Ways to build a map
I know this question sounds noobish, but what are goods ways to start a map (like what do you first do when you make a map.)
(I kind of get stressed when I'm making a map that uses the destroyed or overgrown theme, since it uses so much
I've seen Valve maps, 2 elevators, a box in the middle made of a bts texture, displacements, hints, square beams, panels, ALOT of logic_relays, hints. They seem so perfect and beautiful in the game, but in hammer, it looks like its poorly done.)
I usually start planning, and then perfecting my concepts until I don't see there are any more unwanted solutions, and I know what is the wanted solution, of course - this is the most time consuming part. If you're a student I suggest going through this phase in one of the boring classes 
Also, Skotty once told me that when working on a normal puzzle, you should first place the major test elements and build the room to match the shape of the puzzle, and when you're working on a 'behind the scenes' map you should first build the surrounding environment and try to 'squeeze' the puzzle in it.
If you use a theme that always uses the same 'base'(for example in Old Aperture maps a testing sphere with toxic goo, an elevator and the skybox set up) it's nice to save that for yourself so you need to do fewer 'routine tasks' on future maps.
Judging from my own mistakes, from such a base it's best to start making your puzzle A. working and B. free of exploits, without worrying about the ugly mess of same-textured brushes you create. On my previous maps (of which I've only released one) I started focusing on getting the theme right before even knowing how the puzzle would exactly look in the end. This resulted in putting far too much time in revising my maps to still make a bit of a challenge, and in the end it was just a visually pleasing map with an average at best puzzle.
Because I find puzzle making the most difficult part of mapping I work with this strategy in my current map before even thinking of the visuals:
-Play it, write down EVERYTHING that goes through your mind that would possibly make a good puzzle or allow players to exploit
-Put the useful stuff in your map based on the notes you made to yourself.
-Repeat until you're happy
So what I learned is that mapping can be a lot more fun if your #1 priority is to make a good puzzle and then start worrying about how messy everything looks, add lights, detail props etc.
If you set the build options in Hammer to the fastest possible(set RAD and VIS to fast, disable HDR) and keep Portal 2 in the background(windowed) you should be able to test your puzzle within a minute.
Just like the people before me:
step one: Planning
after I've planned my map I start creating it, always in clean style, 32x32 size textures* and fullbright. That way 1 block of 32x32 texture = one square in hammer. Then I add test elements, white surfaces.
After I layed-out my design I test it, change it where required.
Then I let a friend play the map with the instruction: "try to avoid the intended solution". I then see how he solves it and see if his solution is more fun then mine. If that is the case I remove my own solution and add his. Then once again there's a round of playing, improvement and ways to break my maps.
If the map is working properly and there are no ways to break the puzzles I start detailing. Once I finish detailing I once again play it, then usually taking screenshots of every tiny pixel that isn't on the right spot. I also check if the detailing didn't actually destroy my puzzle. once that's all validated I do a final full compile (LDR+HDR) check if there aren't any lighting problems and THEN the map is finished...
Hi,
Yep, just as above.
I started out with just a plain blank square room, and thought of the route through the room and the puzzles to put in first.
Then after the puzzles are done, I then worked on textures, splitting up brushes and making them more 'shapely', props and stuff.
Then lighting / shadows.
And finally sort out func_details, nodraw, skip/hint blocks, areaportals and performance work.
I try do some func_details as I'm going, as it can have a major effects on compile times also.
I'm close to finishing my first pak soon (I hope)... and I've learnt alot through doing this.
One thing I've learnt to do in the future for other new maps would be to space things out better from the beginning as I quite often ran out of space and had to go back an 're-factor' areas.
Spreading out individual rooms in a map with more distance and give myself more room to play with in the future when I need to change / add stuff.
Also, separate individual rooms by longer corridors and smaller rooms so that its easier to control Env_projectedtexture / shadows / lighting etc with triggers, it makes it easier to do the lighting/shadow parts at the end.
It also helps with control of visleafs and performance.
Rob.
1. All nodraw box rooms
2. Simple textures
3. Add basic gameplay
4. Get playtesters to test
5. Revise basic gameplay to final gameplay
6. Make it look pretty
[spoiler][SP] Alternate[/spoiler]
I personally start with pencil and paper, then model the room in dev textures. I've found it easier to lay the walls out in 128 unit tall sections (4 tiles).After laying out the architecture map, I put in the testing elements in, and fiddle with dimensions of the room to get them to feel comfortable. During this I also mark which walls will be portalable.
After all this, I being using the clipping tool on the walls for detailed metal walls, adding models/observation rooms, etc. to make it all look nice. At the end, I playtest a whole bunch and see how I can break my map, and fix it. Then release.
Note that I did not stick getting playtesters in this process. I do playtesting whenever.
I generally draw puzzle , shape of chamber , design ,elements and steps to solve it.Obviously I always want to build an unknown puzzle , that confuses player.After I draw a example template ,I export it to another place with well-designed , clean drawed look.And I never forget that puzzle of a map is root of it.When I finish my drawings , simply i start to use measure texture and build it with 16 brush size.I apply best textures on them.Play it and see the puzzle works.If puzzle isnt work , fix it.After skipped that puzzle part , I start to design map all over with lighting, props , instances , observation rooms , clipping walls , etc. After I playtest it again, I publish the map file into TWP forums.So thats how i plan myself to build a map. 
For building: I tend to start with a 512x512x512 box covered in either nodraw or dev textures (gray floor/ceiling and orange walls). From there, I expand various sections out of the box until I get a shape that seems right.
For making the puzzle: I keep a list of ideas and add/remove/edit ideas and select whichever one I like best at the time. This doesn't always work as I might come up with something else and work on that instead or decide to mash more than one together. Sometimes what I want to do is too complicated for the moment. When that happens, I leave the idea alone and let my subconscious do its thing.
Once I get a basic form of the idea made (after x-number of rounds of testing depending on the idea), I build the rest of the map around it. That's not to say I put it smack dab in the middle of the aforementioned box, but I put it where I think it fits best (probably not very helpful to you). Usually at this point I have a fair idea of what I want the map to look like and use that as my guide.
[stream of consciousness]
For example: I wanted to make a map based off of Donkey Kong's Barrel Canyon. This, naturally, led me to diversity vents. Vents are usually seen in Clean, BTS, or reconstructing themes. Doesn't mean it can't be in destroyed, but unlikely. Moving vactubes around will take space, make the room bigger and not as square to keep it relatively small. Just expand enough to move things around. Vactubes too cumbersome to work with, use flanges or connectors.
Flanges better: Bigger, can fit player, harder to hide. No collision model. Make one out of brushes. Want flanges to move up/down left/right like in DK. Turn brush collision into func_tank to double its function. Push or Catapult? Push takes time to build up and unsuited for precision. Catapult can be targeted and applies velocity instantly. I need to stop the player once they enter, add togglable clip brushes. Player can't have too much control and too easy to clip into func_tank. Change to movelinear. Catapult needs to be triggered. Use game_ui.
Having portal gun with game_ui is problematic. Particle system error spews in console when gun is hidden and fires when not. No portalable surfaces allowed. Clean is too bland. Why are flanges exposed? Use destroyed. Don't want map to be too long, 3 jumps good enough. Up/down, left/right, and carousel. Vertical carousel causes clipping issues, make horizontal. Destroyed okay, but not quite right. Use reconstructing. Why jump when you can walk? Use toxic slime. Flange firing system will be used multiple times, turn into instance. Third jump over long distance too bland, need dangerous element...
[end stream]
That's kind of how I made map 1 of Test Map Pack 4 and how I've been mapping lately. There have been instances where I know exactly what I want and build it on the spot.
I have a video of me making the moving button map from Pack 3 if you want to watch it. It was recorded in 15 minute increments due to limitations of the recording software, so I had time to ponder what I was doing as it compiled. It isn't a perfect example of how I map, but it is fairly close.