relying on feedback

[HardboiledGargoyle]

We sometimes say here on tavern that you can just try something out in action, and that is sufficient, no need to bother with anything more.
This is a good position to take with other games, that have things like:

• real-time (no discrepancy between frequency of actions taken and "game-time" AKA "monster-time")
• audio-visual feedback (like hearing a "thwak" or "swoosh", seeing where your bullets fly, not needing to scan messages)
• low variance (e.g. it always takes 3-5 bullets to kill a zombie, koopas always deal 1 damage)
• some variables are held contant (e.g. every monster has human durability)
• no confounding factors (e.g. a knife is equally deadly no matter who wields it, it doesn't look at stats/skills, it doesn't trigger headbutts, all knives are the same, etc.)
• few options (e.g. there are only 7 weapons, of which 4 obviously suck)
• strong contrast (e.g. uzi and plasma rifle)
• save games and reproducibility (if something doesn't work you can load and try something else)
• feedback has zero impact on your play (e.g. you must kill a monster and whether you succeed is only a matter of movement and timing)

Crawl is lacking in these things. Feedback is unreliable and hard to collect and interpret, yet important.
Thus IMO Crawl should move away from model where you rely on testing and prior experience.
For example, instead of showing specific damage outcomes in the message bar, it's better to put overall damage as a stat somewhere, even if it's a bunch of !s.