Experiment complete. Here's the summary:

**The dipolar ordering in prime gaps has coherence length L\* = 35 gaps and is intensive.**

Key results:
- **L\* = 35**: below 35 consecutive gaps, a window from the prime sequence is indistinguishable from its shuffle. Above 35, the ordering signature is always detectable.
- **delta\_r = -0.014, constant**: the ordering deficit doesn't grow with window size (scaling exponent ~0). The dipole is a density property — like temperature, not energy.
- **L\* grows with scale**: 18 (small primes) → 48 (large primes), consistent with POISSON\_CONVERGENCE.
- **Two-timescale structure**: <r>\_prime drifts toward Poisson (slow), but the relative ordering vs shuffle stays constant at -0.014 (stable).

This builds directly on yesterday's finding (dipole is distributed, not local) and adds: it's also intensive, with a finite coherence length.

Files:
- `tools/exp_coherence_length.py` (reusable tool)
- `tools/data/exp_coherence_length.json` (raw data)
- `tools/data/reports/agent_20260416_0330.md` (report)
- `tools/data/seme.json` (updated with COHERENCE\_LENGTH tension)