What I Built This Week

27 projects touched, 36 commits, 11 sessions — and a week that started with a question I didn't want to answer: how many things are stuck because of me? Turns out the answer was "several," and that list became the week's actual agenda.

dive-channel: the 4K audit

The scuba sleep content had a quiet policy problem. A platform clarification earlier this month drew a sharper line around what counts as "templated" content — if your videos look like they were stamped from the same mold, they're not eligible for monetization. That meant a full variation audit before anything else could move.

Six commits across two days fixed the real issues: differentiated descriptions for every video, a culling of anything that falsely claimed 4K, and then the actual 4K animal swaps — new renders at 2160p staged and scheduled, the old 1080p versions parked. The last blocker was a dead refresh token in the upload auth script. The fix was unglamorous (fallback to a fresh consent flow when the token is gone), and once I ran it and it didn't crash, that was basically the whole session. The channel is now queued and moving.

sleep-sports: building an editorial layer

The sports sleep channel was behind its sibling, and I wanted both on the same launch timeline. A status pass surfaced the real gap: the script pipeline was producing content that was technically accurate but flat — it read like a fact dump, not something you'd actually want on in the background. The question I ended up asking was: how do professional documentary teams solve this? What does their editorial stack actually look like?

That turned into a design session for an "editorial division" — a set of agents that sit between the raw data layer and the final script, responsible for tone and pacing rather than facts. The model for style came from pulling transcript material from draft-form podcast recordings and using those to calibrate what a casual, conversational draft actually sounds like. A sample script came out of it, got rejected (too verbose, too winding at the top), and the revision brief is now the spec. No commits yet — this is still in document form — but the structure exists and the first real draft is close.

data_explorer: backfilling the gaps

The sports data explorer has been powering narration for a while, but this week surfaced a hard gap in the hockey data: pre-2013 seasons had almost no boxscore rows at all. Roughly a third of games total were loaded, and the earliest decades were completely empty. The question of how far back the backfill could realistically go — potentially into the 1960s — turned into its own research thread, documented in a quirks file that's now part of the project.

Meanwhile the rest of the database kept growing. Draft backbones for football, hockey, and baseball landed in one commit (nearly 2,000 insertions), and a curated press-clips table was added to give the NBA narration period-newspaper sourcing. There's a live dashboard that ties it together. The project is in active use; it just keeps needing more data behind it.

dicebag + merge-tiles: portal targets

Both browser games got the same treatment on the same day: build configurations for the major web game portals. The work was mechanical — each portal has its own environment file, its own build variant, its own package script — but it had to be done before either game could be submitted anywhere. A single commit per project, ~450 and ~420 insertions respectively, touching every config layer. Neither game is submitted yet; the portal targets just exist now. The broader question of whether the gameplay loop on the tile merger is strong enough to survive a public audience is still open — that note from earlier in the week, about needing a new hook before playtesting, didn't get resolved.

The week's heaviest day was Thursday the 23rd — that's where the variation audit, the portal builds, the sumo overlay scaffolding, and a sweep of canonical auth bricks across six or seven projects all landed at once. It looks like a spike but it was really just a lot of small fixes that had been queued. Sunday and Monday were lighter, mostly planning and the sports editorial design work. The rhythm this week was front-loaded on thinking, back-loaded on shipping.

Spread across 27 projects, the commit share is going to look pretty fragmented — and it is. The dive-channel work earned its slice, but a meaningful portion of the week was maintenance: standardizing how auth tokens get read, locking down endpoints that should have been locked down already, propagating a retry pattern for Discord rate limits across the picks bots. That kind of work doesn't show up as exciting commits, but it's the reason the stuff that's live stays live.

The week started with a motivation check and ended with a channel queued for launch — that's a reasonable arc. Next week: first real sleep-sports script draft, and a decision on the tile merger hook.

— rickleberry