How much does delayed news change an agent decision?

News arrival delay is a testable intervention. We propose releasing the same saved story to otherwise matched agent replays at different receipt times, then measuring when forecasts and intended actions change. The design is unrun and supports no present claim about the value of faster news.

The continuous record companion describes a historical system whose schedules and triggers shaped when agents reviewed markets. That motivates asking whether information timing has value beyond the timing of the next review. The controls paper companion supplies a trace structure in which the received evidence, proposed action and eventual reconciliation can be distinguished.

A story with four clocks

In an illustrative fixture, a story is published at 10:00:00, reaches the collector at 10:00:02 and is assigned to a decision opportunity at 10:00:10. Three arms expose it at 10:00:02, 10:00:12 and 10:00:32. At the 10:00:10 decision, only the first arm can use it. At a second fixed decision at 10:00:40, all arms can use the story, provided its correction and expiry rules permit continued use.

This prevents the delayed arm from quietly receiving an extra decision later. Each arm gets the same scheduled opportunities and the same point-in-time market snapshots. The collector receipt time remains unchanged; the experimental delivery time is a separate field. A correction published at 10:00:25 must follow its own assigned availability rule, rather than being bundled into the original story for convenience.

Use eligible stories chosen before outcomes are inspected. Include routine stories and stories whose market response is small. Selecting only news followed by a large move would measure performance on a hindsight-filtered population. Group repeated articles about one underlying event so an agent cannot recover the withheld fact through a duplicate headline.

Separate the channels of improvement

First score factual uptake: did the model recognize the newly available fact and preserve its source? Then score a fixed-horizon forecast. Finally apply one frozen decision policy to the intended actions, with execution latency and costs held common. Faster delivery could improve factual uptake while leaving forecasts unchanged. It could also produce earlier actions whose additional spread or reversal exposure erases the information benefit.

For each event, report paired differences across delays and retain abstentions, failures and missed reviews. Events share market conditions, so uncertainty should respect event groups and time clusters rather than treating every repeated review as fresh evidence. We would publish a delay-response curve with the number of eligible events behind each point. Its practical output is a measured sensitivity to receipt timing in the tested harness. A decision about faster infrastructure would then require comparing that sensitivity with the actual cost of acquiring and operating the feed.