I dropped the BMD kill date and named it flagship

I officially dropped the /bmd kill date and named BMD the flagship software on 2026-09-21.

This decision required updates to the cannon, BMD Desktop, bmdpat, and pricing pages. I am moving away from arbitrary thresholds. The blog will now use a search-led model where dynamic GSC missed-click scoring replaces hard CTR thresholds. I have 0 confidence that the current security scan is actually working. The SecurityAnalyst returned a GREEN verdict with 0 P0 issues, but Gitleaks scanned 0 of 5 repos because the K: volume has been unmounted for 26 days. A check cannot verify what it cannot reach.

Why did my vault health check report so many stale tasks?

The vault health check flagged 92 of 104 root tasks as stale and 35 tasks as stalled for 14+ days because the check only reads folder roots. It does not recurse into subdirectories to find active work.

I found that several drafts were sitting in the Outbox past their 3-day TTL on 2026-09-21. I need to update the crawler logic. I also noticed the VaultAgent-GapResearch NotBefore date of 2026-09-22 closes a five-day Doctor blind window. This is a scheduling collision I have to fix.

What did I ship in bmd-desktop and bmdpat?

I shipped bmd-desktop version 3.47.15, which includes home polish, terminal fit, agent tabs, and build simplification. I also enabled the ability to send reports without a GitHub account via POST /api/bmd/feedback.

The following changes were merged on 2026-09-21:

I also ran a quant ladder on the 5090. It measured Qwen3.8-27B (lmstudio-community) at Q4_K_M, Q6_K, and Q8_0 across 9 sweep rows in the desk corpus. If you are interested in how these weights affect performance, you can read my post on GGUF quantization explained or use my GGUF compare tool. This is useful for anyone running local LLM inference on consumer GPUs.

Who was I on 2026-09-21?

I killed a date instead of a product. The /bmd door carried a kill date of October 1 and targets I set myself, but on the morning of 2026-09-21 I saw that the signed installer only became public on 2026-09-20. A 14-day trial that starts now cannot finish before that date. So there was no result waiting at the end of the test, only a deadline. I dropped the date and said BMD is our flagship software. That is the opposite of the move I usually make, and it comes from the same place. When a measurement stops measuring, I stop trusting it.

Then I shipped all day. Three bmdpat merges, two desktop merges, a signed 3.47.15 a stranger can download, a Docker CI runner on the 5090 box, and a quant ladder that put three quants of one 27B model inside the same error bar.

And I did not plug in the K: disk. Day 26. The brief asked me for the 27th time, and my own security agent reported GREEN after reading zero of five repositories. I can take a gate off a product I believe in within an hour. I let a cable sit for four weeks. Both of those are me.

Accompanying prompt

What the prompt does: Updates blog writer templates to include structured input fields for consistent generation.

Copy/paste this prompt:

Copy-ready prompt

Paste the exact block into your coding agent.

No article chrome, no footnotes, no formatting drift.

Copy promptSelect text

Role: Blog Template Architect Context: I am updating my writing workflow to use a structured Inputs section in every post. Inputs: - Topic: __ - Key Numbers: __ - Broken Items: __ - Shipped Items: __ Task: 1. Create a draft structure for a devlog post. 2. Ensure the first sentence is a plain statement of what happened. 3. Include an H2 for technical details. 4. Use the Inputs provided to populate the body. Output: - A structured markdown post. Constraints: - No preamble or intro text. - Use only the facts provided in the Inputs.

24 lines/539 chars

Ready

This prompt and every other one we publish live in the free prompt library.

Copy the block above.

Weekly measured local runs: https://bmdpat.com/5090-reports