Predictive User Behavior — Agent 1o1

Monte Carlo simulation of 500 sessions across 5 personas · Markov chain state model

500
Completion Rate
44.4%
222 of 500 completed
Drop-off Rate
55.6%
278 abandoned
Avg Fuel / Session
126
GPU Render Capacity spent
Avg Time / Session
4m 42s
wall clock time
Avg Tool Calls
9
per session
Avg Revisions
0.2
revision cycles

User Journey Funnel

How many simulated users reach each stage

Start500 (100%)Project Setup264 (53%)Auto Screenplay318 (64%)Screenplay Writing188 (38%)Screenplay Review366 (73%)Bible Building389 (78%)Breakdown Running358 (72%)Image Generation362 (72%)Image Review349 (70%)Video Generation319 (64%)Video Review283 (57%)Export241 (48%)Completed222 (44%)

Tool Call Frequency

Most called tools across all sessions

universal_generate805read_canvas714write_screenplay589chat_message500read_screenplay470shot_breakdown408manage_ref406create_project264upload_to_canvas246propagate_changes98

State Flow Diagram

Transitions between workflow states (curve thickness = frequency, red = drop-off, amber = revision)

Start500 sessionsProject Setup264 sessionsAuto Screenplay318 sessionsWrite Screenplay271 sessionsReview Script470 sessionsBuild Bible406 sessionsShot Breakdown408 sessionsGen Images425 sessionsReview Images396 sessionsGen Videos360 sessionsReview Videos318 sessionsRevision98 sessionsExport246 sessionsCompleted0 sessionsDropped0 sessionsQuick Image20 sessions

Persona Radar

Sessions · Completion · Fuel

SessionsCompletion %Avg Fuel

Persona Breakdown

How each archetype performs

Speed Demon
Wants it fast, minimal revisions, budget-conscious. Uses cheap models. Loves auto-screenplay.
44%
179 sessions · 72 fuel
Storyteller
Deep narrative focus, writes detailed screenplays, iterates on story before visuals.
50%
122 sessions · 113 fuel
Perfectionist
Many revision cycles, high fuel spend, wants every shot perfect. Uses premium models.
46%
74 sessions · 296 fuel
Explorer
Tries everything, experiments with models, creates multiple projects. High tool diversity.
29%
68 sessions · 120 fuel
One-Shot Wonder
Wants the agent to do everything from one prompt. Minimal interaction, trusts AI fully.
51%
57 sessions · 110 fuel

Fuel Distribution

GPU Render Capacity spent per session

0-5014950-10021100-200257200-40051400-80022800+0GPU Render Capacity

Time Distribution

Time spent per session

<5min3185-15min17715-30min530-60min060min+0minutes

Predictive Insights

Auto-generated from simulation patterns

→44.4% of simulated users complete the full pipeline. 55.6% drop off.
→Biggest drop-off: "screenplay review" — 41 users abandon here. Fix: simplify this stage or add guidance.
→Most called tool: "universal_generate" (805 calls across all sessions).
→Best completion: "speed_demon" persona at 44%. Worst: "explorer" at 29%.
→Average fuel per session: 126. 43 users (9%) spend 2x+ the average.
→4% of users go through 2+ revision cycles. The cascade system saves them from manual dependency tracking.
→Average tool diversity: 6.7 unique tools per session. Users who use 6+ tools have higher completion rates.
Predictive model v1.0 · Markov chain + Monte Carlo · 5 personas · 500 simulated sessions