Daily Cap on Local Humanization & Tbot Fixes — September 12, 2026

Today, I focused on implementing a hard daily cap on the humanization process for our local pages to manage costs effectively while also addressing some significant issues within the tbot functionality. After a long 8-hour day, I successfully shipped three important fixes that are pivotal for maintaining the performance and sustainability of my project.

The first challenge I tackled was the excessive costs associated with the /local humanization process. My initial setup had allowed for a staggering number of calls that were draining resources—around 30,000 calls per day, which translated to about $32 daily. While I’m okay with a weekly budget of $64, I certainly can't sustain $32 a day. To rectify this, I implemented a hard daily cap that stops the per-render re-humanization. The caching fix I had previously introduced was meant to alleviate this, but with over 100,000 pages on the /local surface, crawlers were still hitting new uncached pages, causing the one-time warm-up to run rampant. This fix is crucial not just for cost control but also for ensuring that crawlers can navigate the site more efficiently.

In the realm of tbot, I faced a different but equally pressing issue. The performance of the trading bot had been less than ideal, with a significant financial bleed from entries that were not yielding profitable results. Specifically, the lifetime realized showed a profit of $7.51 from majors, while memecoin scans resulted in a loss of $78.63 over 23 wins and 61 losses. The scan lanes were clearly the source of loss, so I made the strategic decision to disable the pickFresh and pickMomentum features unless the TRADING_BOT_MAJORS_ONLY flag is set to zero. This means that the entries will now focus solely on majors, and the bot will utilize a dollar-cost averaging (DCA) strategy to top up its position. This fix should help stabilize the bot’s performance and, hopefully, turn around its profitability.

Lastly, I addressed the problem of per-render re-humanization for local pages, which was generating approximately 35,000 Anthropic API calls per day. The humanizeCached function was racing against humanizeInhouse, but due to a 4-second timeout, the whole race was being rejected. This was happening on the /local render path, which includes an introduction and three FAQs per page. Since the local surface sees about 73% of crawler traffic, this fix is essential for improving the overall efficiency of the site. By stopping the per-render re-humanization, I can ensure that the rendering process is more streamlined and less resource-intensive.

Looking back at the day, I’m proud of the progress made but also acutely aware of the grind that comes with these fixes. Each of these adjustments required careful consideration and testing. It’s a delicate balance—optimizing performance while keeping costs in check. Working solo means I have to wear many hats, from developer to project manager, and while AI tools like Claude Code help, the final decisions and implementations are still on me.

As I move forward, I’ll keep refining these processes, especially as I work toward my goal of building a one-man-show company with a billion-dollar valuation. Each fix, each day, brings me closer to that vision. The challenges I faced today are just part of the journey, and I’m learning how to navigate them with increasing skill and insight. The agentic web is a complex landscape, but I’m committed to making it work, one fix at a time.