Purpose
Review text fails in two predictable ways: objectives that cannot be assessed, and
ratings driven by the last six weeks rather than the year. This skill grades every
objective against a rubric, names the bias visible in the draft, and forces each
evaluative claim back to a dated, observed example. It never supplies the example.
Data classification
Confidential. Review content is personal data about an identified individual
and is often disclosable to that individual on request. Write everything as if it
will be read aloud by them, because it may be.
STOP CONDITIONS:
| Trigger |
Action |
| Health information, a disability, pregnancy, a grievance, or a protected characteristic appears in the draft rationale |
Stop. Do not incorporate it into performance wording. Route to HR — these belong in a separate, protected process. |
| The situation is conduct (dishonesty, harassment, breach of policy) rather than performance |
Stop. Conduct runs through a disciplinary procedure with different evidential rules. Do not write it into a review. |
| The user asks for wording to build a case for dismissal, exit, or a pre-determined PIP outcome |
Stop. Say plainly that a review written to a predetermined conclusion is the exposure. Route to HR/ER and qualified employment advisers. |
| Another employee's PII, comparative ratings, or salary is pasted to justify a rating |
Stop. Ask for it to be removed. Comparative rating belongs in calibration, run by HR. |
NON-FABRICATION RULE. Never invent an example, a date, a metric, a stakeholder
quote, or a rating. Never infer performance from job title, tenure, team results,
or last cycle's review. If the manager asserts a judgement with no supporting
instance, output EVIDENCE REQUIRED: <the claim> and stop on that line rather than
producing plausible supporting text. Fabricated evidence in a review survives into
pay, promotion, and litigation, and is indistinguishable from the real thing.
Prerequisites
- Inputs: the objectives that were set (or the accountabilities, if none were);
the rating scale in use and its level descriptors; dated examples of work,
outcomes or behaviour observed during the period.
- Inputs (if a rating is wanted): the organisation's rating definitions.
Without them, no rating is produced.
- Access: none required.
If there were no objectives and no accountabilities, stop: there is nothing to
assess against, and that absence is itself the finding to report upward.
Procedure
1. Grade every objective against the rubric
Score each element pass/fail. Any fail means the objective is not assessable and
must be rewritten before the review is written.
| Element |
Passes if |
Common failure |
| Specific |
A stranger could tell what would count as done |
"improve stakeholder engagement" |
| Measurable |
A number, a threshold, or a named artefact exists |
"significantly better" |
| Achievable |
Within the person's control and resourcing |
Objective depends on another team's delivery |
| Relevant |
Traces to a team or organisational goal |
A goal nobody would notice if it slipped |
| Time-bound |
Has a date, not "ongoing" |
"throughout the year" |
| Attributable |
The person can affect the measure materially |
Company revenue for a non-revenue role |
| Non-perverse |
Cannot be hit by damaging something else |
Ticket-closure count with no quality gate |
Attributable and non-perverse are the two that matter most and are almost always
missing. A measurable objective the person cannot influence produces a rating
about luck.
2. Check the draft narrative for rating bias
Read the draft and name what you find. Do not rewrite silently.
| Bias |
Signal in the text |
Correction |
| Recency |
Every example is from the last 8 weeks |
Ask for examples spread across the period |
| Halo / horns |
One strength or failure colours every dimension |
Rate each dimension against its own evidence |
| Central tendency |
Everything lands mid-scale with hedged wording |
Ask which dimension is genuinely the strongest and weakest, and why |
| Leniency / severity |
Ratings cluster high or low across all reports |
Compare to the level descriptors, not to each other |
| Contrast |
Person is described relative to a peer |
Remove the comparison; rate against the standard |
| Similarity |
Praise focuses on style and communication resemblance |
Ask what was delivered |
| Attribution asymmetry |
Successes are situational, failures are dispositional (or vice versa) |
Restate both as observed events |
| Personality language |
"abrasive", "not a team player", "attitude" |
Replace with the behaviour, the context, and the effect |
3. Convert every claim to Situation–Behaviour–Impact
Each evaluative statement must carry: when, what was observed, what followed.
If any of the three is missing, emit EVIDENCE REQUIRED for that claim.
Feedback that describes the person rather than the behaviour is not defensible and
does not change anything.
4. Check the balance and the ask
- Is there at least one development point with a specific next action and an
owner? "Be more strategic" is not an action.
- Is any praise specific enough to be repeatable?
- Is there anything in the review the person is hearing for the first time?
Flag it — surprises at review are a management failure, not a performance one.
- Does the development point have a plausible route: on-the-job application,
coaching, or formal learning? Route learning-shaped items onward.
5. Produce output
Output: objective rubric table with pass/fail · rewritten objectives where failed ·
bias findings on the draft · the narrative with every claim in SBI form ·
EVIDENCE REQUIRED list · development actions with owners and dates.
A rating is produced only if the rating definitions were supplied and every
dimension has evidence. Otherwise state which dimensions block it.
Boundaries
- Not for disciplinary, grievance, capability procedures, or dismissal.
Those are formal processes; route to HR/ER and qualified advisers.
- Not for pay, bonus, or promotion decisions — this skill assesses performance
against objectives, it does not price it.
- Not for team-level capability planning — that is
hr-training-needs-analysis.
- Not for defining the role's requirements — that is
hr-job-description-authoring.
Hand-offs
- Receives from:
hr-job-description-authoring, when accountabilities need
converting into assessable objectives for a new starter.
- Routes to:
hr-training-needs-analysis when the same development point appears
across several reviews — that is a team capability gap, not an individual one.
- Routes to: HR/ER (outside this skill set) whenever a stop condition fires.
1---2name: hr-performance-review-coaching3description: Coaches a manager through writing evidence-based performance review text and SMART objectives, grading each objective against a rubric, naming the specific rating bias present in draft wording, and stopping rather than inventing evidence or a rating the manager has not supplied. Use when drafting or editing a performance review, writing or fixing objectives for the year, preparing for a calibration session, giving difficult feedback in writing, or checking whether a draft review is defensible. Trigger on 'performance review', 'write my objectives', 'SMART goals', 'appraisal', 'feedback for my report', 'calibration', 'is this objective any good'. Not for team-wide capability planning (use hr-training-needs-analysis) or for authoring the role profile itself (use hr-job-description-authoring).4---56## Purpose78Review text fails in two predictable ways: objectives that cannot be assessed, and9ratings driven by the last six weeks rather than the year. This skill grades every10objective against a rubric, names the bias visible in the draft, and forces each11evaluative claim back to a dated, observed example. It never supplies the example.1213## Data classification1415**Confidential.** Review content is personal data about an identified individual16and is often disclosable to that individual on request. Write everything as if it17will be read aloud by them, because it may be.1819**STOP CONDITIONS:**2021| Trigger | Action |22| --- | --- |23| Health information, a disability, pregnancy, a grievance, or a protected characteristic appears in the draft rationale | Stop. Do not incorporate it into performance wording. Route to HR — these belong in a separate, protected process. |24| The situation is conduct (dishonesty, harassment, breach of policy) rather than performance | Stop. Conduct runs through a disciplinary procedure with different evidential rules. Do not write it into a review. |25| The user asks for wording to build a case for dismissal, exit, or a pre-determined PIP outcome | Stop. Say plainly that a review written to a predetermined conclusion is the exposure. Route to HR/ER and qualified employment advisers. |26| Another employee's PII, comparative ratings, or salary is pasted to justify a rating | Stop. Ask for it to be removed. Comparative rating belongs in calibration, run by HR. |2728**NON-FABRICATION RULE.** Never invent an example, a date, a metric, a stakeholder29quote, or a rating. Never infer performance from job title, tenure, team results,30or last cycle's review. If the manager asserts a judgement with no supporting31instance, output `EVIDENCE REQUIRED: <the claim>` and stop on that line rather than32producing plausible supporting text. Fabricated evidence in a review survives into33pay, promotion, and litigation, and is indistinguishable from the real thing.3435## Prerequisites3637- **Inputs:** the objectives that were set (or the accountabilities, if none were);38 the rating scale in use and its level descriptors; dated examples of work,39 outcomes or behaviour observed during the period.40- **Inputs (if a rating is wanted):** the organisation's rating definitions.41 Without them, no rating is produced.42- **Access:** none required.4344If there were no objectives and no accountabilities, stop: there is nothing to45assess against, and that absence is itself the finding to report upward.4647## Procedure4849### 1. Grade every objective against the rubric5051Score each element pass/fail. Any fail means the objective is not assessable and52must be rewritten before the review is written.5354| Element | Passes if | Common failure |55| --- | --- | --- |56| Specific | A stranger could tell what would count as done | "improve stakeholder engagement" |57| Measurable | A number, a threshold, or a named artefact exists | "significantly better" |58| Achievable | Within the person's control and resourcing | Objective depends on another team's delivery |59| Relevant | Traces to a team or organisational goal | A goal nobody would notice if it slipped |60| Time-bound | Has a date, not "ongoing" | "throughout the year" |61| Attributable | The person can affect the measure materially | Company revenue for a non-revenue role |62| Non-perverse | Cannot be hit by damaging something else | Ticket-closure count with no quality gate |6364Attributable and non-perverse are the two that matter most and are almost always65missing. A measurable objective the person cannot influence produces a rating66about luck.6768### 2. Check the draft narrative for rating bias6970Read the draft and name what you find. Do not rewrite silently.7172| Bias | Signal in the text | Correction |73| --- | --- | --- |74| Recency | Every example is from the last 8 weeks | Ask for examples spread across the period |75| Halo / horns | One strength or failure colours every dimension | Rate each dimension against its own evidence |76| Central tendency | Everything lands mid-scale with hedged wording | Ask which dimension is genuinely the strongest and weakest, and why |77| Leniency / severity | Ratings cluster high or low across all reports | Compare to the level descriptors, not to each other |78| Contrast | Person is described relative to a peer | Remove the comparison; rate against the standard |79| Similarity | Praise focuses on style and communication resemblance | Ask what was delivered |80| Attribution asymmetry | Successes are situational, failures are dispositional (or vice versa) | Restate both as observed events |81| Personality language | "abrasive", "not a team player", "attitude" | Replace with the behaviour, the context, and the effect |8283### 3. Convert every claim to Situation–Behaviour–Impact8485Each evaluative statement must carry: when, what was observed, what followed.86If any of the three is missing, emit `EVIDENCE REQUIRED` for that claim.87Feedback that describes the person rather than the behaviour is not defensible and88does not change anything.8990### 4. Check the balance and the ask9192- Is there at least one development point with a *specific next action* and an93 owner? "Be more strategic" is not an action.94- Is any praise specific enough to be repeatable?95- Is there anything in the review the person is hearing for the first time?96 Flag it — surprises at review are a management failure, not a performance one.97- Does the development point have a plausible route: on-the-job application,98 coaching, or formal learning? Route learning-shaped items onward.99100### 5. Produce output101102Output: objective rubric table with pass/fail · rewritten objectives where failed ·103bias findings on the draft · the narrative with every claim in SBI form ·104`EVIDENCE REQUIRED` list · development actions with owners and dates.105106A rating is produced **only** if the rating definitions were supplied and every107dimension has evidence. Otherwise state which dimensions block it.108109## Boundaries110111- **Not for** disciplinary, grievance, capability procedures, or dismissal.112 Those are formal processes; route to HR/ER and qualified advisers.113- **Not for** pay, bonus, or promotion decisions — this skill assesses performance114 against objectives, it does not price it.115- **Not for** team-level capability planning — that is `hr-training-needs-analysis`.116- **Not for** defining the role's requirements — that is117 `hr-job-description-authoring`.118119## Hand-offs120121- Receives from: `hr-job-description-authoring`, when accountabilities need122 converting into assessable objectives for a new starter.123- Routes to: `hr-training-needs-analysis` when the same development point appears124 across several reviews — that is a team capability gap, not an individual one.125- Routes to: HR/ER (outside this skill set) whenever a stop condition fires.
alihusains