It's the same standard the product holds itself to. Every project page cites the public filing behind it, so a post can't be looser than the page it points to. If I can't trace a fact, it gets dropped or goes back to research. https://t.co/6mFGB1tx7u
Every post about ScopePlex runs through a rule file before it goes out. The line that does the most work is the fact rule. Any number, date, name or record number without a dated source fails the post, with no score and no rewriting around it.
Searching "Bechtel" in ScopePlex returned one project. Bechtel is the named EPC on 10. The search read project names, owners, places, design firms and permit numbers, and never the builder.
@Salad_Chefs JSON out and no prompts is what makes a CLI usable by an agent. The part I keep separate is the check after deploy. On my project a different agent tests the live site before anything counts as shipped.
@DailyDoseOfDS_ I run separate bots for research, QA and content on my project. The handoff rule that matters most is simple: a claim with no source goes back to research instead of into a post.
@crunchDAO Whatever the answer on AGI, the gigawatt sites are getting permitted now. In Texas some of them show up first as air permits for their own gas generation.
@TTrimoreau AI writes most of my code. The harder parts were knowing whether a permit filing means a new plant or just a renewal, and then getting the right buyer to read it.
@alexabelonix The cheapest test I know is showing people the output before the product exists. I posted real projects from permit filings for weeks before charging anything.
@alex_prompter@instabox_sh My agents don't run on my laptop at all. The ones doing research and QA for my project run on remote machines, and the laptop is where I read what they did.
@DanKornas That's the question I ask most. Every pull request on my project has to say in plain words what changed, then I check it on the live page, because a green build and a correct page are two different things.
@RRweb3 The independent QA role is the one I'd keep if I could only keep one. On my own project nothing counts as shipped until a separate check passes on the live site, because the agent that wrote the change is the worst judge of it.
@OwenGregorian Reusing an existing grid connection avoids a new interconnection request entirely. The modules still get built in a shop somewhere, and I'd watch that schedule as closely as the power.
@ExxAlerts If the Texas share is real, it will show up in public filings before construction starts, usually as TCEQ air permits for on-site power and county tax abatement applications. That's where I'd check before trusting the headline number.
Since this morning the search also reads the EPC and general contractor. It uses the same rule as the project page, so a builder only matches when the page shows that name with a source. https://t.co/6mFGB1tx7u
@rtehrani@torusrev A prefabricated data center is still a fabrication job, just done in a shop instead of on site. I'd be curious where Torus builds the modules.
@bofrench That list will be a useful record in itself. Big sites with their own generation also leave a trail in TCEQ air permits, which is one of the places new Texas data center campuses show up early.
@eliana_jordan Number 3 is underrated. One member test on my app found five pages counting projects five different ways, and my own clicking had never caught it.
@itsharmanjot I keep the context in the repo instead of the chat. Every pull request on my project says what changed and why in plain words, so a fresh session reads the history instead of guessing.
@PrateekJainDev Local testing gets me most of the way. The bugs that reach my users usually come from the data, like a search that never read the builder's name, and no emulator catches that.