Autoresearch Pro-2: Temporal mode optimization findings (M17)

[realityinspector]

Summary

Temporal cluster (M17: Modal causality) autoresearch sweep — 30 dry-run iterations each on two templates, exploring all 7 temporal parameters.

mars_mission_portal (seed 300)

Pareto frontier: 3 configs, kept 7/30

Run ID	Quality	Cost	Causal Resolution
dry_c097d4225f31	0.8521	$0.1199	0.7098
dry_ef9ee8e60028	0.8803	$0.1200	0.7679
dry_e65a85511c48	0.8933	$0.1354	0.7685

• Best quality: dry_e65a85511c48 (q=0.8933) — branching mode, low dramatic tension (0.34), minimal foreshadowing (0.15), high prophecy accuracy (0.86)
• Best efficiency: dry_ef9ee8e60028 (eff=7.34) — cyclical mode, cycle_length=15, high destiny_weight (0.90)

castaway_colony_branching (seed 400)

Pareto frontier: 6 configs, kept 11/30

Run ID	Quality	Cost	Causal Resolution
dry_149e73994f9b	0.6530	$0.1021	0.3713
dry_a6f328ac3e82	0.8025	$0.1051	0.6174
dry_adbb9f915626	0.8127	$0.1288	0.6140
dry_e9fd0250ac93	0.8726	$0.1319	0.7422
dry_d6290b67f726	0.8748	$0.1675	0.7666
dry_ef4025fad19a	0.8905	$0.1741	0.7930

• Best quality: dry_ef4025fad19a (q=0.8905) — branching mode, moderate tension (0.36), high foreshadowing (0.55), very high prophecy accuracy (0.91)
• Best efficiency: dry_a6f328ac3e82 (eff=7.63) — portal mode, moderate tension (0.60), cycle_length=13, destiny_weight=0.73

Key findings

1. branching mode produces the highest quality scores across both templates (q=0.89 on both), suggesting it is the strongest temporal mode for narrative coherence.
2. cyclical mode dominates cost-efficiency on mars_mission_portal (eff=7.34), with high destiny_weight (0.90) being the key driver.
3. portal mode is the efficiency winner on castaway (eff=7.63), indicating template-dependent mode selection matters.
4. High prophecy_accuracy (>0.85) consistently correlates with top quality — the simulation benefits from deterministic temporal anchoring.
5. Low foreshadowing_probability favors mars_mission_portal while higher foreshadowing favors castaway_colony — branching narratives benefit from more narrative setup.
6. Dramatic tension has a non-obvious sweet spot around 0.34-0.60; extreme values in either direction hurt quality.

Recommended defaults for temporal cluster

temporal_mode:
  active_mode: branching          # best quality across templates
  directorial:
    dramatic_tension: 0.35        # low-moderate sweet spot
    foreshadowing_probability: 0.35  # template-dependent, use midpoint
    coincidence_boost_factor: 1.5    # moderate boost
  cyclical:
    cycle_length: 10              # midpoint of effective range (5-15)
    prophecy_accuracy: 0.87       # high for quality
    destiny_weight: 0.55          # moderate

Next steps

• Live-run validation of top 3 Pareto configs per template
• Cross-template sweep with branching mode locked
• Interaction effects with fidelity cluster (M6) params

Files

Branch: autoresearch/pro/temporal

• autoresearch/results/dry_run_temporal_20260316_083326.jsonl (mars_mission_portal)
• autoresearch/results/pareto_temporal_20260316_083326.json (mars_mission_portal)
• autoresearch/results/dry_run_temporal_20260316_083331.jsonl (castaway_colony_branching)
• autoresearch/results/pareto_temporal_20260316_083331.json (castaway_colony_branching)