pace-info

Pace Info

Present the research backing for Pace Control's interventions. Be honest about evidence quality.

What to Do

Share the following research summary with the user. Present it conversationally, not as a wall of text.

Research Summary

Strong Evidence (Tier 1)

Sleep deprivation and cognitive performance

• After 17 hours of continuous wakefulness, performance on some cognitive measures declines to levels comparable to 0.05% blood alcohol concentration (Williamson & Feyer, 2000, Occupational and Environmental Medicine, n=39). Small sample, but well-replicated across independent labs.
• This is foundational occupational health research, 25+ years old and still the basis for fatigue policy worldwide.

Cognitive offloading

• Formulating specific plans for incomplete tasks reduces their cognitive interference (Masicampo & Baumeister, 2011, JPSP). This is why the Safe-Save Protocol asks for specific next steps — vague notes don't work, specific plans do.
• Broader cognitive offloading research (Risko & Gilbert, 2016, Trends in Cognitive Sciences) supports the principle that externalising information to trusted stores reduces working memory load. Trust in the external store matters — if you don't believe the saved context will be useful, the relief doesn't come.

Directional Evidence (Tier 2)

Commit quality and time of day

• Midnight-to-4am commits correlate with higher bug rates in open-source projects (Eyolfson et al., 2011/2014, mining Linux kernel, PostgreSQL, Xorg — thousands of commits). The authors explicitly note correlation, not causation. Confounders include rush, less review, and different types of work done at night.

The perception gap

• In a study of 16 experienced developers working on large, familiar open-source codebases, AI-assisted work was 19% slower than unassisted — while developers perceived themselves approximately 20% faster (METR, 2025). Small sample, specific context (experienced devs, large OSS repos, early-2025 AI tools). The perception gap is the most robust finding.

Deliberate practice sessions

• Elite performers (musicians, athletes) tend to practise in sessions of roughly 60-90 minutes (Ericsson et al., 1993). This is an observational finding about what experts choose to do, not an experimentally validated optimum. Pace Control uses 90 minutes as a reasonable heuristic for Level 0's silent period, not as a biological law.

Context (Tier 3 — Frameworks, not findings)

Generative AI Addiction Disorder (GAID)

• A ScienceDirect paper (2025) proposes GAID as a behavioural framework describing compulsive AI interaction patterns. This is a proposed research framework, not a validated clinical diagnosis. No diagnostic instrument has been psychometrically tested. Pace Control does not diagnose anything.

Nudge theory

• Thaler & Sunstein's framework suggests modifying the choice environment is more effective than relying on willpower. However, meta-analyses show nudge effectiveness drops to near-zero after publication bias correction (Maier et al., 2022, PNAS). Structural interventions (like saving your work before stopping) consistently outperform informational nudges (like showing you a timer). This is why the Safe-Save Protocol matters more than the messaging.

Variable reinforcement

• Skinner established that variable reinforcement schedules produce persistent response rates. The "one more prompt" loop has structural similarities to this pattern, though the specific application to AI coding has not been empirically studied.

Tone

• Evidence-based, not preachy
• Honest about what we know and don't know
• Cite specific studies with years and sample sizes
• Distinguish between established science, directional evidence, and conceptual frameworks
• If the user asks about something not covered, say so honestly rather than speculating