The story of Studies — the blog's quietest corner (and what it hides)

One post out of 78 — and the discovery it hides

On the day LifeLog had 78 posts, the Studies section had exactly 1: the Ollama truncation loop (ollama-truncation-loop-4096). One post, two months of journey. Any reader would look and think: “studies? this blog doesn’t study.”

Nothing could be further from the truth. It was in the studies that TatuEngine was born — wave field theory, Mamba SSM, the BitMamba-2 1B — and it was by studying that Ollama revealed the bug that cost 2h of debugging and became a permanent lesson. The blog’s most neglected section is the one that produces the most content… it just didn’t publish it.

Context — why Studies fell behind

LifeLog was born to document the journey. Each project had its voice: Arachne, Dogwalk, Capivara, Portifólio, LifeLog. The studies — paper reading, concept tests, discarded experiments — were orphaned:

• They weren’t a project — they were the raw material of projects. A read paper became code in TatuEngine, not a post.
• They felt “unfinished” — a concept test that didn’t work felt like shame, not content.
• They fell into oblivion — what was studied in June was already “obvious” in August. The process disappeared.

The result: when the posting pipeline looked at the grid, Studies was always last in line. Because nobody knew what to publish from it.

The struggle — what the studies actually produced

The truncation loop (the first lesson)

The only studies post existed because the pain was too big to ignore: Hermes entered a continuation loop 4x in a row, every response “incomplete”, 2h of debugging. The cause was in the Ollama documentation, but in a corner nobody reads:

model num_ctx ≠ slot num_ctx

The qwen3.5:4b supports 262,144 tokens of context — but the inference slot allocates only 4096 by default. The model truncates the prompt without warning on stdout. The fix: OLLAMA_CONTEXT_LENGTH=32768 in the environment. A 30-minute documentation study saved hours of debugging — and became a rule: Docs First.

Wave field theory (the study that became a project)

While project posts counted features, the studies accumulated the theoretical side: wave-based phase field theory — a way to model information as waves in a field, instead of discrete vectors. Paper read, notes in the vault, concept tests in notebooks.

That study gave birth to TatuEngine, which became the BitMamba-2 1B model: an SSM (State Space Model) based on Mamba, trained with local GPU inference. The study became code, the code became a model, and the model became the ecosystem’s most ambitious project — with a security sandbox and an autopoietic agent.

The experiments that didn’t work out (what nobody saw)

Not everything the studies produced became a project. There was an attention architecture that didn’t converge, an FTS5 concept test that was shelved, an embedding approach the Arachne RAG rejected after benchmarking. None became a post — and that was the mistake.

A documented discarded experiment is worth more than a forgotten successful one. It stops the next project from walking the dead path again.

Resolution — why publishing now is worth it

The turning point was realizing the Studies section doesn’t need a “perfect” post — it needs an honest record of the process:

1. Study is process, not result — the Ollama post isn’t about the bug, it’s about how you study documentation when everything seems to be working.
2. The discarded is also content — noting the dead path stops it from being walked again.
3. Study feeds everything — wave field theory → TatuEngine; Mamba → BitMamba-2; num_ctx → Docs First rule. Publishing the study is publishing the root of the projects.

Metrics

Learnings

1. Neglected section ≠ idle area — Studies had 1 post and produced the ecosystem’s most ambitious project. The blog didn’t reflect reality.
2. Documented study is worth double — what was learned studying Ollama’s docs became a rule that crosses every project.
3. The process is the content — you don’t need the final result to publish; the path to it is already the story.
4. Every project has a root in studies — publishing the root gives depth to feature posts.

╔══════════════════════════════════════╗
║  Samuel Medeiros                    ║
║  Senior Software Engineer           ║
║  Stack: Python · TypeScript · Rust  ║
║  Projetos: Arachne, Dogwalk,        ║
║            Capivara, TatuEngine      ║
╚══════════════════════════════════════╝