The Prosoche Protocol v1.0.

Plain-language summary

AI capability is compounding quickly while operator cognition remains constrained. This protocol tests whether prosoche, a Stoic attention discipline centered on impression-checking and deliberate assent during action, performs differently from mindfulness and productivity controls in AI-supervisor work. The design is a four-arm RCT with pre-registered hypotheses, matched intervention dose across active arms, and a pre-specified primary contrast: prosoche vs mindfulness on NASA-TLX change during a standardized AI-mediated task.

• Population: professional knowledge workers using AI tools at least 10 hours/week.
• Intervention: four-week protocol with distributed daily micro-practices and weekly matched calls.
• Primary inference: cognitive-load reduction under AI-supervised work, not generic wellness outcomes.
• Key contribution: methodologically strict active-control design intended to stay publishable under both positive and null outcomes.

1. Background and rationale

1.1 The bandwidth gap

The protocol grounds itself in a capability-vs-operator mismatch: frontier compute and agent horizons increase faster than human conscious processing, making operator attention the bottleneck. The trial builds on The Bandwidth Gap thesis and prior evidence on interruption costs, attention residue, and cognitive-load constraints in complex work.

1.2 Why prosoche, why now

Prosoche is operationalized as continuous attention to impressions, judgement separation, and action alignment under pressure. The protocol hypothesis is not universal superiority over mindfulness; it is domain-fit superiority for sustained AI-supervisor workflows where action-time evaluation matters.

1.3 Existing literature gap

The protocol reports no published RCT matching this exact comparison set: Stoic-derived attention intervention versus active mindfulness and productivity controls in AI-intensive work, with behavioral AI-tool outcomes and standardized supervisor tasks.

References

Primary references in the source include Allen (2001), Cresswell (2017), Engeser & Rheinberg (2008), Epictetus (Hard trans.), Epoch AI (2026), Gill (2010), Goyal et al. (2014), Hadot (1995, 1998), Hart & Staveland (1988), Imai et al. (2010), Khoury et al. (2013), Leroy (2009), Linardon et al. (2019), Marcus Aurelius (Hays trans.), Mark et al. (2008, 2014), METR (2026), Posner & Petersen (1990), Robertson et al. (1997), Sellars (2006), Seneca (Campbell trans.), Sweller (2020), Tapal et al. (2017), and Zheng & Meister (2024).

Full citation list preserved from the protocol source document and available in partner review materials.

Apparatus / citation

Cite this protocol.

@techreport{zenocenter2026prosoche,
  author      = {{Zeno Center}},
  title       = {The Prosoche Protocol v1.0: A Four-Week Randomized
                 Controlled Trial of Stoic-Derived Attention Training
                 in AI-Saturated Knowledge Work},
  institution = {Zeno Center},
  type        = {Pre-registered study protocol},
  version     = {1.0},
  year        = {2026},
  month       = apr,
  url         = {https://zeno.center/prosoche},
  urldate     = {2026-07-05},
  note        = {Protocol frozen at v1.0. OSF registration precedes
                 first enrolment; no participants enrolled at this
                 version.}
}

Related / research program

The rest of the program.