WCP Error Trends Reporting · AI Pro Writers Studio

WCP Error Trends Reporting

I was looking at the Error logs for WordCrafter.Pro this morning and I'm seeing some trends in the last several days. If you are having issues with errors let me know. I built this back side log so I can be a little proactive when folks are having issues but not saying anything. Sometimes it's the program, sometimes it's the ai model and the provider. And yes sometimes it's the guy sitting behind the coding console listening to his buddies Claude or Gemini gas lighting him.

"I can't fix it if I don't know it's broke" was a sign in an early shop I worked in and I've never forgotten it.

I don't have an easy way to pull how much $$ you have left on an api key but I can try to make the error messages you see make more sense. Something like "Hey add more $$ to your openai key"

I renew the list of models about once a week to cover anything new from our providers and I have a couple of tests I run to see if they work in WCP before you see them. The failures get deactivated but some of them slip through (sigh..more work developing better testing).

NONE of the gpt5 models passed this morning. Not sure what has changed on their side or this one so I will do more testing.

Most notably I'm seeing failures in the reasoning and higher end models mostly. Fable, gpt 5.6 dol, luna and weirdly gpt-4o-mini. They are blowing the context window for the OUTPUT call. So the reasoning it is doing is taking a huge amount of tokens and not giving a full answer which makes me wonder if this is by design to make openai more money costing us for the reasoning to get better answers but not really.

I'm thinking also that the increased guard rails on the system prompts that these models use before we send our prompts through are stealing some of our context windows. The system prompt for Anthropic's Claude Fable 5 is approximately 120,000 characters long (roughly 30,000 tokens consumed before the user types a single word). This is why they have started caching repeated prompts not to save us money, but to save face when they have to put stronger rules in place that takes up what we have to use.

I've seen this form other groups and coders as well. I'm rethinking and working with the skill rooms to try to tighten up the responses there to save tokens and give us a bigger context window. If you want to see what I am talking about try using deepseek, because I know what its reasoning model looks like and it has gaslit one of our users before as well. If you use it with one of the skill room's in WCP it will show it's reasoning "I think the user wants me to do this..." and then it will answer and hide the reasoning in the output panel under a toggle. You can click this toggle and read it to see how it thinks.

I've never liked openai as a company, never cared for the answers from chatgpt, and now this has me wondering more. The images from gpt2 are great, but always grainy, dark, and very wordy. As far as AI tells go these are easy to pick out, but as I said the imagery is pretty great at the same time.

Is anyone interested in my adding any other api key providers to the list? I have one I'm looking at that might be a little cheaper than open router across some models. If you know provider I should be looking at let me know.

One last semi-unrelated item, because if I did not throw a tangent in what would you think of me? Has anyone else noticed this? Because I've been grabbing the free book deals for years I get a couple hundred authors updates weekly. More and more are using (at least) Ai generated covers more than traditional artists. I'm only noticing it really because I've been following some of these folks since before Ai covers were a thing. So my question is with all the Ai haters out there "It's got an Ai cover I'd never buy that" is it the authors who are creating this or are they still paying artists "I'm a real artist I'd never use AI" (but almost all graphics tools have some ai built in now) to create these for them and the artists are using ai more or are they doing it themselves?

I also take suggestions and hints

9 comments

WCP Error Trends Reporting