Learning Loop — Rebuilding Atlas's Feedback Culture

A four-round, advanced people-leadership simulation set inside Lumera Cloud Pty Ltd, a Sydney B2B SaaS scale-up (410 staff, 240 engineers in 22 squads, AUD 96M ARR, 78% gross margin, Series C 2024 with an eight-quarter break-even mandate). After the Series C, Lumera rolled out PulseHR, an algorithmic platform that scores engineers weekly on velocity, code-review latency and peer-recognition and feeds a quarterly pay calibration. Instead of making performance management objective, it taught engineers to game the metrics: candid feedback dried up and the flagship Atlas squad (9 engineers) has now missed three straight sprints, delaying a feature for the two largest customers (AUD 7.2M ARR). Voluntary feedback-seeking fell from ~14 to 3 events per engineer per quarter, the Atlas engagement index sits at 52/100, the trust item 'I trust how my performance is evaluated' is 34/100, and rework on 'done' work rose from 9% to 22%. A fourth miss triggers an AUD 540,000 service-credit clause; replacing a flight-risk senior engineer costs AUD 180,000. Playing the Engineering Team Lead, you (1) diagnose why a system built to increase feedback suppressed it — separating the feedback environment from the measurement system and naming the cost (image cost, ego cost) PulseHR inflated using Anseel's cost/benefit lens; (2) design a feedback-and-reflection operating model that fits inside a HARD budget of 6 engineer-hours per person per week, closing the loop so reflection feeds the next sprint; (3) re-tune PulseHR under an explicit trust constraint — the board will not switch off a AUD 1.1M platform — by re-weighting signals, choosing individual vs squad scoring, drawing the developmental-vs-evaluative boundary, and granting transparency; and (4) sequence a 90-day recovery and scale-out plan with leading and lagging indicators, a stop condition, and a pilot-before-fiat rollout. The math rewards lowering the image cost of candour over merely adding feedback events, fitting the 6-hour budget, separating development from pay, and aligning the algorithm change to trust — and punishes the five classic errors: confusing feedback volume with culture, detonating PulseHR, blurring development and evaluation, reflection that never closes the loop, and scaling the fix by fiat. Final KPIs track Feedback-Seeking events, Trust (0-100), Adaptive Performance (0-100), and Delivery Risk (the AUD 540,000 service-credit exposure).