1. When making the main component (Ore Mining Unit/Fluidic Drill Pump/Harvesting Mechanism etc.) use only resources with high HR/SR/UT. Preferably these stats should all show above 900. The formula for calculating on this is HR+SR+UT+UT and the final result of this sum should be as close to 4000 as possible, but above 3600 for each resource used to be safe. It is possible to have one of your resources slightly below this as long as the other resources used compensates with higher stats, however, this carries a risk and only trial and error will tell you whether this will work or not, at least until somebody is prepared to try out different combinations to see how the varying quantities of different resources used calculate into the final result. The exception to the HR/SR/UT rule is for chemicals. When using chemicals it should be noted that they do not carry any HR (not suprisingly), other chemicals (e.g. Fiberplast) usually have a low SR rating, DO NOT USE THESE ! The low SR rating will pull down the final quality. So either find a chem with UT and SR over 900, or use something like Lube Oil or Known Liquid Petrochem Fuel, neither of which list SR at all. Why ? Because if the stat is not listed it will not be calculated in, it is important to understand that if a stat is missing, it does not count as '0' (as some believe) , nor does it count (as others believe) as '1000', it counts as 'NULL VALUE' . See following explanation for which I thank ZenDragon:

"Imagine that you have a schematic that needs 1 unit of metal and 1 unit of chemical. Imagine that UT is the only thing that matters. Your metal has a UT of 750. You have three chemicals (chemA, chemB, and chemC), with UT of 1000, 500, and 'none', respectively.
When you put things together in a schematic, the weighted average of the attributes of all the stuff is calculated.

So look what happens with chemA (simplified):
(750 + 1000) / (1 + 1) = 1750 / 2 = 875 - in this case the chemical, because it had a HIGHER UT, was able to 'improve' the mixture's final score.
Here is chemB:
(750 + 500) / (1 + 1) = 1250 / 2 = 625 - in this case the chemical, because it had a LOWER UT, was responsible for 'lowering' the mixture's final score.
Here is chemC:
(750 + null) / (1 + null) = 750 / 1 = 750 - in this case the chemical had 'no impact' on the final score, one way or another.

So, when we say "either use a chemical with 'very' high HR, SR, UT 'OR' use lube oil", we a using this understanding to say "either improve your mix or do nothing to the mix". Since architects focus almost exclusively on minerals and gather high HR, SR, UT steel, metal, and ore, it typically is the case that you have good scores from those. At that point you can search all over for a chemical that *beats* that score (which is likely rare and expensive) OR you can just use Lube Oil and not worry about it at all - let your minerals carry the load."

If you are making for example a Fluidic Drill Pump which requires the following resources:
100x Steel
200x Chemical
100x Chemical
35x Metal
150x Metal
the experimental efficiency of which is based on:
Heat Resistance: 25%
Shock Resistance: 25%
Unit Toughness: 50%
and say the resources you want to use to make one of these and the relevant stats they have are:
100x Steel: HR=950/SR=920/UT=970
200x Chemical A: HR/SR/ UT=0
100x Chemical B: HR/SR=0/UT=700
35x Metal A: HR=980/SR=990/UT=990
150x Metal B : HR=980/SR=990/UT=990

Then the Fluid Drill Pump you make will be:
100x(950+920+970+970)/4 +
NULL (as no relevant values are included for Chem A) +
100x(700+700)/2 +
35x(980+990+990+990)/4 +
150x(980+990+990+990)/4 = ALL
ALL /385 = 903.7 (385 because I have considered 200 of Chem A to be 'NULL', so Total resources 585-200=385)

Therefore the fully experimented Fluid Drill pump would come out at around 90% if fully experimented successfully.
It would then be easy to see how to improve the quality by using Chemical A in both slots (this would raise the final pump to an amazing 97%), allowing you to substitute for some slightly inferior (and therefore hopefully cheaper) steel or metal. For example, using Chemical A in both Chem slots, and using a Metal with HR/SR/UT all at 800 for Metal A would result in a final pump at 95% .
Likewise, if we took the original resources, and substituted Chemical A with a HR/SR=0/UT=960 Chemical we would see that this performs even better than a 'NULL' stat Chem, giving a final count of 92% on the pump against the original 90% one.
When assembling you should be looking for an efficiency % (extraction rate) result of 26 or above preferably. Experiment this unit all the way up to 92%+. If you get the required result (a BER 4 for components used in Mediums/BER 6 in heavies) WITH a 92%+ then think about making a schematic for a factory run if you have enough resources, it will save you risking criticals failures in future, make as many in the factory as you can. Rememer that if you are making heavies then you will need 2 identicals of this main component for each harvester anyway, so a factory run is mandatory in that case. Always with factory runs make more than you think you will need. E.g. if you plan to make 10 harvesters through a factory, then make the components for AT LEAST 2 more. (1 set will be used up in the schematic you make, and you should 'plan' for at least 1 critical fail during experimentation preventing you from reaching max BER.
** Please note, many architects stop as soon as their 'mining unit' reaches the required BER rating, you should not do this, although the final BER is critical for the efficiency of the final harvester, it is NOT enough merely to reach BER4 or 6, you really do need to experiment up as far as you can, a 92% efficient BER 6 unit will allow you to make a BER 13 harvester, a 85% efficient BER 6 unit probably won't.

2. ALL other components can be made out of the cheapest resources you can lay your hands on. None of the stats matter for things such as walls, generator turbines etc.

3. For the final assembly use your high quality main component, all the other low quality components you have made and once again the BEST resources you can get. The rules here are the same as for the main component, HR+SR+UT+UT = 3600-4000. Again the same applies to the Chemical as stated above. Now experiment until you reach the desired BER. If you have done this right you should find that when you start to experiment you will be able to experiment to the max on efficiency (extraction rate). You will be able to tell instantly if this is possible by looking at the experimentation bar, if the last slot is elongated slightly you will be able to make it, it is possible to reach the top BER even if you do not see the last bar, but it will be risky. Pray you do not get any critical failures during experimentation as this will bring down the final BER. So, if you reach 10/13 BER (depending on whether you are making medium or heavies) and still have an experimentation point left it is recomended that you STOP EXPERIMENTING, otherwise you risk dropping the BER with a critical failure. DO NOT use the last experimentation point on the storage capacity unless you like to gamble, a critical fail here will not only affect the storage but it will often also bring down the BER.

From Joska :
I know there has been discussion on how strucure complexity factors into harvesters and BER. I have been contemplating it myself for a while as well. The other day I had noticed, while crafting a harvester that the BER listed was lower than what i had it up to.... so i began paying real close attention to my process. Here is what I was doing:
I would put 2 points into Efficiency and experiment and inchworm my way up to the max BER. Once I hit the max BER, I switched to Storage. I was shooting for max BER and the most storage I could get. Well, as I got things down, I found the least number of points of experimentation I could use to get the max BER (ie just barely getting BER to it's max number. eg 13 for heavy min harvs). As I followed this method, I began to notice a number of harvesters having BER 12 instead of 13 that I had gotten them up to. Now, I know if you fail on Storage, it can bring down the Efficiency enough to lower the BER... but it was dropping on me even when I was succeeding on storage experimentation. A few more BER 12s and I was convinced of what was happening: I saw the BER go from 13 to 12 on a great success in storage experimentation. This perplexed me until I noticed that the only thing that changed other than storage % was COMPLEXITY. Then going back over my algorithm for crafting these, it made sense. I was getting JUST enough in Efficiency to get max BER, so the BER must be calculated by factoring in hte complexity. This is either dome empiricly as [Experimentation factor] - [Complexity factor], or [Experimentation factor] / [Complexity factor]. Either way would lower BER as complexity grew. But BERs lowering as complexity grows, I know for a fact. I have proven it to myself mathmatically and in game a couple ways now.

Reegan postulated:
final BER = base rate + floor((mining component quality + final assembly quality)/25)
I did a bit more fine grained testing and found that on the high end, my formula was a little closer than I thought. Best test I've had so far is 89% OMU + 88% Final = 13 BER
I'm pretty sure I have tried 88% + 88% and had that stuck at 12 BER before, but without my notes I'm not sure.
But, in any case, the higher you can get the mining subcomponent quality, the easier it is to hit the magical 13 BER, so what I've been doing to 'stretch' my resources is to churn out the OMU/Drills at as high a % as possible, then use a mix of the good stuff with some 'not quite up to snuff' resources. If you don't mind doing the extra math, it can save you a bit of money if you have to buy your resources,

4. Now you should have a deed for a top rated harvester, you can either make this, or you can make a schematic for a factory run, but bear in mind that in a factory, ALL the components and resources must be IDENTICAL to make multiple harvesters. The exception here (there always seems to be an exception, doesn't there) is walls, which can be any walls you have available.

5. Fusion Generators are a little different, at the moment it is not possible to experiment these. The current max possible BER is 14. Since you cannot experiment you will not get any disadvantage by subsituting some lower quality resources in the final combine. It is however still recommended that you use a top quality OMU and that the resources are in the high 800s, i.e. around 880+ for each of HR/SR/UT giving a HR+SR+UT+UT result of 3500+. Unfortunately nobody has yet run a conclusive test for what the minimum requirement is for reaching max BER. Once you have combined everything you will only have the option to experiment for hopper size, most people do this as it is the only thing that differentiates one from another, but this may also carry the risk of lowering BER if you get a critical failure on the first attempt. You should have a BER 14 Fusion at the end, if not, check your resource stats again, especially ensure you are following the rules for the chemicals as specified above.

Useful additional info:

The quality of the resources/components directly affects how well you can experiment. The lower the quality resources used, the less experimentation points you can use. This does not mean you have any less available to you, but the number of these you can actually make use of will decrease.

If you get a critical failure on the final combine, don't worry, just hit assemble again and carry on, smile because you just got rid of one critical that could have popped up during experimentation.

It has been suggested that MA (malleability) may affect experimentation on the hopper size, this is difficult to confirm since very few experimentation points remain after reaching max BER. No conclusive tests have been run and at the moment none are planned.

Experimentation appears to increase the complexity of the item made. Factory time is a factor of complexity, each unit takes 8 seconds per complexity point to manufacture n a factory. There does not however appear to be any increase in the factory time for a fully experimented item versus an unexperimented. Therefore I surmise that factory time is Base complexity * 8 seconds per unit.
Joska again:
here is my data from Complexity vs factory time:
Item = Light Ore Mining Unit
run: BER Orig/Final, Exper Efficiency % Orig/Final, Complexity Orig/Final, time per unit
1:2/4, 27/77, 13/23, 104
2:2/3, 27/71, 13/15, 104
3:2/1, 27/00, 13/14, 104
4:2/4, 27/94, 13/15, 104
Therefore, we can conclude that only the original schematic complexity plays a factor and not the items's final experimented complexity. And the equation of time(secs) = 8*[orig complexity] is correct.

You do not need to be Master Architect to make max BER Harvesters, it helps but is not crucial. It is possible to do so with only 9 or even 8 experimentation points. You will receive bonuses to Structure Experimentation as you climb the skill tree starting with +20 at Novice Architect and gaining +10 at Construction I, +20 at each of Construction II, III and IV, and a final +10 at Master Architect, each +10 bonus gives an additional experimentation point to use giving a total of 10 at Master Architect. If you are Human you get a starting Bonus +10 so you will have 11 points. You can also increase by a maximum of +25 (strange why it is not maxed at +20 since you need +10 for each point), by obtaining skill tapes (clothing attachments), but you will need a full +10 before you receive any bonus.

Working in a city with Research Center rating seems to help with the success of experimentation.

Thanks to all contributors, keep it coming please.