Confirmation strength must match consequence

What it is

Confirmation comes in tiers: none, implicit, explicit. A tier is not a tone of voice. It is a fee paid against the cost of being wrong. An office voice that dims the lights can be corrected with the next sentence; it should not collect a round of “is that right.” A graduate student submitting an appeal to the registrar cannot take the title back with the next sentence; the system should not bury the title in a passing clause. The matching principle is that confirmation strength follows consequence, not the presence of a confirmation module.

Why it happens

Every tier spends the same kind of resource: turns, time, the listener’s verification attention. None saves the resource and lets consequence carry the miss. Implicit stuffs verification into the progress sentence: fewer turns, but the listener must scan. Explicit makes consent its own turn: the surest stop, and the most expensive. Matching means the fee does not exceed the bill for being wrong, and is not less than it.

Consequence here is only a scale for choosing a tier: light consequence, low strength; heavy consequence, high strength. How the scale is scored in detail, whether an after-the-fact reverse can replace a pre-confirm, and whether the same confirm sentence said too often becomes a reflexive “yes,” belong to the next layer. What has to stand first is the direction: one strength across every skill overtaxes the light ones and leaves the heavy ones running naked. If dimming the office lights and submitting an appeal share a single “okay, done,” matching has already failed.

Studying it

Coarse-sort skills into light / mid / heavy consequence (can a miss be corrected in the next breath, and is there still time), then cross the three confirmation strengths. Dependent measures: false execution, time to completion, and whether people judged “this confirm was sized for this act.” The question is not “which confirm is better” but where mismatch appears: does explicit on light consequence only add turns; do implicit or none on heavy consequence surface only after execution.

In the field, tag each skill with a designed strength and compare it with the strength the logs actually took. Designed explicit that is skipped when the system is busy is matching being dismantled at runtime. Do not pick a tier from mean satisfaction: the weight people put on being over-asked about lights is not symmetric with the weight they put on an appeal that was never confirmed.

Where it stops holding

When consequence for the user and for the institution disagree (light for the user, heavy for the registrar), match to the heavier side and say in the prompt why this turn is heavier; otherwise people only see a mysterious “is that right.” When recognition confidence is very low, a temporary step up is a surcharge on uncertainty, not a change in consequence itself. Labelling every skill “mid” and always using implicit is fake-tiering to avoid matching. The principle does not teach how to score consequence. It requires that once scored, the tier actually follows the score.

Applying it

• List skills, split consequence into at least three bands, and bind a confirmation strength to each: light → none or implicit, mid → implicit, heavy → explicit. Put it on the skill table, not in a launch-week habit.
• Do not step a skill down because “this dialogue is already long.” Length is a different bill; stepping down removes protection when people are most tired.
• At runtime, if recognition confidence drops under a threshold, strength may go up, never down for throughput.
• How to check: sample live sessions from light and heavy skills. Frequent explicit on the light band is over-matching. Heavy skills with no separate consent before execution are under-matching. Change the table, not one sentence of copy.

Related

• Same group: M2.03.1 Explicit confirmation is reliable but slow · M2.03.2 Implicit confirmation embeds the ack in the next move
• Nearby: M2.09 Matching confirmation to consequence · M2.08 Explicit and implicit confirmation · M1.03 Dialogue turns
• Search terms: confirmation-consequence matching · confirmation strength · grounding strategy