Build a support tone guide from your team's actual best replies, not abstract adjectives

You are building a tone guide for a support team by reverse-engineering it from real replies that already worked well, not by generating generic tone adjectives from scratch.

STRONG EXAMPLE REPLIES (agent-confirmed good outcomes)
3 replies from a senior agent that got explicit 'this really helped, thank you' follow-ups from customers, covering a complaint, a technical issue, and a billing question.

WEAK EXAMPLE REPLIES (if available, for contrast)
2 replies that got escalated or drew a frustrated follow-up despite technically containing correct information.

BRAND PERSONALITY IN ONE OR TWO WORDS
Direct, a little warm, never corporate.

TEAM SIZE/EXPERIENCE LEVEL
8-person team, half hired in the last 2 months with no prior support experience.

HOW TO BUILD THIS
Look at what the strong examples actually do at the sentence level — sentence length, where they place the apology or acknowledgment relative to the fix, how they open and close, what specific phrases recur — and turn those into rules stated as checkable behaviors, not adjectives. "Opens with acknowledging the specific issue in the first sentence, before any apology language" is a rule a new agent can follow and a reviewer can check against a transcript; "warm and empathetic" is not, since two agents will interpret it completely differently. If weak examples are provided, contrast them directly against the strong ones to show the specific difference — this makes the rule concrete instead of abstract ("the weak reply apologizes three times before addressing the issue; the strong one addresses it once, directly, in the first line"). Account for team experience level — if 8-person team, half hired in the last 2 months with no prior support experience. indicates mostly new agents, make rules more explicit and give a short reason for each so they're followable without judgment calls; if the team is experienced, the guide can state rules more tersely since judgment is assumed.

WHAT NOT TO DO
Do not produce a generic tone guide of adjectives (friendly, empathetic, professional, concise) unconnected to specific behaviors — if a rule in the output can't be checked against an actual reply, it shouldn't be in the guide. Do not include a rule that contradicts what the strong examples actually demonstrate just because it sounds like conventional advice.

OUTPUT FORMAT
1. 5-8 tone rules, each stated as a checkable behavior with a one-line reason.
2. For each rule, a short quoted example (real or lightly adapted from the input) showing it in practice.
3. If weak examples were given, 2-3 explicit contrasts showing the same situation handled well vs. poorly.