Skip to content

Git Recon: menus as a toy-soldier briefing, hill climb to 8/10, rounds 5+ (from #3569) #3575

Closed

Description

@QuinnsCode

Score: 5.2/10 (target 8)

Carried from #3569 (round 4, PR #3574; rounds 1 to 3 were PRs #3552 and #3559): Git Recon's menus as a toy-soldier briefing. A toy-soldier third-person shooter fought in a codebase, with glossy black clay soldiers. The menus are the toy box and the field manual: moulded-plastic sprues and runners as frames, a glossy clay sheen on buttons, stencilled field-manual headings, mission cards like punched dog tags. Our own ideas only, never the look of any commercial toy line or game.

Goalpost: 8/10 from the owner on the menus

  1. In the game's own look. Title, lobby or front, settings, results and any panel read as the same world as the play view: material, type, ornament.
  2. Artwork, not only CSS. Backplates, frames, textures and insignia as small SVG, procedural canvas or headless-Blender tiles; say what each weighs.
  3. Still readable, every screen size. Contrast at or above today, play view unobscured, checked with bun run probe git-recon --shot from phone to desktop.
  4. eqosys menus wear it. The sign-in chip, report form, achievement toast and any lobby come through the skin hook from eqosys menus: a game injects its skin once and every shared surface wears it, hill climb to 8/10, rounds 2+ (from #3530) #3541, never a second stylesheet.
  5. House kit underneath. Components come from packages/ui and packages/game-controls re-themed by tokens; a forked copy of one is a gap.

Where it stands after round 4

Next

  1. Settings and loadout are still CSS, not artwork.
    • Give settings stencilled section headings (its "Keyboard" header is still sans) and a moulded panel. Its mono rows read as a terminal, not a manual: try the stencil or a typed-manual face instead.
    • Give the loadout's head swatches moulded sockets instead of plain squares.
  2. The backdrop figures are soft olive blobs, almost invisible on phone. In menu-art.ts, light them harder (rim light, contrast) and frame fewer, bigger parts so they read as toy soldiers. Re-bake.
  3. Pick one button material per page. Results mixes black clay role pills with steel dog-tag buttons.
  4. Small fixes.
    • On tablet, the backdrop's gate labels are cut at the edge.
    • On desktop, the map card overlaps the sprue labels.
  5. eqosys menus. When the skin hook (eqosys menus: a game injects its skin once and every shared surface wears it, hill climb to 8/10, rounds 2+ (from #3530) #3541, PR eqosys skin: a game injects its look once, and the chip, toast, report form, lobby and spaces wear it (climb round 1) #3540) is on main, put this skin through it for the sign-in chip, report form and toast. Not before.

Activity

  1. QuinnsCode commented on Nov 5, 2026

    @QuinnsCode
    OwnerAuthor

    Stalled: four rounds tonight have left the score flat at about 5/10 (target 8). Each round's fresh judge, corrected two points down, gave:

    round PR judge's BEFORE judge's NOW
    1 #3552 1.5 (main) 5.0 (a second judge agreed: 1.5 and 5.0)
    2 #3559 4.4 (round 1's result) 5.2
    3 #3559 4.8 (round 2's result) 5.3
    4 #3574 4.5 (round 3's result) 5.2

    Judge drift is as big as the gains. Every judge saw its own round as a rise of +0.5 to +0.8. But each new judge put the result it was handed 0.2 to 0.8 lower than the judge before had scored it. So the headline number sits at 5.0 to 5.3, and three rounds have bought no gain on the climb's own measure.

    Did the rendered backplate move the judges? A little, not enough. The judge called round 4's art "the real gain": a sprue of olive clay soldiers baked by scripts/menu-art.ts with the game's own figure, and the posed soldiers on results. But it read the moulded soldiers as "soft olive blobs", almost invisible on phone. The page is still a CSS card laid over the render, so the score rose by about the same as the CSS rounds did.

    What I think the real gap is:

    1. The screens are web forms wearing a skin. Settings is a list of key bindings; loadout is radio chips and inputs. Every judge marked those two down whatever the frame. Reaching 8 probably means rebuilding them as objects: the loadout as an actual parts tray you pick pieces off, settings as a field-manual page with diagrams. That's layout and design work, not more texture.
    2. The renders need a real art pass. The game's figure is built from round clay primitives, so moulded in olive and lit flat it reads as blobs. Sharper lighting, fewer and bigger parts, or a figure with a sculpted silhouette for the menus would carry more than another CSS layer.
    3. The judges can't settle which direction is right. More toy box or more field manual is a taste call, and agent judges run high and drift. That's why the next step is the owner's own score: MANUAL RUN issue linked below.

    Open PR: #3574 (round 4: the baked art, plus ?screen= links that open each menu screen without playing a round).

  2. QuinnsCode commented on Nov 5, 2026

    @QuinnsCode
    OwnerAuthor

    The owner's score and choice of direction: #3576 (MANUAL RUN). This climb waits on it.

  3. QuinnsCode commented on Nov 5, 2026

    @QuinnsCode
    OwnerAuthor

    Score: 3.5/10 (target 8)

    Round 5 is PR #3593. This round changes direction: the owner, through Nibble, asked for a sniper's mission briefing (the scope, the dossier, the classified marks) on the clean white crafting-sandbox canvas, still rugged, keeping the dog tags. The olive toy box is gone.

    A fresh judge gave BEFORE (main, round 4) 3.5 and NOW 5.5, each corrected two points down. It scored against the new brief, so these numbers don't line up with round 4's 5.2.

    Shots are in apps/git-recon/docs/shots-2026-11-05-briefing/ (before and now; title, loadout, settings, won and lost at desktop, tablet and phone).

    What landed:

    • A white dot-grid table with grit, under a faint scope reticle.
    • The card is a dossier: corner brackets, an orange edge, an EYES ONLY strip and torn tape.
    • Buttons are white tiles and chips. Sealed missions are flat, hatched dog tags.
    • Results are a DEBRIEF with the stamp in the one orange.
    • Settings is the same dossier through packages/ui tokens.
    • The olive backplate is gone (55 KB less), and the climb's CSS is one section.

    Next: real briefing artwork. Ideas: a manila folder edge or tab, a paper clip, redaction bars, the soldier as a clipped photo. Then a crisp reticle that frames the title or marks the picked mission, and settings and the debrief made more than a white form.

  4. QuinnsCode commented on Nov 5, 2026

    @QuinnsCode
    OwnerAuthor

    Moved up to #3594

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions