cut total dose first and see whether reading of high-value hints returns. Changing only copy at the same dose usually leaves the ratios up—that is a channel problem misread as a copy problem.

Related

• Within the group: H2.10.1 Shared hint budgets need a single priority order across teams · H2.10.2 High-value hints should displace low-value ones, not stack on them · H2.10.4 When the budget is spent, new-feature hints wait in queue
• Adjacent: H2.07 Hint Restraint · H5.08 Alarm fatigue · H2.05 Feature Discovery
• Search terms: dismiss rate · hint ignore rate · channel fatigue