I am very skeptical that, for example, a flavor bias of 6 generally means three times as much of an effect as a 2!
You can have a look at Civ5's source code yourself if you don't believe me: it's accessible via the Civ5 SDK, though I do warn you that a lot of the AI-related bits are written rather obtusely, to put it lightly.
Almost every time a flavor value or a boldness value is called, it is either straight-up multiplied by a constant or added to other flavor values before being multiplied by a constant. It's simply a matter a mathematics that a flavor bias of 6 will produce numbers that are three times larger than the numbers a flavor bias of 2 would produce. For example, the Early Expansion economic strategy only fires if, among other things, the AI's city count is less than (baseline city count)*(expansion flavor)/(growth flavor)*(tile count)/(default map size); with starting expansion and growth biases of 5, that ratio can range from 2.33 (7/3) to 0.43 (3/7), which is the difference between 18 desired early cities and 3 desired early cities with the default baseline of 8 cities. When settling a new city, the potential yields received always get multiplied by the civ's flavor value for that yield, so one potential gold from a newly settled city is counted as six gold with a gold flavor bias of 6, but a gold flavor bias of 2 would make it count as two gold.
Some things people have found explicit code for, e.g., nuking or not. That kind game mechanic, which uses RNG against the run-time flavor value, might or might not be used elsewhere. With this example, an AI with a nuke value of 6 is much, much more likely to nuke than an AI with a nuke value of 2. And this is pretty much the most transparent example of flavor values impacting the game. Everything else is pretty opaque.
It's not opaque at all, it oftentimes is just really hard to find in the jumbled mess of AI routines that is the Civ5 source code.
It just so happens that I stumbled upon the algorithm for when the AI launches nukes a few months back while I was tweaking nuclear missile target scoring for my mod: each turn that the AI is losing any wars and has a nuke, it will roll two random numbers between 0 and 9 inclusive, and if both numbers are less than its use nuke flavor, it launches a nuke. The exact probabilities are 100%, 81%, 64%, 49%, 36%, 25%, 16%, 9%, 4%, 1%, and 0% for flavor values of 10+, 9, 8, 7, 6, 5, 4, 3, 2, 1, and 0. This means that if an AI has starting use nuke flavor of 5, their actual value could range from 7 (49% chance to use nuke each turn) to 3 (9% chance to use nuke each turn), so yeah, a range of +/-2 can be huge.
My point is that, for all anyone knows, a run-time Boldness value of 6 might result in dramatically different game behavior from a Boldness value of 5.
Boldness is a weird one, as it's baked into all sorts of decisions. It's also quite easy to find all locations it's used, you just need to find all references to the CvDiplomacyAI::GetBoldness() function. All examples of where it's used in the unmodded game: multiplied by 10 as the base chance of an AI giving into a demand, as the base flavor for scoring of the Conquest grand strategy (which can control quite a lot, given how often the active grand strategy is checked for in AI routines ranging from what free GP to select to whether or not a World Congress proposition is beneficial), as the extra amount of military units the AI wants in its empire if it's running the Conquest grand strategy, whether the AI wants to settle near enemy cities (boldness greater than 7 increases the fertility score of tiles that are at most 5 tiles away from another major's city by 50%, boldness less than 4 decreases the fertility score of tiles that are at most 5 tiles away by up to 75%).
- and yes i lost a battleship to a spearman back in the day
