Test Captaincy: Declarations, the Follow-On and Reviews Policy & Terms

Batting first is the default in Test cricket, and the AI knows it. The exception is the classic bowl-first call: a genuinely green, seaming surface with morning moisture. The captain reads this match'sprepared pitch, not the venue's reputation, so an unusually grassy Lord's tempts a bowl while a flat road is an almost automatic bat. On a typical surface, bat.

When should a captain close the innings? Mined from thousands of real first innings, the answer surprised us: captains bat far longer than intuition suggests. With real batting still possible, genuine declarations cluster around a 510-520 total regardless of era, and even the aggressive quarter still bat past 390. The AI reflects that: it will not declare a strong first innings early just because it has passed 400. Time pressure on days three and four lowers the bar, and an exposed tail lowers it too, because there is little to gain batting the last pair on.

The third-innings declaration is a different calculation: balance a safe lead against the overs left to bowl the other side out. The AI aims to set a target that would be a hard chase rather than a gettable one, holds out for a genuinely safe lead, and gets more willing to pull out on the final day as time runs short. Test captains do not leave gettable targets, and neither does the AI.

When a side is bowled out far enough behind, the fielding captain can send them straight back in. Modern captains enforce it only about half the time it is available, and the AI starts there, then adjusts for the two things that actually drive the call. Time is the big one: late in the match, with the clock the enemy, it enforces far more often to buy an extra innings of bowling. A tired attack pushes the other way, because enforcing means bowling two long innings back to back with no rest. There is a dedicated guide to the follow-on rule itself.

In the modern era each side gets three unsuccessful reviews per innings, and they are a resource to manage, not spend freely. The AI reviews a strong shout readily and a weak one rarely, guards its last review more jealously, and is a little more willing to gamble a fielding review late in the match when it is hunting a result. Because the umpiring-error layer means a review is never a guaranteed wicket, this is a genuine judgement, the same one you make when the prompt comes to you.

Between decisions, the captain also sets the tempo. Chasing, the AI compares the required rate to a gettable clip and shuts up shop for the draw when the target is out of reach, because nobody slogs at a Test target. In the field it crowds the bat when the tail is in or the batting side is blocking for a draw, and sits back to squeeze when defending a small fourth-innings total.

Take the armband yourself in The Follow-On, or study how real captains handled these calls across 150 years in the Test archive.