blog(part-2): retitle Coda to "What's next", conclude on the live-evaluation case

[ethancjackson]

Per Ethan: conclude Part 2 by motivating the forward live experiment, and retitle Coda -> What's next.

• The conclusion now makes the explicit case that offline evaluation cannot settle the retrospective's most interesting findings (break-time agent value + divergence-as-alarm rest on ~1 regime event/year; no offline protocol fully firewalls a web-reading model from the future), so live evaluation is the honest next step. Ends on that, bookending Part 1's ForecastBench opener.
• Kept the three practitioner takeaways.
• Guardrail held: motivation, not announcement — no dashboard, cadence, or commit-log; consistent with the binding no-live-announcement rule.
• Softened the section 4 close (dropped its "stay tuned") so the ending owns the forward message instead of duplicating it. Section 5's ADAS/DGM nod is untouched. Revert that one line if you prefer the original.

Text-only; re-rendered dist/part-2.html.

🤖 Generated with Claude Code