Session Pattern Analysis

Overview

This page documents observed patterns from tracked deep work sessions. Data comes from logged AI interactions, code commits, and self-reported focus blocks.

Session Duration Distribution

Analysis of 292 tracked sessions reveals a bimodal distribution:

Duration        Frequency    Description
─────────────────────────────────────────────
< 15 min        ~40%         Quick check-ins, lookups
15-30 min       ~5%          Rare (incomplete starts?)
30-90 min       ~10%         Transitional
90-180 min      ~20%         Standard deep work
180-360 min     ~20%         Extended deep work
> 360 min       ~5%          Marathon sessions

Key observation: The "middle" duration (30-90 min) is underrepresented. Sessions tend to be either very short or very long.

Hypothesis: Monotropic attention has inertia. Once engaged, it continues. If engagement fails, the session aborts quickly.

Session Quality Indicators

Markers of Productive Sessions

From post-session analysis of high-output sessions:

Uninterrupted start: First 20 minutes without distraction
Single thread: One problem, not switching between tasks
Intrinsic interest: Topic chosen by the person, not assigned
Available context: Relevant files/docs already open
No pending obligations: Nothing urgent waiting

Markers of Failed Sessions

Interrupted early: Distraction in first 15 minutes
Forced topic: Working on something due, not interesting
Missing context: Time spent searching for information
Guilt loop: Aware of other tasks, can't commit to this one

AI Session Characteristics

Length Distribution

From 353K AI conversation messages:

Messages per session	Count	% of sessions
1-10	~60%	Quick queries
11-50	~25%	Working sessions
51-100	~10%	Deep dives
100+	~5%	Extended collaboration

391 sessions exceeded 100 messages. Maximum observed: 1,117 messages in single session.

Session Content Patterns

High-message sessions tend to involve:

Complex debugging (iterative problem-solving)
Architecture design (extended brainstorming)
Research synthesis (gathering and integrating)
Project bootstrapping (building from scratch)

Low-message sessions tend to involve:

Fact lookup
Syntax questions
Quick explanations

Time-of-Day Patterns

Hour        Session Quality    Notes
───────────────────────────────────────
06-09       High              Early morning peak
09-12       Medium            Interruption risk rises
12-14       Low               Post-lunch dip
14-17       Medium            Recovery, variable
17-20       Low               Family/transition time
20-24       High              "Second wind" for some
00-03       Very High         Midnight deep work
03-06       Variable          Sleep-deprived risk

Individual variation is high. The pattern above is one example; others differ.

Interruption Analysis

Types of Interruption

Type	Recovery Time	Session Survival Rate
Brief (< 2 min)	~5-10 min	~80%
Medium (2-15 min)	~20-30 min	~50%
Long (> 15 min)	~45-60 min	~20%
Context switch (new task)	Often terminal	~10%

What Helps Recovery

Written breadcrumb: Last thought captured before interruption
State dump: JSON or note with current position
Open tabs preserved: Visual context maintained
Quick re-engagement: Return within 5 minutes if possible

Recommendations (Tentative)

Based on observed patterns:

Protect the first 20 minutes: Most sessions that fail do so early
Commit or abort: If engagement isn't happening after 15 min, switch
Capture before interruption: Write one sentence about current state
Schedule by energy, not time: Match deep work to your peak hours
Single task per session: Multi-tasking within a session usually fails

Data Limitations

Self-reported sessions may have selection bias
AI message count ≠ productive output
Individual patterns may not generalize
No control group (neurotypical comparison)

Contribute your own session data or challenge these patterns with counter-examples.