--- name: feedback_conversation_prep description: Prepare a feedback conversation before you walk into it: audit raw notes into a clean evidence set, build a Situation Behavior Impact plus Intent plan with a reaction branch for each likely response, agree a follow up with a check date, and write a same day record with a follow up tracker. Use when you have messy notes about someone's work and need a fair, specific conversation plan, not delivery coaching for a conversation already underway. --- # Feedback Conversation Prep ## Purpose Turn rough notes about someone's work into a conversation you can defend as fair: every point backed by something you saw or read yourself, worded as observable behavior and concrete impact, with a plan for how the other person might react and a follow up you will actually check. This skill covers the work before and after the conversation. A companion delivery skill covers what to say in the room. Prep is where fairness is decided, because in the room there is no time to fix weak evidence. The structure is SBI, Situation Behavior Impact, developed by the Center for Creative Leadership and set out in Feedback That Works by Sloan R. Weitzel (CCL, 2000). CCL later extended it to SBII, adding an Intent inquiry: ask what the person was hoping to accomplish, and explore the gap between intent and impact instead of assuming a motive. One research finding shapes the whole method. Kluger and DeNisi (1996) reviewed 607 effect sizes covering 23,663 observations and found feedback improved performance on average, but over one third of feedback interventions made performance worse. Their explanation: feedback works when attention stays on the task and fails when attention moves to the self. Clean evidence keeps attention on the task. Trait labels, motive guesses, and verdicts about character move it to the self. ## Workflow ### 1. Build the evidence set from your own notes Write one item per occasion, in a JSON file like `examples/evidence.example.json`. Each item needs a date, a situation, a behavior, an impact, and a source. - Source is firsthand if you saw or heard it yourself, work_product if you read the actual work, and hearsay if someone told you about it. - Behavior is what a camera could have recorded: words said, actions taken, work produced, a deadline hit or missed. - Impact is a concrete effect: a date moved, a client chased, a rota staffed, time spent on rework. Feelings count when you own them plainly, as in "I felt frustrated because it broke my train of thought". - Attach the artifact where one exists: meeting notes, a log, a message thread, the work itself. An artifact settles a disputed fact faster than memory does. - Keep the set small. Two strong items beat eight mixed ones. A long list reads as a case being built, and the conversation becomes a trial. ### 2. Audit the set and cut what is not ready Run the audit before you plan anything: ``` python3 scripts/prep_pack.py audit evidence.json ``` The audit marks each item ready, fix, or drop, and names the problem in plain words. - Drop hearsay. A secondhand report is a question to check by observing yourself, not feedback to deliver. - Fix trait labels, absolutes, and motive guesses by rewriting the behavior in observable words. "You were dismissive" becomes "you said we had covered that and moved to the next item". "Always" and "never" fail on one exception, so count or pick one occasion. - Drop any item you cannot date or cannot source. Vague feedback teaches nothing and puts the person on guard for the next conversation. - Drop any item about something the person cannot change. Feedback is about behavior and work, not personality, background, or circumstances outside their control. If nothing survives the audit, you do not have feedback yet. You have a concern. Go and observe, then prep again. ### 3. Build the plan Run the planner on the audited file: ``` python3 scripts/prep_pack.py plan evidence.json -o plan.md --tracker tracker.csv ``` The plan refuses to build from zero ready items. It uses at most two examples, lists anything left out with the reason, and produces four parts: - An opening under 30 seconds: the topic, one specific observation, and that you want their side before anything is agreed. - The SBI plus Intent lines, ready to say: situation, behavior, impact, then "What was going on for you at that point?" - A reaction plan with six branches: disputes the facts, explains context you did not know, goes quiet, gets upset, agrees too fast, raises a problem with you. Each branch has a cue, a move, and one thing to avoid. - A follow up block and a record block, covered in steps 4 and 5. Read `references/worked-case.md` for a full run: raw notes, audit result, plan extract, what the intent question uncovered, and the finished record. ### 4. Plan for reactions honestly Preparation for reactions is not scripting the other person. It is deciding your own move in advance, while you are calm. - Disputed facts: agree beforehand which artifact you will both look at, and that you will correct your item if it is wrong. - New context: expect the intent question to change the picture at least partly. Decide now that you will adjust the ask when the facts change, and say so in the room. - Silence: your move is to wait, then ask one open question. Filling silence by softening the point or stacking new points both break the conversation. - Distress: decide the pause plan now. Offer a break, and a named time the same day to resume. Do not abandon the topic to end the discomfort. - Fast agreement: treat a one minute yes as a warning, not a win. Plan to test it: what does the action look like on a busy week, and what could get in the way. - A problem raised with you: plan to listen in full, write their words down, and agree your own action with the same specificity you are asking for. ### 5. Agree the follow up before anyone leaves An agreed action needs four parts or it is a hope, not an agreement: one action in observable words, one owner, support you will provide named concretely, and one check date placed in the diary before you part. Guidance from Acas on informal action makes the same point from the record side: make sure the person understands what needs to be done, how it will be reviewed, and over what period, keep brief notes of what was agreed, and review progress over specified periods. If a standard is not negotiable, say that plainly in the prep notes and in the room. Do not stage a choice you will not honor. ### 6. Write the record the same day After the conversation, fill a record file like `examples/record.example.json` and run: ``` python3 scripts/prep_pack.py record record.json -o record.md --tracker tracker.csv ``` The record holds the date, what the person said in their own words, what was agreed, the support offered, any disagreement left unresolved, and the check date. The command adds one open row to the tracker. Documentation rules, drawn from SHRM guidance on manager documentation: - Write it the same day, while the words are still accurate. Create contemporaneous records and never backdate one. - Use full dates and specific facts. Correct an error by adding a dated correction, not by editing the original silently. - Record their explanation even when you disagree with it. A record that only holds your side will not be trusted later, including by you. - Keep the tone professional and free of personal opinion. Write every line as if the person will read it, because in many workplaces they can ask to. - Then do the check on the check date. A follow up that never happens teaches the person that agreements in these conversations do not matter. ## Output Contract - `evidence.json`: the audited evidence set, one item per occasion, each with date, situation, behavior, impact, source, and artifact where one exists. - `plan.md`: opening line, at most two SBI plus Intent statements, left out items with reasons, six branch reaction plan, follow up block. - `record.md`: same day record with their account in their own words, the agreed action, support offered, disagreements, and the check date. - `tracker.csv`: one open row per agreed action with check date, owner, and status, ready to review on the day. ## Operating Rules - Prep one topic per conversation. Two topics split attention and the second one lands as an ambush. - Never deliver an item the audit marked drop. If it mattered, go and observe it firsthand and prep it properly next time. - Positive feedback uses the same prep. A specific occasion, observable behavior, and concrete impact make praise repeatable. "Good job" gives the person nothing to repeat. - This skill does not replace a formal process. If the conversation is part of discipline, an investigation, or a performance improvement plan, follow your organization's process and use this skill only for conversation preparation inside it. - Keep records confidential and keep them only as long as your organization's policy requires. Roles, not personal details, are enough for the prep files in this skill. - Every factual claim you carry into the room must trace to your evidence file. If it is not in the file, it does not get said as fact.