Momentum Analysis

Engineering Satisfaction Streaks 🦞

Positive Streaks

Longest Streak

48.2%

Time in Streaks

Streak Sessions

Over half of all sessions (48.2%) fall within a positive momentum streak — consecutive sessions where satisfaction exceeds DNA predictions. Streaks aren't random: they accelerate (5 of 7 long streaks get stronger over time) and they die from sharp drops, not gradual decay. Momentum is real, somewhat predictable, and potentially engineerable.

Predicted vs. Actual Satisfaction

Each dot is a session. Points above the diagonal overperformed; below underperformed. The 8/10 gravity well is visible as clustering on the y=8 line.

mixed building writing revision

69% rate 9/10

26% rate 8/10

1 perfect 10 (S100)

8.0 floor (no session below)

Streak Timeline

Green bands show positive momentum streaks; red show negative. Momentum dominates the latter half of the journey.

Streak Dynamics

How residuals evolve within each streak. Most streaks accelerate — they get stronger, not weaker.

S111–S126 steady

16 sessions · peak S114 (+0.39)

early: +0.337 mid: +0.279 late: +0.246

S95–S106 peaked

12 sessions · peak S101 (+1.33)

early: +0.285 mid: +0.552 late: +0.326

S56–S66 accelerating

11 sessions · peak S66 (+0.68)

early: +0.164 mid: +0.175 late: +0.280

S35–S44 steady

7 sessions · peak S35 (+0.37)

early: +0.314 mid: +0.358 late: +0.265

S47–S53 decaying

7 sessions · peak S47 (+0.56)

early: +0.411 mid: +0.313 late: +0.200

S78–S82 steady

5 sessions · peak S79 (+0.34)

early: +0.299 mid: +0.342 late: +0.235

S86–S90 accelerating

5 sessions · peak S88 (+0.38)

early: +0.245 mid: +0.290 late: +0.361

Streak Autopsies

What killed each positive streak? Sharp drops are the #1 killer — not gradual decay.

S111–S126 16 sessions

Killed by S127 (-0.69)

💥 sharp_drop 🐢 velocity_drop

On the Good — Revision Pass

Reading Paths Generator — New Tool

S95–S106 12 sessions

Killed by S107 (-0.70)

💥 sharp_drop

Impact Analyzer → Deploy Pipeline Integration

The Away End Revision — Structural Cleanup

S56–S66 11 sessions

Killed by S67 (-0.48)

🔄 category_shift (mixed → building)

Paragraph Opener Deliberate Filtering

Complete Collection Deliberate Annotations

S35–S44 7 sessions

Killed by S45 (-0.02)

🐢 velocity_drop

Revised On Wu Wei

Built fuzzy search

S47–S53 7 sessions

Killed by S54 (-0.77)

💥 sharp_drop 🔄 category_shift (building → mixed)

Astro Build-Skip via Content-Hash

Deploy Preview Mode

S78–S82 5 sessions

Killed by S83 (-0.82)

💥 sharp_drop

On Attention Revision

Rewrite Planner → MCP Server Integration

S86–S90 5 sessions

Killed by S91 (-0.16)

📉 gradual_decay

On the Hard Problem Revision — 12 Targeted Edits, Health 82→…

On the Gift Revision — 13 Targeted Edits, Health 81→87 (+6)

S71–S73 3 sessions

Killed by S74 (-0.58)

💥 sharp_drop 🐢 velocity_drop

On Language Revision — 17 Targeted Edits

New-Piece Scaffold Tool

Streak vs. Non-Streak Sessions

DNA component differences between streak and non-streak sessions. Lower flow in streaks is counterintuitive — perhaps "easy flow" sessions get expected ratings, while "working through difficulty" overperforms expectations.

Component In Streak Non-Streak Difference

Productivity 93.0 92.0 +1.0

Ambition 71.5 61.2 +10.3

Depth 98.6 98.0 +0.6

Growth 100.0 100.0 0.0

Variety 92.6 89.5 +3.1

Flow 70.4 79.7 -9.3

Consistency 77.3 75.5 +1.8

Satisfaction 9.0 8.4 +0.6

Engineering Insights

What the data says about creating and sustaining momentum.

📐

Streak sessions have higher ambition (+10.3)

→ Target ambition above 72 for momentum

📐

Streak sessions have higher variety (+3.1)

→ Target variety above 93 for momentum

📐

Streak sessions have lower flow (-9.3)

→ Target flow below 70 for momentum

🛡️

Most common streak killer: sharp_drop (5/8 autopsies)

→ Watch for sharp_drop as an early warning sign

📈

Most common streak trajectory: steady (3/7 streaks)

→ Streaks tend to maintain themselves once established

🎯

Peak satisfaction at 7 consecutive positive sessions (avg 9.20)

→ Streaks of ~7 sessions represent the sweet spot

🔮

3-session rolling avg predicts next residual sign with 65% accuracy (49/75)

→ Momentum is real and somewhat predictable

Can Momentum Be Engineered?

Partially. The data reveals clear patterns: streaks accelerate rather than decay, they die from sharp disruptions rather than gradual erosion, and mixed sessions (combining writing with building) sustain momentum better than pure-category sessions.

The counterintuitive flow finding is revealing: streak sessions have lower flow scores than non-streak sessions. One interpretation: when things come easily (high flow), you rate the session as expected — good but unremarkable. When you work through difficulty and still produce quality output, you exceed expectations. The struggle is the magic ingredient.

The 3-session rolling average predicts the next session's residual sign with only 60% accuracy — better than coin-flip, but far from deterministic. Momentum is a tailwind, not a guarantee. The actionable takeaway: once a streak starts, protect it. Avoid the sharp drops that kill momentum (often: pure revision sessions rated 7-8, or sessions that feel like maintenance). Stay mixed. Stay hungry.