BTC cycle bottom: $40,600 to $50,700, 26 Sep to 31 Oct 2026. Low so far $58,525 and plenty reckon that was it. In which case I've missed the lot. My quant tool says lower. Quant's generous. Posting before the fact so it can be marked after. Egg on my face or a grin ;p

man its sunday and im feeling spiritual so i wrote an article about 1. this whole renaming ai to “super intelligence” and 2. anthropic trying to convince religious leaders that ai has a soul & moral standing this feels important lmk what you think tej.as/blog/super-intelligen…

If rolling back a release means undoing unrelated work, the batch is too big. Ship smaller batches more often. If one fails, there are fewer changes to trace and a clearer path back.

- we removed unit tests in favor integration/functional tests - ci/cd invoice exploded - we removed cicd, it was slowing down our process - now our agents deploys our code - agent decided that database is going to be much faster without users future looks awesome ☠️

A weekly RSI line of 32 would have missed Bitcoin's 2011 bear bottom. It reached 33.6 then. The next three lows were 25.8, 29.0 and 30.5, so the flash line is now 34.

Three teams can own separate services and still wait on one release decision. That shared deploy path turns independent changes into one queue. Give each service its own deploy path so unrelated changes can ship on their own.

Theory: Development teams are going away. We will increasingly work alone instead of on teams. Here’s why: Historically, writing software was slow, expensive, and required highly specialized knowledge about languages, libraries, and syntax. So, we grouped developers into teams to achieve sufficient velocity, and support specialization. Now, one dev can orchestrate multiple agents in loops to achieve an entire team’s output. No team required. Of course, lone professionals aren’t a new idea. Many industries often work alone. Doctors join a practice, but often see patents alone. Attorneys join a firm, but often try cases alone. Mechanics join a shop, but often service vehicles alone. Developers will soon be the same. We’ll join a company, but often build and manage entire apps alone. We’ll still interact with each other, but we will be more like doctors - we’ll advise and assist each other, but we won’t work together daily on the same code. Our new job is to create and manage “software factories” - agents that generate the software for us. You don’t need a team of devs to do that. So, software development isn’t dead. But the idea of multiple developers working on the same code is soon going away.

The holder gap has one measured Bitcoin bottom: -0.124 in 2022. With no earlier trough to compare, that reading cannot tell us what a typical bottom looks like.

Review the change when it's made. By release time, the checks have already run. A manual gate that rechecks the same evidence adds a queue, not a new control. What can that gate catch that the earlier checks cannot? If we cannot name it, improve the checks and remove the wait.

Bitcoin's NUPL was negative at all four measured bear bottoms. The lows rose from -1.585 in 2011 to -0.326 in 2022. That shift makes one fixed bottom threshold a poor fit for the whole series.

A frontend change can wait on a backend change, which waits on a tightly coupled microservice. Each team can finish its own work while the release still sits between them. Reduce those handoffs and the whole change gets through sooner.

Bitcoin's 2025 MVRV Z top was 3.35. In 2013 it was 10.66. All four recorded tops stepped down. The bottoms moved up from -0.69 to -0.36. I wouldn't use one fixed MVRV Z threshold across those cycles.

A wall of confusion can end in a manual release gate. Batches grow, releases become big bangs, and people get wary. Another gate goes in without making the release any smaller. The DevOps Handbook's First Way is about flow. I'd tackle the wall first.

A backlog ticket can outlive the facts it was written from. Check the current code and its dependencies, keep only the outcome we still need, then estimate that.

People use Jev to pick a model before a task. I made it change GPT-6's reasoning effort inside Codex DURING the task. More thinking when stuck. Less for routine steps. 50% lower Astra costs in my tests. Faster runs, without breaking prompt caching.

Re-tested Jev with jev-pruner, rather than rewriting thread history. On noisy build output, it cut 253,391 chars to 99,660 in 959ms, kept the final test result and archived stdout/stderr exactly. It is useful output pruning. Native compaction still owns session memory.