Techies — Fair-Process & Performance Improvement Framework (Proposal)

Important — not legal advice. This framework is an operational proposal, not a legal document. Employment law varies significantly by jurisdiction, and metric-based performance management carries real risk (unfair-dismissal and discrimination/adverse-impact claims). Nothing here should be used in any employment decision until it has been reviewed and approved by qualified employment counsel and HR for the relevant location(s). The checkpoints in §7 make that review mandatory, not optional.

1Principles & Non-Negotiables

The rules that keep accountability fair and defensible.

A score is an input, not a verdict. KPI numbers inform a human judgement; they never auto-decide anyone's employment.
Support before exit. No one moves toward exit without first being told clearly, given real help to improve, and given reasonable time.
Calibrate before it counts. Scores carry no employment consequence until targets are baselined and the model passes a bias check.
Consistency. The same criteria, thresholds, and process apply to everyone in the same role — documented and repeatable.
Transformation is the goal. The aim is to grow software engineers into AI engineers; exit is the last resort when support hasn't worked, not the opening move.

Why this layer exists: the scoring system was built for coaching. Pointing it at employment decisions without these safeguards is unfair to people and legally fragile. This document is what makes the difference.

2Human-in-the-Loop Rule

The single most important safeguard.

The system may…	The system may never…
Measure, score, and flag low or declining performance	Trigger, recommend by default, or execute a termination
Surface a pattern for a human to review	Rank people for cutting without human judgement
Track improvement-plan progress	Close out a plan or decide its outcome automatically

Every consequential step is owned by a named person who can explain and justify it. The automation prepares evidence; people make decisions.

3The Fair-Process Ladder

How a sustained low score escalates — never a cliff, always a staircase. (Timelines are proposed defaults.)

Observe · 1 weak week
A single low or amber week is noted, not actioned. Could be variance, a hard project, or PTO. No conversation required beyond the normal 1:1.

Coach · 2–3 consecutive weak weeks
Manager raises it directly and supportively, identifies the specific gap, agrees concrete actions and support. Documented as coaching notes. Most cases resolve here.

Formal notice · pattern persists after coaching
Written notice that performance is below expectation, with specifics. Triggers a Performance Improvement Plan (§4). HR informed. Gate A applies.

Improvement plan · 30–60 days (proposed)
Clear targets, training, support, and check-ins toward the AI-engineer competency. Genuine opportunity to succeed. Progress reviewed at midpoint and end.

Review decision · at plan end
A human panel reviews evidence and decides: pass, extend, or exit. Gate B applies — no exit without legal/HR sign-off.

A developer can enter the ladder no faster than the steps allow. Skipping rungs (e.g. straight to exit on one bad month) is explicitly not permitted under this framework.

4The Improvement Plan (PIP)

A real opportunity to succeed — and the bridge into the AI-engineer transformation.

4.1 What every plan must contain

Element	Detail
Specific gaps	The exact metrics/behaviours below expectation, with evidence from the scorecards.
Clear targets	What "back on track" looks like, measurable, realistic for the period.
Support & training	Claude Code upskilling, pairing, mentorship, time allocated — the resources to actually get there.
Check-ins	Scheduled midpoint and end reviews; regular feedback, no surprises.
Duration	30–60 days (confirm with HR/legal)
Outcomes defined	Pass / extend / exit — stated up front so the process is transparent.

Transformation built in: because the objective is engineers → AI engineers, the PIP's support component is an upskilling track (Claude Code fluency, AI-assisted workflows), not just pressure. A plan with no real support attached is not a valid plan under this framework.

Fairness test: a plan must be one a capable person could realistically pass with the support offered. Plans designed to be unwinnable ("paper PIPs") are both unethical and a significant legal liability — they read as pretext for a decision already made.

5Decision Authority

Who decides what — no single person, no algorithm, owns an exit.

Step	Owner	Required input
Observe / Coach	Line manager	Scorecards, 1:1 notes
Formal notice / PIP	Manager + HR	Documented coaching history
Final review	Review panel (≥2 people incl. a director)	Full record + PIP evidence
Exit approval	Director + HR + legal sign-off	Gate B clearance (§7)

6Adverse-Impact & Bias Check

Run before the scoring is ever used in a decision, and periodically after.

Check for disparate impact: review whether low scores cluster by age, gender, disability, caregiving status, or other protected characteristics. AI-adoption metrics are the highest-risk area — slower tool uptake can correlate with age or disability.
Job-relatedness: confirm every scored metric genuinely reflects the role's requirements.
Small-sample caution: with a team of 1–10, individual scores are statistically noisy; weight monthly trends over single weeks and treat borderline cases conservatively.
Document the check: keep a record that the review was done and what it found.

If the check shows disparate impact, the scoring must be adjusted before use — proceeding anyway is where most legal exposure lives.

7Legal / HR Checkpoint Gates

Mandatory sign-off points. The process cannot advance past a gate without them.

Gate 0 — Before launch: employment counsel + HR review and approve this framework, the scoring model, and the bias check for each jurisdiction where staff are based. Scores carry no consequence until cleared.

Gate A — Before a formal notice / PIP: HR reviews the documented coaching history and confirms the case is consistent, evidenced, and ready for a formal plan.

Gate B — Before any exit: director + HR + employment-law sign-off that process was followed, the PIP was fair, documentation is complete, and the action is compliant for that jurisdiction.

Gate C — Periodic audit: recurring review (e.g. quarterly) that the framework is being applied consistently and the bias check still holds.

These gates are the "review checkpoint" safeguard. They convert legal review from an afterthought into a structural requirement the process literally cannot skip.

8Performance Exit vs Layoff — Keep Them Separate

These are legally distinct and must not be conflated.

Performance exit

Individual, conduct/capability-based, follows the ladder and PIP above. Requires documented underperformance and a fair chance to improve.

Layoff / redundancy

Role- or economics-based, not about the individual's performance. Has its own legal rules — selection criteria, notice, and sometimes consultation — that differ by jurisdiction.

Do not use the KPI score as a redundancy-selection tool without specific legal advice — using a performance metric to choose who is made redundant blends two regimes and is a common source of claims.

9Documentation Requirements

If it isn't written down, it didn't happen — for fairness and for defensibility.

Scorecards retained for each period, per developer.
Coaching conversations summarised and dated.
The PIP document, its support commitments, and each check-in outcome.
The bias-check record and any adjustments made.
Gate sign-offs (who approved, when, on what basis).
Final review rationale, including dissent if the panel wasn't unanimous.

10Decisions for Sign-Off

Decision	Proposed default	Needs
PIP duration	30–60 days	HR + legal
Weeks before formal notice	Pattern over ~3–4 weeks post-coaching	Director
Review panel composition	≥2, including a director + HR	Director
Jurisdictions in scope	[list staff locations]	HR + legal
Bias-check cadence	Before launch + quarterly	HR

10.1 Recommended sequence before anyone's job is on the line

Clear Gate 0 (legal + HR + bias check).
Run a 4-week baseline with scores visible but non-consequential.
Calibrate targets, then communicate the full process to the team openly.
Only then does the ladder carry employment consequences.