Map renderer theme system and harness
Transcript
Thinking01:28:41
I want to explore the renderer directory thoroughly, starting with theme.lua in full, then elements.lua, homepage.lua, sidebar.lua, main.lua, settings.lua, and other files as needed, plus the ricochet folder, tests/tools directories, and Anchor/engine docs like SURFACE.md and AGENT.md. I'll begin with a broad directory listing and reading theme.lua completely.
Bashcd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && ls -la && echo "---ri...01:28:41Output not recorded
cd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && ls -la && echo "---ri...Bashwc -l "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/theme.lua" "C:/Users/a...01:28:41Output not recorded
wc -l "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/theme.lua" "C:/Users/a...Thinking01:28:46
Settings.lua's line count seems odd, so I want to check its actual content directly, and also look at the anchor directory for engine docs, plus the tools and tests directories. I'll read theme.lua in full and explore these files in parallel.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/theme.lua01:28:46Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/theme.luaBashcat "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/settings.lua"; echo "---...01:28:46Output not recorded
cat "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/settings.lua"; echo "---...Thinking01:28:56
I notice a naming mismatch in theme.lua: the THEMES table stores a key like "posts_old" but the apply function checks against "site_old" instead, which looks like a possible dead code path worth flagging.
I want to grep main.lua for theme_apply calls and related names like 'posts_old', 'site_old', plus F2, theme toggle, settings.lua usage, A327EX_ADMIN, MOBILE_BP, and sidebar references. I should also peek at tests/attribution.lua since it shares a name with the top-level attribution.lua, which seems odd.
Greptheme_apply|posts_old|site_old|theme_active|F2|A327EX_ADMIN|MOBILE_BP01:28:56Output not recorded
theme_apply|posts_old|site_old|theme_active|F2|A327EX_ADMIN|MOBILE_BPReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/run.bat01:28:56Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/run.batGreptheme_apply|posts_old|site_old|theme_active01:28:56Output not recorded
theme_apply|posts_old|site_old|theme_activeBashcat "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/tests/attribution.lua" |...01:28:56Output not recorded
cat "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/tests/attribution.lua" |...Thinking01:29:07
I see there's just a single Lua test file, attribution.lua, likely run directly with lua54's asserts. I should check how it's invoked—maybe there's a runner script—and look at PLAN.md and tools/ for related references, plus check the Anchor engine docs for relevant commands.
I also need to review several sections of main.lua: theme application logic, settings persistence and dirty-flag handling, F2 debug key handling, and the mobile top bar with hamburger menu. Beyond that, I should look through sidebar.lua, homepage.lua, and elements.lua for chip/card/hairline/button/color constants, dig deeper into settings.lua's persistence functions, and check where the A327EX_ADMIN env var gets used elsewhere. I'll run these reads in parallel.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.lua01:29:07Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.luaReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.lua01:29:07Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.luaReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.lua01:29:07Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.luaReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/sidebar.lua01:29:07Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/sidebar.luaThinking01:29:12
With sidebar.lua covered, I want to check main.lua's F2 handler, settings persistence, the mobile top bar/hamburger drawing code, and the settings page handling, plus where settings are saved and loaded.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.lua01:29:13Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.luaReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.lua01:29:13Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.luaGrepsettings_save|settings_load|settings\.lua|__settings__|home_hidden|font_scale|text_width|hide_radio01:29:13Output not recorded
settings_save|settings_load|settings\.lua|__settings__|home_hidden|font_scale|text_width|hide_radioThinking01:29:20
I want to read the full settings section to see how the theme setting is saved and applied, and check whether the settings page UI actually exposes a theme selection row alongside things like font scale and text width.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.lua01:29:20Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/main.luaThinking01:29:28
The settings page toggles between light and dark theme only, which is key context for figuring out where a third theme would need to be registered. Now I want to check for A327EX_ADMIN env var usage and look into the PLAN.md commit about the lua54 homepage harness and headless snapshots.
I should also look at the Anchor engine docs (SURFACE.md/AGENT.md), the tools directory for snapshot logic, the anchor drive and anchor check (LuaLS) commands, and elements.lua/homepage.lua for hardcoded chip/card/hairline/button constants — I'll run several searches in parallel.
GrepA327EX_ADMIN01:29:28Output not recorded
A327EX_ADMINGreppx\(3\)|gold|QUOTE_BORDER|CHIP|chip|radius|rounded_rectangle|px\(1\)|hairline|HAIRLINE01:29:28Output not recorded
px\(3\)|gold|QUOTE_BORDER|CHIP|chip|radius|rounded_rectangle|px\(1\)|hairline|HAIRLINEBashfind /c/Users/a327e/Desktop/a327ex/Anchor -maxdepth 3 -iname "SURFACE.md" -o -in...01:29:28Output not recorded
find /c/Users/a327e/Desktop/a327ex/Anchor -maxdepth 3 -iname "SURFACE.md" -o -in...Thinking01:29:35
I should check SURFACE.md and AGENT.md for mentions of "anchor drive", "engine_snapshot", and "anchor check" commands, but I'll grep for those specific terms rather than reading the whole file since it could be large.
Bashwc -l /c/Users/a327e/Desktop/a327ex/Anchor/engine/docs/SURFACE.md /c/Users/a327e...01:29:35Output not recorded
wc -l /c/Users/a327e/Desktop/a327ex/Anchor/engine/docs/SURFACE.md /c/Users/a327e...Grepanchor drive|engine_snapshot|engine_render_setup|engine_step|snapshot|headless|anchor check|LuaLS01:29:35Output not recorded
anchor drive|engine_snapshot|engine_render_setup|engine_step|snapshot|headless|anchor check|LuaLSThinking01:29:39
I should read AGENT.md more thoroughly around those sections for the exact commands, and also check for an "anchor check" command and LuaLS references elsewhere, like workflow docs or the Anchor root README, by broadening my search.
Read/c/Users/a327e/Desktop/a327ex/Anchor/engine/docs/AGENT.md01:29:39Output not recorded
/c/Users/a327e/Desktop/a327ex/Anchor/engine/docs/AGENT.mdGrepanchor check|LuaLS|lua54|homepage harness|headless snapshot01:29:39Output not recorded
anchor check|LuaLS|lua54|homepage harness|headless snapshotThinking01:29:54
The Anchor CLAUDE.md gives exact commands for LuaLS checks and drive eval. Now I want to check the PLAN.md section about the lua54 harness and headless snapshots, plus look in tools/ for anything related to headless rendering.
I'll also check the git log around the "unified feed" commit that mentions the lua54 homepage harness and headless snapshots, checking the delegate run report or git log in the a327ex-site repo directly.
Bashcd "C:/Users/a327e/Desktop/a327ex/a327ex-site" && git log --oneline -15 -- rende...01:29:55Output not recorded
cd "C:/Users/a327e/Desktop/a327ex/a327ex-site" && git log --oneline -15 -- rende...Greplua54|headless|harness|homepage harness01:29:55Output not recorded
lua54|headless|harness|homepage harnessThinking01:30:00
I want to check commit ac7e500's message for mentions of the lua54 homepage harness and headless snapshots, then verify where lua54.exe is currently located now that a8c1199 fixed the path finder.
Bashcd "C:/Users/a327e/Desktop/a327ex/a327ex-site" && git show -s ac7e500 --format="...01:30:01Output not recorded
cd "C:/Users/a327e/Desktop/a327ex/a327ex-site" && git show -s ac7e500 --format="...ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/PLAN.md01:30:01Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/PLAN.mdThinking01:30:08
I notice PLAN.md still references the old E: drive path, which is now stale since the machine switched to C:/Users/a327e/Desktop/a327ex and the recent fix removed the hardcoded E: path for locating lua54. So the correct command on this Windows machine should reference lua54.exe under Anchor/workflow/tools instead.
I want to check how tools like convert.lua actually auto-detect lua54, so I can state the exact command precisely — I should grep the deploy scripts for that detection logic. I also need to remember the run.bat instruction is user-only and should never be executed directly, which I'll flag in the verification section as mechanical checks only, with no interactive run.
Greplua54|lua5\.4|LUA_BIN|find_lua01:30:09Output not recorded
lua54|lua5\.4|LUA_BIN|find_luaBashcd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && grep -rn "lua54\|LUA_...01:30:09Output not recorded
cd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && grep -rn "lua54\|LUA_...Thinking01:30:17
I've confirmed the exact syntax-check command for Lua files using lua54.exe from the renderer directory, distinguishing it from the luac -p approach mentioned in Anchor's CLAUDE.md, since this Windows setup requires the standalone Lua interpreter instead.
Let me check if luac54.exe also exists in workflow/tools, then look for a headless testing harness that might directly load theme.lua or homepage.lua without the engine—searching for dofile/require patterns in the tools directory, unify_feed.py, make_report.py, and PLAN.md for any references to a homepage test harness.
Bashcd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && grep -n "homepage har...01:30:18Output not recorded
cd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && grep -n "homepage har...Bashcd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && ls Anchor/workflow/de...01:30:18Output not recorded
cd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && ls Anchor/workflow/de...Bashgrep -n -i "lua54\|homepage harness\|headless snapshot\|harness" "/c/Users/a327e...01:30:18Output not recorded
grep -n -i "lua54\|homepage harness\|headless snapshot\|harness" "/c/Users/a327e...Thinking01:30:25
I want to trace what the "lua54 harness" and "headless snapshots" actually refer to, likely a custom script from the subagent's work, so I should check the tools/_audit directory and the .repo-visibility file for context.
Read/c/Users/a327e/Desktop/a327ex/Anchor/workflow/delegate/runs/20260911-unified-feed/report.md01:30:25Output not recorded
/c/Users/a327e/Desktop/a327ex/Anchor/workflow/delegate/runs/20260911-unified-feed/report.mdBashls -la "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/tools/_audit" 2>&101:30:25Output not recorded
ls -la "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/tools/_audit" 2>&1Thinking01:30:32
I'm realizing that pegasus/lfs/dkjson/discount are actually the server-side Lua web framework dependencies used by server.lua, not part of the renderer itself — so the harness the subagent built was stubbing out those server dependencies to test homepage rendering in isolation.
So this harness tests server/content.lua's homepage generation (headless browser snapshots of the server's HTML), separate from the Anchor renderer engine's canvas rendering. Let me check the tools/_audit/unified-feed directory for the actual harness script to confirm.
Bashfind "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/tools/_audit/unified-fe...01:30:33Output not recorded
find "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/tools/_audit/unified-fe...Thinking01:30:55
This confirms the harness is server-side—it renders home.html by running server/content.lua and server.lua through lua54 with pegasus/lfs/dkjson/discount stubbed, then takes headless browser screenshots of that output. So this is testing the server's HTML rendering, not the Anchor engine's theme.lua/elements.lua/homepage.lua, which belong to the separate desktop/web app renderer.
That's distinct from anchor's own engine_snapshot for desktop instances — this is a separate headless browser screenshot tool for the web build.
For section 5, I need to note that tests/attribution.lua is a lua54 assertion script, not an interactive game, and its dofile paths reveal it must be run from the a327ex-site root directory rather than from renderer/ itself, since it references paths like 'renderer/attribution.lua' and 'renderer/data/horse-game-19.lua'.
I'm working out the exact invocation using lua54.exe on Windows via Git Bash, then considering how tools/convert.lua (the content-to-data converter) could be syntax-checked with a loadfile assertion instead of a full run, and whether `anchor check renderer` would work given the framework copies present in the renderer's anchor folder — though I need to confirm whether anchor.meta.lua actually exists there.
Then I found the key answer directly in main.lua's comments: offscreen verification is meant to be done via a driver, not custom code — specifically `anchor drive start renderer` followed by `anchor drive eval renderer` running setup and stepping the engine to capture every frame, with `engine_snapshot(path)` available for on-demand PNGs. This confirms the exact commands for headless engine snapshots of the renderer app itself. Now I want to check if anchor.exe exists within the render directory.
Let me check for anchor.meta.lua in renderer and see if an "anchor" command is available on PATH.
Bashfind "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/anchor" -iname "*.meta....01:30:55Output not recorded
find "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/anchor" -iname "*.meta....Thinking01:31:05
So "anchor" isn't found via `command -v` but exists as a script at Anchor/workflow/anchor — it's likely added to PATH through the user's normal bash profile, just not available in this tool's environment. That's enough detail to note in the report.
I'm also checking whether renderer has an anchor.meta.lua or anchor.toml config anywhere, since Anchor games are typically recognized just by having a main.lua at the project root with the require('anchor') init call — which renderer already has, so that should suffice for "anchor drive start renderer" to work.
Bashfind "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" -maxdepth 1 -iname "*....01:31:06Output not recorded
find "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" -maxdepth 1 -iname "*....Thinking01:31:17
I notice there's no anchor.meta.lua or VERSION stamp in this renderer/anchor/ folder, suggesting it's an older or custom-trimmed framework copy rather than a standard one. That's worth flagging as a caveat—since anchor.meta.lua isn't present, standard commands like `anchor check` or `anchor framework status` may not function normally here.
I have enough context now, so I'm turning to elements.lua and homepage.lua to pin down exact line numbers for each visual component I need: filter chips, role chips (already located around lines 825-902), mediacard, tweet card, video poster/play button, code blocks, blockquote bar, heading hairlines, links, feed row, and the ARTIFACTS label.
Grepfunction draw_mediacard_element|function draw_tweet|function draw_video_element|function draw_code|function draw_blockquote|QUOTE_BORDER|function draw_.*link|ARTIFACTS|play_button|poster01:31:17Output not recorded
function draw_mediacard_element|function draw_tweet|function draw_video_element|function draw_code|function draw_blockquote|QUOTE_BORDER|function draw_.*link|ARTIFACTS|play_button|posterGrepfilter chip|FILTER_CHIP|function draw_feed_row|feed_row|ARTIFACTS|meta line|read more|function draw_home|CHIP01:31:17Output not recorded
filter chip|FILTER_CHIP|function draw_feed_row|feed_row|ARTIFACTS|meta line|read more|function draw_home|CHIPThinking01:31:22
I found the filter chips drawing around line 531 and the ARTIFACTS label around lines 613-622 in homepage.lua. I want to pin down the exact function names and lines for both the filter chip rendering and the feed row rendering, so I'm going to look more closely around 525-650 and further down near 700+ for the feed row code.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/homepage.lua01:31:22Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/homepage.luaGrepfunction draw_row_element|function.*feed.*draw|function draw_home|function draw_entry|title.*meta|meta_y01:31:22Output not recorded
function draw_row_element|function.*feed.*draw|function draw_home|function draw_entry|title.*meta|meta_yThinking01:31:27
I want to check line 759 where the row entry with title and meta is rendered.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/homepage.lua01:31:27Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/homepage.luaGood, I have enough now. Let me check a couple more precise spots: links color in elements.lua, blockquote bar/QUOTE_BORDER definition, and the code block drawing signature, then write the report.
GrepQUOTE_BORDER|prose\.color\.link|function draw_quote|blockquote01:31:31Output not recorded
QUOTE_BORDER|prose\.color\.link|function draw_quote|blockquoteReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/elements.lua01:31:31Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/elements.luaThinking01:31:36
I want to find where the vertical bar for blockquotes is actually rendered, so I'm searching for the quote element type's drawing logic.
Grepkind == 'quote'|kind=='quote'|quote_bar|>>|right-aligned blockquote01:31:36Output not recorded
kind == 'quote'|kind=='quote'|quote_bar|>>|right-aligned blockquoteGrep'quote'01:31:36Output not recorded
'quote'Grepquote01:31:36Output not recorded
quoteThinking01:31:49
I don't find a literal blockquote bar drawn in elements.lua by grep, so it might just render as indented italic text with a quote color rather than a vertical bar. Since the task explicitly mentions a "blockquote bar," I'm searching for alternate spellings or a quote-kind check that might draw a left border rectangle instead.
Grepquote01:31:49Output not recorded
quoteThinking01:31:54
I found the QUOTE_INDENT and QUOTE_BORDER constants in canvas.lua around lines 15-16, and the actual left border bar drawing is likely near line 417 — let me look at lines 400-430 to check.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/canvas.lua01:31:54Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/canvas.luaNow I have all the pieces needed. Let me do one final check on the ricochet folder relationship (is it imported by main.lua — yes, require('ricochet') at main.lua:35) and confirm it's actively used, not vestigial.
ReadC:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/ricochet/init.lua01:31:57Output not recorded
C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer/ricochet/init.luaThinking01:32:10
I've confirmed main.lua calls ricochet_setup with smooth mode, light palette, and effects disabled, giving access to ui_layer, an unused effect system, a palette that gets overridden by theme.lua's own color roles, and the ricochet.ui widget toolkit—now I need to check whether the site renderer actually uses those ricochet.ui widgets or draws its own custom UI instead.
Checking sidebar.lua, I find it calls `ui_text{...}` directly, confirming this is a ricochet/ui.lua widget function, so ricochet is genuinely used for basic text rendering and mouse-handling globals. But the site seems to draw its own rectangles and hairlines rather than calling ricochet's skinned buttons and panels, which suggests ricochet's skin_rough.lua, skin_smooth.lua, and theme.lua (the widget chrome/profile system) may be effectively bypassed by the site's custom drawing approach.
I want to verify this by searching for usage of things like ui_button, ui_panel, skin_rough, skin_smooth, ricochet_theme_set, and ui_layer across the renderer files to see exactly which ricochet exports elements.lua, homepage.lua, sidebar.lua, and main.lua actually call.
Bashcd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && for fn in ui_button u...01:32:20Output not recorded
cd "C:/Users/a327e/Desktop/a327ex/a327ex-site/renderer" && for fn in ui_button u...Thinking01:32:35
These are just converted markdown blog posts about UI work that happen to mention ui_button and similar terms in their text content, not actual calls from the renderer's source files. So none of the real renderer files—main.lua, elements.lua, homepage.lua, sidebar.lua, canvas.lua—actually invoke ui_button, ui_panel, ui_slider, skin_rough, or skin_smooth.
This confirms ricochet's widget skin layer is entirely unused; only the low-level ui_text/ui_state hot-tracking, a disabled effect_setup, and a one-time palette_init (immediately overridden by the site's theme.lua) are actually wired in. The higher widgets in skin_rough.lua, skin_smooth.lua, and ui.lua are dead weight inherited from the template.
I've also spotted a naming mismatch: the THEMES table uses keys like site_light, site_dark, posts_old, but theme_apply checks for 'site_old', which never matches any of those keys—so the old-family activation logic tied to that check is currently dead code.
Since theme_apply is only ever invoked with 'posts_old' from main.lua, the old lazy-family font swap never fires, meaning OLD_FONTS stays locked to its initial SANS fallback rather than being refreshed with the intended old-style fonts. This looks like a genuine latent bug worth flagging with the specific file and line references in my conclusions, though not something to fix here—I just want to confirm there's no other call site using 'site_old' anywhere else in the renderer tree before I note it.
Grepsite_old01:32:35Output not recorded
site_oldThinking01:32:42
This confirms it's a genuine dead-code mismatch: the variable was renamed from 'site_old' to 'posts_old' in the THEMES table and its call site in main.lua, but the two internal branches in theme_apply were left unupdated. I now have everything needed to write the final CONCLUSIONS response across the six numbered sections with precise file:line references.
CONCLUSIONS
1. Theme system (renderer/theme.lua, 413 lines, fully read)
prose (theme.lua:19) is the single live global the renderer reads: {fonts={}, color={}, metrics={}, gap_before={}, treat={}}. fonts maps role names (body,bold,italic,h1,h2,h3,code,small,small_italic,banner,tag,tweet,tweet_bold,tweet_italic, defined at theme.lua:118-139) to registered font handles; color maps role names to stable, mutated-in-place color objects (theme.lua:386-390) so a color-only theme swap needs no relayout — cached layouts hold references to the same object. metrics (theme.lua:140-146, 399-403) holds column_w, text_w, element_gap, body_line_height, head_overflow, all DPR-scaled. gap_before (theme.lua:149, 405-406) is per-kind top margin (h1,h2,h3,image,quote). treat (theme.lua:150, 407-408) is a small flags table, currently only heading_rule (bool).
theme_apply(name, scale, text_w) (theme.lua:341-409) looks up THEMES[name], re-registers every font at size*font_scale*dpr (reusing cache by (file,size) key, theme.lua:312-326), rewrites every prose.color[role] in place, and rescales metrics/gaps by DPR. Three themes exist in THEMES (theme.lua:210-304): site_light, site_dark (share FONTS/METRICS/GAP_BEFORE/TREAT, differ only in colors — a color-only swap, no relayout) and posts_old (theme.lua:275-303, its own OLD_FONTS/OLD_METRICS/OLD_GAP_BEFORE/OLD_TREAT — switching to/from it needs relayout, done by main.lua's load_page). Color keys and site_dark values (theme.lua:245-271): text/ink (230,230,230,255), bg (17,17,17,255), off_bg (255,255,255,10), link (140,180,255,255), dim (255,255,255,128), medium (255,255,255,168), code (230,230,230,255), quote (187,187,187,255), rule (255,255,255,20), rule2 (255,255,255,38), spoiler_bg (255,255,255,255), spoiler_text (0,0,0,255), spoiler_link (48,84,255,255)... 48,84,235,255, selection, find/find_cur, artifact (255,203,92,255), driver (217,119,87,255), add (120,200,120,255), del (235,110,110,255)`.
Fonts: base Latin faces are bundled TTFs (SANS=Arial, theme.lua:32-34; TWEETF=Segoe UI, theme.lua:92-95). Lazy families (cjk, kr, symbol, old, theme.lua:36-114, 355-382) are fetched into MEMFS on web only when a page needs them (theme_want/theme_poll_lazy); desktop wants them all at boot. A real bug: theme_apply guards the old-theme lazy-family activation with if name == 'site_old' (theme.lua:357, 382) but the actual key is posts_old (theme.lua:275, main.lua:818) — that branch is dead, so refresh_old_fonts never runs and posts_old silently keeps its Arial fallback instead of the real Source Sans faces.
Outside theme.lua, hardcoded chrome constants: canvas.lua:15-16 QUOTE_INDENT=px(20), QUOTE_BORDER=px(4) (blockquote left bar width, reused for the tweet-thread user-turn bar); elements.lua:2613 reuses px(4) for the mediacard accent edge; elements.lua:2605 picks gold = prose.color.artifact or prose.color.link for artifact/session-made accents; role chips (elements.lua:825-919) and filter chips (homepage.lua:531-620) use pill radius math.floor(c.h/2) and are drawn with layer_rounded_rectangle, not theme constants; sidebar.lua:14-21 (PAD, NAV_H, SEC_H, ROW_H, INDENT) are its own px()-scaled layout numbers.
2. Theme switching
Runtime: F2 (input_pressed('theme_toggle'), main.lua:2284-2289) flips new_theme between 'site_dark'/'site_light' and calls apply_theme — color-only, no relayout, and it's suppressed inside an old-Blot doc. Settings page (main.lua:1183-1185) exposes "Theme" as a binary stepper row (Light/Dark) that toggles the same two values via settings.theme. Persistence: settings_save/settings_load (main.lua:1106-1135) write a literal-table Lua file settings.lua on desktop (file_write_string) and localStorage on web (web_store_set), loaded at boot (main.lua:1256); settings_load clamps any non-site_light value back to site_dark (main.lua:1134) — a third theme name would currently be coerced away here. A327EX_ADMIN=1 (homepage.lua:86) only gates the admin review UI, unrelated to theming. To add a third theme: (a) add a THEMES.my_theme entry in theme.lua (~line 304); (b) fix/extend the settings.theme clamp at main.lua:1134 to accept it; (c) replace the binary toggle at main.lua:1183-1185 (and the F2 flip at main.lua:2285) with a 3-way cycle; (d) if it needs its own fonts/metrics (like posts_old), route it through load_page's apply_theme(...) + relayout() (main.lua:818-819) rather than the F2 fast path.
3. Chrome
Sidebar (sidebar.lua, fully read): fixed nav rows (home/search/settings/privacy/archive, lines 83-92) then scrollable, stencil-clipped (layer_stencil_mask/test/off, lines 96-155) disclosure sections built by sidebar_build (line 27) and drawn by sidebar_draw (line 61), all drawn with raw layer_rectangle/layer_text/ui_text, not ricochet widgets. Mobile top bar: draw_mobile_topbar (main.lua:2930-2966) draws an opaque bar + hairline, "a327ex" title (tap → home), hamburger (bars/X toggle mobile_menu_open) and a find magnifier. The breakpoint is MOBILE_BP = 768 CSS-px (main.lua:112), compared in compute_layout (main.lua:121-135) as width < MOBILE_BP*UI_S (main.lua:122); when true, left_col_w/sidebar_x = 0 and content goes full-width. There is no existing flag to force the mobile top-bar layout while on a wide desktop viewport — mobile_mode is purely width-derived. To offer "no sidebar, top-line nav" as a theme option: add a boolean (e.g. prose.treat.top_nav) read by compute_layout (main.lua:121-135, alongside/instead of the mobile_mode check) and by the draw dispatch that currently branches on mobile_mode to call draw_mobile_topbar/skip the sidebar (search around main.lua:2967+ draw() and content_view() main.lua:290-294).
4. Element treatments (file:line, one line each)
- Filter chips (pills):
homepage.lua:531-620(chip_width,draw_chip_rows) — readshome_hidden/chip.count/.label/.fold. - Role chips:
elements.lua:825(ROLE_CHIP_DEFS),elements.lua:857layout,elements.lua:902draw — readsel.controls,doc.models. - Mediacard:
elements.lua:2598draw_mediacard_element— readsel.media, image/title/platform. - Tweet card:
elements.lua:2323draw_tweet_element— reads tweet author/handle/body/quoted tweet. - Video poster/play button:
elements.lua:1057draw_play_button,elements.lua:1470draw_video_element— readsel.image,el._video. - Code blocks:
elements.lua:1672draw_code_element— reads code lines/lang,prose.fonts.code. - Blockquote bar:
canvas.lua:417-428(QUOTE_BORDERfrom canvas.lua:16) — readsel.kind=='quote',el.role. - Headings hairline:
elements.lua:642-645— readsprose.treat.heading_rule,el.kind. - Links: color role at
elements.lua:77-81(style.link/style.artifact), underline atelements.lua:573. - Buttons: no generic themed button; closest is settings' stepper box (
main.lua:1223-1250) and the game touch-chip row (elements.lua:2794-2806). - Feed row (title + meta):
homepage.lua:759-774— readsf.title,entry_runs(f)(homepage.lua:284). - ARTIFACTS label:
homepage.lua:617(draw_tracked('ARTIFACTS', prose.fonts.banner, ...)), defs athomepage.lua:248.
5. Verification without running interactively
renderer/tests/attribution.lua is a plain Lua-assert script (no framework) exercising renderer/attribution.lua; it dofiles paths relative to a327ex-site/, so run it from there: Anchor/workflow/tools/lua54.exe renderer/tests/attribution.lua (Git Bash on Windows; PLAN.md:1078-1084 references a stale E: drive/path, since fixed to auto-detect — see renderer/tools/deploy.sh:33, uncached_media.py:18-19, which try lua5.4/lua54 on PATH then fall back to Anchor/workflow/tools/lua54.exe). Syntax-check any file with lua54.exe -e "assert(loadfile('renderer/theme.lua'))". renderer/tools/convert.lua (content→data) is runnable/checkable the same way. The "lua54 homepage harness and headless snapshots" mentioned in the unified-feed commit (git log: ac7e500, and Anchor/workflow/delegate/runs/20260911-unified-feed/report.md:74-77,124-132) is a server-side harness (stubs pegasus/lfs/dkjson/discount) that renders server/content.lua's HTML homepage under lua54 and takes headless-browser PNGs of it into renderer/tools/_audit/unified-feed/*.png (gitignored) — it exercises the HTML feed, not the Anchor canvas renderer directly. For the Anchor canvas itself, main.lua:1258-1260 documents the real mechanism: anchor drive start renderer, then anchor drive eval renderer "engine_render_setup('render') engine_step(60)" — an agent (headless) instance that steps frames and can call engine_snapshot(path) for an on-demand PNG (see Anchor/engine/docs/AGENT.md:22-25,38-47,71-78 and SURFACE.md). This requires the anchor CLI wrapper (Anchor/workflow/anchor) on PATH — not found via command -v anchor in this shell, so it may need Anchor/workflow added to PATH first. anchor check <game> (LuaLS) is documented in Anchor/engine/.claude/CLAUDE.md, but renderer/anchor/ has no anchor.meta.lua/VERSION stamp (unlike a stamped Anchor3 game), so it's unclear it's wired for this project without extra setup. renderer/run.bat (renderer/run.bat:1-11) cds to the parent dir and launches renderer/anchor.exe "<game-folder-name>" — per Anchor/engine/.claude/CLAUDE.md and PLAN.md, the user runs this interactively, not the agent.
6. renderer/ricochet/
This is the generic Ricochet UI-kit template (effect.lua, palette.lua, rectcut.lua, theme.lua, ui.lua, ui_state.lua, skin_rough.lua, skin_smooth.lua), aggregated by ricochet/init.lua and booted once from main.lua:35-41 (ricochet_setup({mode='smooth', palette='light', effects=false})). The site only keeps a thin slice of it alive: the ui_layer it creates, the ui_state hot/active tracking and the low-level ui_text{...} primitive (used directly in sidebar.lua, e.g. sidebar.lua:72-73), and rectcut. Its effect system is explicitly disabled (effects=false, main.lua:36-40, to avoid a shader-compile stall), and its higher widgets/skins (ui_button, ui_panel, ui_slider, ricochet/skin_rough.lua, ricochet/skin_smooth.lua, ricochet_theme_set) are never called anywhere in the actual renderer source (elements.lua, homepage.lua, sidebar.lua, canvas.lua, main.lua all draw with raw layer_rectangle/layer_text/layer_rounded_rectangle and the site's own theme.lua/prose table instead) — confirmed by grepping for those symbols across the renderer tree, where every hit is inside data/*.lua (converted blog-post content that happens to talk about Ricochet, not code that uses it). So renderer/ricochet/ is carried-over UI-kit scaffolding from the ricochet-template fork, mostly vestigial for the site's own visual system, which the site's theme.lua/elements.lua fully own.