The Promotion Metric That Isn't on Your Performance Review
Promotion criteria in Chinese tech teams are often opaque to outsiders. This account makes the unwritten rule explicit: sustained delivery on tasks slightly above your current level, not tool-count or visibility, is what triggers a manager to fight for your raise.
A junior hire in 2024 started with a simple cancel-workflow feature. The code was structurally clean — validate, execute, sync, persist — and the developer could explain exactly why his DDD-style in-memory state change avoided a partial-failure scenario. That clarity, plus consistently low bug counts and fast incident response, earned him progressively larger assignments.
The turning point came during a third-party integration. The vendor proposed batch-syncing orders via scheduled jobs. The developer pushed back, arguing that a batch spike from multiple clients would destabilize the vendor's system and cascade to their own. He insisted on real-time, order-by-order sync with a cancel API. The lead backed him immediately.
By 2025 he was promoted to mid-level with a ¥2,500 raise and an A rating. By 2026 he owned entire modules — store scheduling, material estimation — from design through delivery. The through-line: he consistently delivered on work that sat just beyond his current job description.
The developer's early DDD answer — that root.cancel() only mutates an in-memory object, so no partial-failure race exists — is technically correct but incomplete; the lead still pushed for a compensation mechanism, which is the real production-hardening step many juniors skip.
Pushing back on a vendor's architecture is a high-risk move for a junior. It worked here because the objection was grounded in a concrete failure mode (multi-tenant batch spikes), not in preference, and the lead was willing to absorb the political cost.
The promotion logic described — 'always handle things slightly beyond your current ability' — is effectively a continuous stretch-assignment filter. It selects for learning velocity, not for current competence, which is the opposite of how many Western performance-review grids are calibrated.