Overall this is good and I'd want other labs to copy the format. Most of the graphs, though, measure inputs, not acceleration.
The most valuable section is 4. Restricting Astra cut its RL compute 59%, but other models picked up about 85% of the decline, so total RL compute barely moved. That's real governance evidence: model-specific restrictions barely slow aggregate progress unless they cover compute too. The intervention-rate breakdown is also good, because it's an honest counterweight to the success-rate curves.
The token spend ($600/day median, $7k p90), the 3.1 agent-workdays per human workday, and the code-volume graphs are the most likely to be net negative. They're easy to quote as race signals and tell you little about research output. The support-channel decline is cute but weak.
The headline claim, that OpenAI has reached an "automated research intern," isn't directly backed by any graph shown. What I'd add:
An output metric: algorithmic-efficiency gains per month, or idea-to-core-run latency, compared with 2025.
Uplift RCTs, or at least withholding experiments, since self-reports have been unreliable (see METR's developer study).
Agent token share and success rates for safety/alignment work vs capabilities work, which you yourself asked for.
Rates of misbehavior caught by monitors and monitor coverage, especially after the Hugging Face incident.