**Scope statement** This review covers the literature on psychological safety in software engineering teams, drawing on foundational organizational behavior research (Edmondson 1999) and industry-specific work from 2015-2023. It focuses on the relationship between psychological safety and team performance outcomes; it does not cover the leadership behaviors that produce psychological safety (a related but separate literature). **Map of the field** The literature is organized around three positions. The strong position, building on Edmondson's foundational work, holds that psychological safety is a strong predictor of team performance across contexts including software teams. The moderate position holds that psychological safety matters but is one of several team characteristics that jointly determine performance. The skeptical position, less prominent but visible in some industry analyses, holds that the construct has been over-applied and that the effect sizes in real software teams are smaller than the popular narrative suggests. **Themes** **1. Definition and measurement.** Edmondson (1999) defines psychological safety as "a shared belief that the team is safe for interpersonal risk-taking." Subsequent work largely adopts this definition, but operationalizes it differently. Google's Aristotle study used a 7-item self-report scale. Microsoft Research uses behavioral indicators (rate of dissent in code reviews, willingness to flag concerns in retrospectives). The construct is the same but the measures aren't, which complicates direct comparison. **2. Psychological safety and team performance.** Aristotle (Google, 2015) identified psychological safety as the single largest predictor of team effectiveness across 180+ teams. Microsoft Research papers find a positive correlation but with effect sizes that vary by team type (R&D teams larger effect; maintenance teams smaller). State of DevOps surveys consistently find psychological safety correlates with high-performing organizations but cannot establish causation. The convergence is on direction; the divergence is on magnitude. **3. The "safety vs. accountability" tension.** Several Microsoft papers and Edmondson's later work address a misreading of psychological safety as freedom-from-criticism. The empirical finding is that high-performing teams combine high safety with high accountability — disagreement is expected and welcomed, but standards are not lowered. Industry interpretations sometimes collapse safety into "be nice," which the literature does not support. **4. Mechanisms.** Why does safety affect performance? The literature offers several candidate mechanisms: faster problem surfacing, better knowledge sharing, lower turnover, more candid retrospectives. Microsoft's 2022 paper isolates "willingness to surface bugs early" as a specific mediator. The aggregate picture is that safety likely affects performance through multiple channels, not one dominant one. **Gaps in the literature** - Most studies are correlational. Experimental manipulation of psychological safety is rare and limited to small lab settings. - Cross-cultural generalizability is under-studied. Most data is from North American and European companies. The construct may operationalize differently in cultures with different norms around hierarchy and disagreement. - Long-term effects of safety-building interventions are not well measured. We have snapshots, not trajectories. - The interaction between safety and remote / hybrid work modalities is just beginning to be studied. **Where my read sits** The literature collectively supports the claim that psychological safety is a meaningful predictor of software team performance. The strongest version of the claim (Aristotle's "single largest predictor") is likely an overstatement on the magnitude; the moderate position is more defensible. The reframe from safety-as-niceness to safety-with-accountability is well-supported and underappreciated outside research circles. **Useful next reads** - Edmondson's "The Fearless Organization" (2018) for the practitioner synthesis of her 20-year program. - The DORA "Accelerate" book for the industry-survey methodology. - Recent critiques of the construct in organizational behavior journals; the literature now has its own self-critical voices.