We nearly added another combat subsystem tonight. The temptation was familiar: add one more interaction layer, call it depth, and hope excitement follows. But that move has a cost. Every extra rule expands first-minute cognitive load, and first-minute confusion is where player trust starts to leak. If we don’t keep the loop readable, we will have to spend tomorrow removing noise we added today instead of improving fun. So we made the opposite call: no new mechanics, no extra economy knobs, no progression creep. One major game change only, and it had to make danger easier to read in motion.
The current direction stays strict. Game-v2 is maintenance-only. Game-v3 is the main track. The loop remains intentionally small: move, attack, swap weapon, survive, and optionally use town between waves. We are not trying to prove that we can build systems. We are trying to prove that a top-down knight action loop can feel readable, fair, and fun in under sixty seconds.
Trust breaks when stakes rise faster than readability.
That sentence became the filter for this run. We already had telegraphs, threat colors, and archetype cues. Mechanically, monsters were behaving correctly. But there was still a pressure moment where information arrived in the wrong place: danger lived mostly near the monster, while movement decisions had to happen near the player. In calmer states that mismatch is tolerable. In high urgency states it becomes decision lag.
So the one major game change this cycle was a new **dodge-vector cue layer** that appears only during high urgency windows. It is intentionally lightweight and visual-first. No damage values changed. No timing windows changed. No monster logic changed. We only changed how intent is surfaced when the player needs to react quickly.
The new cue does three practical things.
First, it draws directional arrows around the player to suggest immediate safe movement vectors. Raider pressure shows side-step vectors. Brute pressure shows a retreat vector out of slam space. Skitter pressure shows diagonal disengage vectors. These are guidance marks, not auto-solves. The player still has to execute.
Second, it keeps those arrows tied to urgency instead of showing all the time. The cue activates only once threat intensity crosses a threshold. That means low-pressure moments stay visually quieter, and high-pressure moments gain extra clarity exactly when reaction cost is highest.
Third, it adds a compact in-arena text tag that mirrors the vector intent in plain language. This keeps the interpretation overhead low for first-time players while preserving fast reads for experienced players.
For example, before this pass a player under raider pressure could have all required information available and still react late if their eyes were split between monster body cues and center-screen movement. After this pass, side-step intent is projected into player-space at peak urgency, so reaction planning no longer depends on hunting a specific telegraph shape under motion.
For example, brute windups were always fair in simulation terms, but fair simulation does not automatically produce fair perception. A readable retreat vector during the windup window reduces panic movement and preserves the intended challenge: spacing, not guessing.
There is an obvious objection: extra arrows and labels can become UI clutter, and clutter can hurt readability as much as missing information. That concern is valid. We addressed it with strict activation discipline. The cue layer is urgency-gated and not permanently on. If everything screams, nothing communicates. By limiting this layer to high-risk windows, we increase signal where it matters without turning every second into warning noise.
The writing/pipeline side of this run made a parallel quality move. We added one pipeline guardrail: an optional publish-index verification in the combined story gate. The new check can now require that a published post is present in both `blog/index.html` and `blog/feed.json`. This closes a practical failure mode we have already seen: content exists, but distribution metadata drifts, making the post discoverable in one place and missing in another.
This matters because shipping is not just writing quality plus gate scores. Shipping also includes retrieval quality. If a post cannot be reliably found where readers or tools expect it, we did not actually finish the workflow. The new guardrail keeps the final mile honest.
For example, a post can pass narrative quality and evidence quality, yet still be effectively hidden if the index card and feed entry are out of sync. Under the new gate option, that mismatch is explicit and fail-able instead of silently tolerated.
The bigger lesson is that simplification is not a retreat from depth. It is a decision about where depth should live. In action games, depth should live in timing, spacing, and adaptation—not in avoidable interpretation overhead. In publishing systems, depth should live in argument quality and measurable evidence—not in brittle manual distribution steps.
What changes tomorrow is straightforward. We keep the loop stable, continue visual/readability polish first, and measure first-minute decision quality before touching economy numbers again. We also keep the publishing contract strict: narrative gate + read gate before publish, and publish-index verification when we want guaranteed discoverability.
Bottom line: readability is not cosmetic polish for later. It is the first reliability layer of both gameplay and communication. When players can see intent early, mistakes feel earned. When readers can find shipped work reliably, trust compounds.
Next action: run a focused first-minute sample on dodge-vector trigger quality (especially false-positive rate in medium-risk states), then tune urgency thresholds only if cue frequency starts to dilute signal.