agent

+0:18
reasoning

Reading the specified file portion, it appears to contain further HTML content, likely related to the overall webpage structure, and additional markdown files. The intention is to gather all relevant context from the specified locations for analysis. Determining the relationship between these files is critical to understanding the overarching design.

  • read_file{"EndLine":428,"Path":"~/context/reflect/sortie.html","StartLine":232}

    input

    {
      "EndLine": 428,
      "Path": "~/context/reflect/sortie.html",
      "StartLine": 232
    }

    result, 198 lines

    Showing lines 232-428 of 428
    232:     </div>
    233:   </div>
    234: </section>
    235: 
    236: <section class="part" id="research">
    237:   <header><div class="tag">Part 2</div><h2>How mechs get made</h2></header>
    238:   <p class="note">How far to trust each point: <span class="c s">source</span> I read the page. <span class="c e">excerpt</span> I only saw a search excerpt, the site blocked me. <span class="c g">guess</span> my own reading. Japanese pages were read through machine translation.</p>
    239: 
    240:   <div class="rows">
    241:     <div class="row"><h3>The design comes first</h3><div><ul>
    242:       <li>Yoji Shinkawa starts with <span class="q">many doodle-like, rough sketches</span>. <a href="https://blog.playstation.com/?p=207343">PlayStation Blog</a> <span class="c s">source</span></li>
    243:       <li>Kanetake Ebikawa, who designed the Exia and 00 Gundam, holds silhouette and character first and leaves small details for later. <a href="https://tamashiiweb.com/topics/detail/t_kokkaku_61.html?wovn=en">Tamashii Web</a> <span class="c s">source</span></li>
    244:       <li>Some think with their hands. Kunio Okawara presented with wooden models because it was <span class="q">quicker</span>. <a href="https://gundamunofficial.com/production/kuniookawara_walker.html">Translated interview</a> Titanfall's Ogre began as a kitbashed maquette, and <span class="q">physical restrictions guide the look of the art</span>. <a href="https://www.cgchannel.com/2014/04/qa-titanfall-lead-artist-joel-emslie/">CG Channel</a> <span class="c s">source</span></li>
    245:       <li>BattleTech: iPad sketch, VR blockout, SketchUp concept, then a contract 3D artist. <a href="https://forum.paradoxplaza.com/forum/threads/dev-diary-15-sharkys-machine.1292614/post-26072144">HBS diary</a> <span class="c e">excerpt</span></li>
    246:       <li>One solo Blender mech: about 12 concepts, one developed, blockout <span class="q">60% of the entire job</span>, 3 to 4 focused weeks. <a href="https://80.lv/articles/breakdown-futuristic-stalker-mech-made-with-blender">80 Level</a> <span class="c s">source</span></li>
    247:       <li>Code Geass's Lancelot siN: about a year and 15 takes, matching hand-drawn four-view sheets. Directors often didn't specify the final look, so the CG team found it by trial and error. <a href="https://cgworld.jp/feature/201903-codegeass1.html">CGWORLD 1</a>, <a href="https://cgworld.jp/feature/201903-codegeass3.html">3</a> <span class="c s">source</span></li>
    248:     </ul></div>
    249:       <div class="means"><b>For sortie</b>Modelling is translation of a design that already exists, and we have nothing to translate. Your difficulty naming the look is normal: professional directors have it too. Their answer is many cheap tries.</div></div>
    250: 
    251:     <div class="row"><h3>Silhouette and proportion</h3><div><ul>
    252:       <li>Proportion is a character choice. Ebikawa on one design: <span class="q">Hand is large, the foot is small, the body is long</span>. <a href="https://tamashiiweb.com/topics/detail/t_kokkaku_61.html?wovn=en">Tamashii Web</a> <span class="c s">source</span></li>
    253:       <li>Kimitoshi Yamane says Yoshikazu Yasuhiko's mecha <span class="q">win you over with their form and weight</span>. <a href="https://www.gundam-the-origin.net/en/special/interview08.html">The Origin</a> <span class="c s">source</span></li>
    254:       <li>Our own measures show three body plans. Line mech: crotch .53 to .57, knee .39 to .43. GN-X: crotch .63. Knightmare: crotch .50, knee .28, hands .40 to .50, hunched. docs/light-frames.md, docs/study-proportions.md <span class="c s">source</span></li>
    255:     </ul></div>
    256:       <div class="means"><b>For sortie</b>Aiming at the middle of a range gives an average machine <span class="c g">guess</span>. Pick one body plan per maker and push it.</div></div>
    257: 
    258:     <div class="row"><h3>Surface and detail density</h3><div><ul>
    259:       <li>Work in order: main forms, secondary forms, details. Details add function, story and scale. Going back late <span class="q">usually makes it a little bit worse</span>. <a href="https://80.lv/articles/hard-surface-modeling-tips-and-tricks">80 Level</a> <span class="c s">source</span></li>
    260:       <li>Yamane aims for <span class="q">an appropriate and satisfying level of density</span> and avoids <span class="q">a sort of nervous aesthetic</span>. <a href="https://www.gundam-the-origin.net/en/special/interview08.html">The Origin</a> <span class="c s">source</span></li>
    261:       <li>It depends on style. GQuuuuuuX's team kept adding detail and digital markings. <a href="https://www.animenewsnetwork.com/interview/2025-06-20/cgi-veteran-hiroyasu-kobayashi-breaks-down-how-he-mech-ed-the-magic-happen-in-mobile-suit-gundam-/.224743">ANN</a> The Geass team found more detail wasn't always right. <a href="https://cgworld.jp/feature/201903-codegeass3.html">CGWORLD</a> <span class="c s">source</span></li>
    262:       <li>Game standard for clean plates: real bevels plus weighted normals, often with no bake. One Polycount poster says it removes about 90 percent of the unwanted gradients on flat faces. <a href="https://polycount.com/discussion/comment/2577181">Polycount</a> <span class="c e">excerpt</span>, <a href="https://docs.blender.org/manual/en/5.1/modeling/modifiers/normals/weighted_normal.html">Blender manual</a> <span class="c s">source</span></li>
    263:       <li>Katoki's Ver.Ka kits stand apart largely through decals and warning labels. <a href="https://www.gunpla101.com/version-ka-101">Hobby guide</a> <span class="c e">excerpt</span></li>
    264:     </ul></div>
    265:       <div class="means"><b>For sortie</b>The kit uses weighted normals nowhere. Maker plates have seams off (<code>kit/makers/shell.py:107</code>) because the rule said detail comes from decals, and the makers have no decals, only a service band and stencil blocks. Bare plates are my top suspect for "plastic" and "bland" <span class="c g">guess</span>. This is the cheapest thing to test.</div></div>
    266: 
    267:     <div class="row"><h3>Joints and armour</h3><div><ul>
    268:       <li>Harness moves on overlapping lames with leather straps and sliding rivets, and mail covers the gaps. <a href="https://www.medievalcollectibles.com/glossary/terminology/armor-terminology/">Glossary</a>, Met catalogue notes <span class="c e">excerpt</span></li>
    269:       <li>One armourer notes that many visible rivets connect to nothing. <a href="https://www.ageofarmour.com/education/gorget.php">Age of Armour</a> <span class="c e">excerpt</span></li>
    270:       <li>The Geass rig was kept simple on purpose, and the hand rig was the hardest, since a robot hand can't close like a fist. <a href="https://cgworld.jp/feature/201903-codegeass1.html">CGWORLD</a> <span class="c s">source</span></li>
    271:       <li>Designs change once things move. Shinkawa revised details after seeing in-game movement, and Titanfall's designs were changed for the animators. <a href="https://blog.playstation.com/?p=207343">PlayStation Blog</a>, <a href="https://www.cgchannel.com/2014/04/qa-titanfall-lead-artist-joel-emslie/">CG Channel</a> <span class="c s">source</span></li>
    272:       <li>Gunpla kits let plates slide and rotate with the joint so a pose still looks closed. I found no good source today. <span class="c g">guess</span></li>
    273:     </ul></div>
    274:       <div class="means"><b>For sortie</b>I can't find a range-of-motion sheet in any round. Add reach, crouch and boost poses before any detail. That would catch the shoulder bar problem early, and a mechanism only has to look believable.</div></div>
    275: 
    276:     <div class="row"><h3>Scale</h3><div><ul>
    277:       <li>Pacific Rim: <span class="q">to communicate huge scale, you want to move things slowly</span>, with <span class="q">little hits, and hard stops with mechanical recoil</span> in the small parts. <a href="https://artofvfx.com/?p=4715">Art of VFX</a> <span class="c s">source</span></li>
    278:       <li>An action figure beside the maquette made the Ogre read as 20 feet tall. <a href="https://www.cgchannel.com/2014/04/qa-titanfall-lead-artist-joel-emslie/">CG Channel</a> <span class="c s">source</span></li>
    279:       <li>The Origin sized ships with human models: <span class="q">If the window is this big, then how many meters high is the bridge?</span> <a href="https://www.gundam-the-origin.net/en/special/interview09.html">The Origin</a> <span class="c s">source</span></li>
    280:     </ul></div>
    281:       <div class="means"><b>For sortie</b>We already have pilots, a hangar and a catapult. Put a pilot in every hero shot, add hatches and rungs at human size, and let small parts snap while big masses drift.</div></div>
    282: 
    283:     <div class="row"><h3>Anime and real time</h3><div><ul>
    284:       <li>Lancelot siN: flat fills, line weight painted in large textures, limbs stretched to fit each layout. Some other cuts had about 10 staff retouching frames in Photoshop. <a href="https://cgworld.jp/feature/201903-codegeass1.html">CGWORLD 1</a>, <a href="https://cgworld.jp/feature/201903-codegeass3.html">3</a> <span class="c s">source</span></li>
    285:       <li>A game can't cheat per shot, so the model must hold from every angle. Our chase camera mostly sees the back. <span class="c g">guess</span></li>
    286:     </ul></div>
    287:       <div class="means"><b>For sortie</b>Anime references flatter a front three-quarter drawing. Judge from the chase camera first. That may be why backs keep coming up.</div></div>
    288:   </div>
    289: 
    290:   <div class="sub">
    291:     <h3>Your four influences</h3>
    292:     <div class="scroll"><table class="compare">
    293:       <thead><tr><th></th><th>What I could source</th><th>My reading <span class="c g">guess</span></th></tr></thead>
    294:       <tbody>
    295:         <tr><th>Gundam 00</th><td>Designs by Takayuki Yanase, Ebikawa and others (<a href="https://mechabay.com/gundam-00">Mechabay</a>). Our measure: waist a tenth of height, legs near 60 percent, limbs of three or four overlapping plates over a dark inner frame. No designer interview found on why.</td><td>Long straight lines, sharp wedges and thin plates. The frame does show here, as a dark slim core. Ours reads as grey scaffolding.</td></tr>
    296:         <tr><th>Code Geass</th><td>On the Re;surrection film: Akira Yasuda's original designs, Eiji Nakata's main frames, Kenji Teraoka's cockpits (<a href="https://cgworld.jp/feature/201903-codegeass1.html">CGWORLD</a>). Our measure in docs/light-frames.md.</td><td>The character is in the stance: a hunch, long arms, skating on wheels. Our sheets stand upright, which removes it.</td></tr>
    297:         <tr><th>Front Mission</th><td>Modular: body, two arms, legs (<a href="https://www.mobygames.com/game/13788/front-mission/?s=date">MobyGames</a>). Shinkawa drew one new wanzer for Left Alive. No credits for the classic wanzers found.</td><td>Military vehicles first. Mismatched parts must still look right, so it's the best model for your parts system.</td></tr>
    298:         <tr><th>Lancer</th><td>A baseline maker plus four big makers (<a href="https://lancer.wiki.gg/wiki/The_Big_Four">wiki</a>). No source on its art rules.</td><td>Useful for maker structure, as docs/lore.md already says. It teaches little about 3D form.</td></tr>
    299:       </tbody>
    300:     </table></div>
    301:   </div>
    302: 
    303:   <div class="two">
    304:     <div class="sub">
    305:       <h3>Where code-built falls down</h3>
    306:       <ul>
    307:         <li><b>Differences you can't see.</b> Kate Compton's "ten thousand bowls of oatmeal": unique on paper, the same to the eye. <a href="https://gamedeveloper.com/design/practical-procedural-generation-for-everyone-">Game Developer</a> <span class="c s">source</span> Our heavies carry up to 73 plates, and Codex still says medium and heavy look alike at 2 km on two of the four lines.</li>
    308:         <li><b>Seeing.</b> Vision models struggle with Blender tasks that are <span class="q">relatively easy for human Blender users</span>. <a href="https://arxiv.org/abs/2504.01786v1">BlenderGym</a> <span class="c e">excerpt</span> A similar system's stated limit is missing subtle spatial artifacts. <a href="https://api.emergentmind.com/papers/2508.08228">LL3M</a> <span class="c e">excerpt</span></li>
    309:         <li><b>The vocabulary shows through.</b> Two-view shells look like slabs, lofts look like tubes. Each kit has a look. <span class="c g">guess, from our rounds</span></li>
    310:         <li><b>No hand.</b> An artist pushes a form until it's right. I change a number and wait minutes.</li>
    311:         <li><b>Thin geometry breaks.</b> Coplanar slivers on the light shins, a pinhole on the knee wings.</li>
    312:       </ul>
    313:     </div>
    314:     <div class="sub">
    315:       <h3>Where it wins, so keep it</h3>
    316:       <ul>
    317:         <li>The joint contract, mount pads, material roles and livery swaps.</li>
    318:         <li>Rebuilding every model after one frame change, with byte-identical proof.</li>
    319:         <li>Effects and shaders, as the reel showed.</li>
    320:         <li>Variation you then curate. Compton kept the Spore planet seeds that looked good and threw out the rest: <span class="q">You don't need to 'do it right', just do what works.</span> <a href="https://gamedeveloper.com/design/practical-procedural-generation-for-everyone-">Game Developer</a> <span class="c s">source</span></li>
    321:       </ul>
    322:     </div>
    323:   </div>
    324: </section>
    325: 
    326: <section class="part" id="tools">
    327:   <header><div class="tag">Part 3</div><h2>Tools</h2></header>
    328:   <div class="prose"><p><b>You need to buy nothing.</b> The first four rows need no install and are what I'd use. The well-known Blender hard-surface add-ons I could check are built for a person with a mouse, so they barely help a scripted pipeline.</p></div>
    329:   <div class="scroll"><table class="tools">
    330:     <thead><tr><th>Cost</th><th>Tool</th><th>What it does</th><th>Fit here</th><th>Licence, signs of life</th><th>Verdict</th></tr></thead>
    331:     <tbody>
    332:       <tr class="band"><td colspan="6">Free, already here</td></tr>
    333:       <tr><td class="cost">$0</td><th><a href="https://docs.blender.org/manual/en/5.1/modeling/modifiers/normals/weighted_normal.html">Weighted Normal modifier</a> with Bevel</th><td>Keeps flat faces flat and puts the highlight on the bevel</td><td>Scriptable headless. Unused in the kit today.</td><td>GPL, in Blender. Manual current at 5.x.</td><td class="v-try">Try first</td></tr>
    334:       <tr><td class="cost">$0</td><th><code>lib.subd</code> with creases</th><td>Domes and swelling plates with crisp edges</td><td>Ours, already built. Your approved heads came from this line of work.</td><td>In the repo</td><td class="v-try">Default for hero plates</td></tr>
    335:       <tr><td class="cost">$0</td><th><a href="https://docs.blender.org/manual/en/latest/modeling/geometry_nodes/volume/index.html">SDF grid nodes</a></th><td>Mesh to SDF, boolean, offset and fillet, then back to mesh: forms that blend into each other</td><td>Listed for 5.2 LTS, we run 5.2.3. Graph built from Python. Thin plates may need fine grids <span class="c g">guess</span>.</td><td>GPL, in Blender. I read the node list only.</td><td class="v-try">One-day test on a shoulder</td></tr>
    336:       <tr><td class="cost">$0</td><th>SVG silhouette sheets</th><td>Dozens of rough outlines in minutes</td><td>No install. The kit already takes 2D outlines.</td><td>Ours to write</td><td class="v-try">Use for the sketch step</td></tr>
    337:       <tr class="band"><td colspan="6">Free, to install</td></tr>
    338:       <tr><td class="cost">$0</td><th><a href="https://github.com/gumyr/build123d">build123d</a></th><td>Python CAD with true fillets and chamfers</td><td>Scriptable, but outside Blender. The page lists STL and STEP export. Complex fillets may fail <span class="c g">guess</span>.</td><td>Apache 2.0. 3.6k stars, 3,454 commits.</td><td>Only if bevels fall short</td></tr>
    339:       <tr><td class="cost">$0</td><th><a href="https://github.com/ahujasid/blender-mcp">BlenderMCP</a></th><td>Lets an agent drive a live Blender and see viewport shots</td><td>Could shorten the loop. Adds a live session, and telemetry to switch off.</td><td>MIT. About 12k stars per one listing.</td><td class="v-no">Skip for now</td></tr>
    340:       <tr><td class="cost">$0</td><th><a href="https://extensions.blender.org/add-ons/hardflow/">Hardflow</a></th><td>Draw-to-cut booleans, snapping, decals, kitbash</td><td>Mouse-driven. No scripting described.</td><td>GPLv3+. 1.22.0, Oct 2026 note, 7,504 downloads.</td><td class="v-no">Only if you want to model by hand</td></tr>
    341:       <tr><td class="cost">$0</td><th><a href="https://www.renderjuice.com/blender/extensions/bool-tool">Bool Tool</a></th><td>Boolean shortcuts</td><td>The kit already calls booleans.</td><td>GPL. About 1M downloads.</td><td class="v-no">Skip</td></tr>
    342:       <tr><td class="cost">$0</td><th><a href="https://github.com/hlorus/CAD_Sketcher">CAD Sketcher</a></th><td>Constraint-based 2D sketches</td><td>Interactive. Calls itself experimental.</td><td>GPL-3.0. Needs Blender 5.0+.</td><td class="v-no">Skip</td></tr>
    343:       <tr><td class="cost">$0</td><th><a href="https://github.com/microsoft/TRELLIS.2">TRELLIS.2</a></th><td>One image to a textured 3D mesh</td><td>Needs an NVIDIA GPU with 24 GB, and C1 has a Radeon. A fused mesh won't ride our joints <span class="c g">guess</span>.</td><td>MIT.</td><td class="v-no">Can't run here</td></tr>
    344:       <tr><td class="cost">$0 if it runs</td><th>Local image model on C1</th><td>Rough concept sheets for you to react to, private, nothing ships</td><td>Untested on C1, and the GPU is shared. I didn't research models today.</td><td>Depends on the model</td><td>Your call, question 5 below</td></tr>
    345:       <tr class="band"><td colspan="6">Paid, none recommended</td></tr>
    346:       <tr><td class="cost">$38</td><th><a href="https://superhivemarket.com/products/hard-ops--boxcutter-ultimate-bundle">Hard Ops / Boxcutter</a></th><td>The standard hard-surface toolset</td><td>Menus, pies and clicks. No scripting mentioned.</td><td>GPL. Blender 3.5 to 5.2.</td><td class="v-no">Only for a human modeller</td></tr>
    347:       <tr><td class="cost">$44.99</td><th><a href="https://superhivemarket.com/products/meshmachine">MESHmachine</a></th><td>Chamfer and fillet tools, normal detail plugs</td><td>Page doesn't say how it's driven.</td><td>GPL. Blender 4.2 to 5.2.</td><td class="v-no">No</td></tr>
    348:       <tr><td class="cost">From $54.99</td><th><a href="https://superhivemarket.com/products/decalmachine">DECALmachine</a></th><td>Mesh decals in Blender</td><td>The game already places decals at runtime.</td><td>GPL. Blender 4.3 to 5.1.</td><td class="v-no">No</td></tr>
    349:       <tr><td class="cost">$20 to about $70</td><th><a href="https://blendermarket.com/products/fluent/ratings?page=5">Fluent</a></th><td>Interactive boolean modelling</td><td>Same problem as Hard Ops <span class="c g">guess</span>. Listings conflict on price.</td><td>Not checked</td><td class="v-no">No</td></tr>
    350:       <tr><td class="cost">$29, or $69 commercial</td><th><a href="https://superhivemarket.com/products/robot-designer">Robot Designer</a></th><td>Geometry Nodes robot generator, 800+ sliders, 25 presets</td><td>Someone else's look, with a real oatmeal risk.</td><td>Terms mix GPL with resale limits. Published about 7 months ago, Blender 4.5 to 5.2.</td><td class="v-no">No</td></tr>
    351:       <tr><td class="cost">Not priced</td><th>A concept artist</th><td>One real four-view sheet to model against</td><td>It's the step every source has and we lack.</td><td></td><td>Only if the free week fails, and only on your go</td></tr>
    352:     </tbody>
    353:   </table></div>
    354: </section>
    355: 
    356: <section class="part" id="forward">
    357:   <header><div class="tag">Part 4</div><h2>When you start this again</h2></header>
    358: 
    359:   <div class="two">
    360:     <div class="sub">
    361:       <h3>What changes</h3>
    362:       <ul>
    363:         <li><b>Design and building become separate steps.</b> Rough and many first, then model one.</li>
    364:         <li><b>One hero at a time, judged moving.</b> Clips are the review. Stills are for diagnosis.</li>
    365:         <li><b>You veto, you don't choose.</b> I bring variety and ask "what's off?"</li>
    366:         <li><b>Surface gets equal time with outline.</b> Weighted normals, sculpted plates, designed seams, real decals, one settled look.</li>
    367:         <li><b>Rules come last.</b> <code>docs/visual-language.md</code> goes back to draft until something lands.</li>
    368:         <li><b>The critic changes job.</b> I ask Codex "what's the one idea here, and what feels cheap?" It never decides done. Your friends see clips too, since they gave useful reads early on.</li>
    369:       </ul>
    370:     </div>
    371:     <div class="sub">
    372:       <h3>What I'd ask you, one at a time</h3>
    373:       <ol>
    374:         <li>Think of a moment in any game or show when a mech made you feel "that's it". What was it doing?</li>
    375:         <li>In your ideal first hour of sortie, where's your mech on screen when it matters most?</li>
    376:         <li>You've asked about the inner frame three times. Do you want to see it at all, and where?</li>
    377:         <li>Is Gundam 00 still the target for the plain line mech, or did the knight board move that?</li>
    378:         <li>Are rough private AI sketches fine for this, and do you already have an image tool you like?</li>
    379:       </ol>
    380:       <p class="note">No boards or menus during this. I'd play your answers back in your words and keep one short settled / open note. The clean-or-worn question gets answered by the reel, so I wouldn't ask it.</p>
    381:     </div>
    382:   </div>
    383: 
    384:   <div class="sub">
    385:     <h3>The first week</h3>
    386:     <div class="scroll"><table class="week">
    387:       <thead><tr><th>Day</th><th>What happens</th><th>Your part</th><th>Done when</th></tr></thead>
    388:       <tbody>
    389:         <tr><th>1. Interview</th><td>The questions above. No building.</td><td>About 30 minutes</td><td>A short settled / open note</td></tr>
    390:         <tr><th>2. Reel</th><td>No modelling. A 10-second clip of each existing mech (striker, pattern, all four knight lines): turn, boost, stop, plus one shot next to a pilot. Same clip in clean and worn looks. Makers need a small hook-up to the thruster bench.</td><td>Watch, say what's off, send it to friends if you like</td><td>The look is settled, and we know which mech already has some pull</td></tr>
    391:         <tr><th>3. Sketches</th><td>30 or so rough silhouettes in SVG, pushed hard, none averaged from references. Rough AI sheets too if you said yes.</td><td>Cross out what's dead</td><td>Three survivors, no scoring</td></tr>
    392:         <tr><th>4. Surface test</th><td>One chest and shoulder from a survivor, built with weighted normals, sculpted plates, the SDF fillet, drawn seams and decals. Shown at the chase camera and close up, moving.</td><td>Say which looks expensive and which looks plastic</td><td>One method chosen, the rest dropped</td></tr>
    393:         <tr><th>5. Hero</th><td>One mech at one grade, with a reach, crouch and boost check before detail. Shown only as a reel.</td><td>React as you did to the thrusters</td><td>You say "that's done" about some of it, or name what's off</td></tr>
    394:       </tbody>
    395:     </table></div>
    396:     <p class="note">You've since chosen to start over with one mech from scratch, so the plan at the top replaces this week. The test for the week was one part of one mech reaching the thruster reel's feeling. Grades, other makers and rules wait until then. The open shared shoulder bar question waits on your answer to question 3.</p>
    397:   </div>
    398: 
    399:   <div class="two">
    400:     <div class="sub">
    401:       <h3>What I'd stop doing</h3>
    402:       <ul>
    403:         <li>Numbered polish rounds on one model</li>
    404:         <li>Building twelve models before one is loved</li>
    405:         <li>Turning one comment into a rule for every maker</li>
    406:         <li>Putting ratios and plate counts in your reports</li>
    407:         <li>Grey sheets of small tiles as the verdict</li>
    408:         <li>Asking you to pick, or carrying a rolling list of naming decisions in every report</li>
    409:         <li>Filling spare usage with systems while the look is open</li>
    410:         <li>Long terminal reports: one page, clip first, one question at the end</li>
    411:       </ul>
    412:     </div>
    413:     <div class="sub">
    414:       <h3>Risks I can't remove</h3>
    415:       <ul>
    416:         <li><b>Code-built may cap out.</b> My guess is that it holds at the chase camera and falls short in hangar close-ups. If the week's hero doesn't land, I'll say so plainly. The next step would cost money: one commissioned four-view sheet, not priced, on your go only.</li>
    417:         <li><b>Sketches can promise more than the kit can build.</b> So day 4 comes before day 5, and only proven methods go on the hero.</li>
    418:         <li><b>I can't watch video.</b> I judge frames. You and your friends are the only judges of motion.</li>
    419:       </ul>
    420:     </div>
    421:   </div>
    422: 
    423:   <p class="note"><b>What I couldn't read.</b> Paradox forum (HBS diaries), Polycount, the Met and Age of Armour blocked direct fetches, so those are search excerpts. Kate Compton's original zine returned 403, so I used Game Developer's summary. CGWORLD and Tamashii Web were machine translated. I found no designer's own account of Gundam 00's shape choices, no source on Armored Core 6's or Lancer's art rules, and nothing usable on Gunpla inner frames. Prices are from listing pages today and can change.</p>
    424: </section>
    425: 
    426: </div>
    427: </body>
    428: </html>
  • read_file{"Path":"~/work/sortie/docs/sizes.md"}

    input

    {
      "Path": "~/work/sortie/docs/sizes.md"
    }

    result, 89 lines

    Showing lines 1-88 of 88
    1: # Frame sizes
    2: 
    3: Three size classes. Every class keeps the joint names in `src/render/mech.ts` (JOINTS), so poses, part
    4: swapping and the rig work on all of them. A frame may add joints of its own (harkens, spinners) but
    5: never renames or drops one.
    6: 
    7: | class    | height  | hip    | reads like                                  | hangar         | role                                              |
    8: |----------|---------|--------|---------------------------------------------|----------------|---------------------------------------------------|
    9: | light    | 6 m     | ~54%   | Code Geass knightmare, Front Mission wanzer | half a bay     | agile, close range, harkens, hull and station fights |
    10: | standard | 17.6 m  | 60%    | Gundam 00 mobile suit (the striker)         | one bay        | all round                                         |
    11: | heavy    | 24 m    | ~50%   | Lancer size 2 to 3, heavy wanzer            | one reinforced bay | armour, heavy mounts, slow turns              |
    12: 
    13: The merc ship's launch tube is 18 m wide and 26 m tall and the bay gantries stand 26 m, so 24 m is the
    14: ceiling for anything a merc can launch. Bigger machines belong to factions with carriers.
    15: 
    16: A bay is 12 m by 14 m: two light frames fit side by side. A heavy frame fills a bay edge to edge and
    17: needs the reinforced bay upgrade, which is a hangar-versus-research trade like every other upgrade.
    18: 
    19: Each bay is fitted for one class (`kit/ship.py`, `src/sim/hangar.ts`). The light fit holds two cradles
    20: with boarding platforms at 4.4 m, under the core block, and parks the work decks up high. The standard
    21: fit has work decks at 8 and 15.5 m. The reinforced fit jackets and braces the posts, adds clamp arms at
    22: the torso and moves the decks to 11 and 21 m, with a deeper back deck. On the catapult a light rides a
    23: sled on the striker clamps and a heavy gets wider clamps on outriggers. A heavy crouches deeper than
    24: the striker to pass under the tube's signal box, and every frame sinks onto its clamps. Try a hangar
    25: with `?hangar=light,light,heavy`.
    26: 
    27: ## Proportions
    28: 
    29: - **Light.** Near human. Hip at about 54% of height, small head (about 1/8 of height), narrow waist,
    30:   long shins, slim arms. The pilot does not sit in the chest: the cockpit is a core block in the back
    31:   of the torso, and it dominates the back and side silhouette.
    32: - **Standard.** The striker: hip at 60%, Gundam 00 slim and angular.
    33: - **Heavy.** Hip at about 50%, shoulders about half the height wide, a small head sunk between the
    34:   shoulders (heads.md principle 3), thick thighs, wide stable feet, layered armour, a big reactor pack.
    35: 
    36: Detail density stays the same in real size, so a light frame has fewer, larger plates and a heavy
    37: frame has more of them. A shrunk striker is not a light frame and a grown one is not a heavy frame.
    38: 
    39: ## Code Geass, adapted
    40: 
    41: Code Geass joins Gundam 00, Lancer and Front Mission as an influence. We take traits, not designs: no
    42: knightmare names and no lookalikes of a specific frame.
    43: 
    44: - **Core block.** The pilot rides a bike-style seat inside an armoured block at the back of the torso.
    45:   The block ejects as a whole.
    46: - **Harkens.** Rocket-fired wire anchors, a pair in the chest sides or hips, optionally a pair at the
    47:   wrists. In space they anchor to hulls, rock and other frames: swing round an anchor without fuel,
    48:   reel in, pull a target off line.
    49: - **Spinners.** Wheel units at the heels. Here they are magnetic hull wheels for skating along ship
    50:   hulls and station surfaces.
    51: - **Factsphere.** An armoured sensor behind shutters that open for a scan mode.
    52: - **Knightly armour.** Pointed crests and horns (never a V-fin), tassets over the hips, a narrow
    53:   waist, angled pauldrons.
    54: - **Float pack.** A winged thruster pack. A light frame needs one to fight in space.
    55: 
    56: Harkens and factspheres are not light-only: a standard or heavy frame can mount them as equipment.
    57: 
    58: ## Gameplay (provisional)
    59: 
    60: - Light frames accelerate and turn faster, have a lower top speed and a shorter boost.
    61: - Heavy frames turn slowly and carry more armour and more mounts. Top speed in a straight line is
    62:   close to standard.
    63: - The hit radius scales with height.
    64: 
    65: ## In the sim
    66: 
    67: Classes live in `src/sim/frames.ts`, harkens in `src/sim/harken.ts`. Try a class with `?frame=light` or
    68: `?frame=heavy`: the light flies the knight (`kit/light.py`), the heavy flies frame A (`kit/heavy.py`).
    69: `?variant=line|lancer` or `?variant=B|C` picks another. Each class has its own muzzle and harken mount
    70: in `frames.ts`, measured from the posed model (`__sortie.mounts()` in the game prints them). Weapons are
    71: striker sized and a frame holds them scaled to its height.
    72: 
    73: | class    | hit radius | hp  | accel | top speed | boost | yaw rate | boost lasts | harken wire | reel   |
    74: |----------|------------|-----|-------|-----------|-------|----------|-------------|-------------|--------|
    75: | light    | 1.8 m      | 60  | 96    | 118 m/s   | 250   | 2.3 rad/s | 2.1 s      | 120 m       | 48 m/s |
    76: | standard | 5.3 m      | 100 | 70    | 140 m/s   | 320   | 1.5 rad/s | 3.3 s      | 250 m       | 38 m/s |
    77: | heavy    | 7.2 m      | 180 | 46    | 132 m/s   | 300   | 0.95 rad/s | 3.8 s     | 250 m       | 30 m/s |
    78: 
    79: Harkens: `R` or right mouse fires along the aim, and again lets go. Hold `T` to reel in. One anchor
    80: flies at a time, alternating chest sides. It bites drones and rock, retracts at full length and lets go
    81: when its drone dies. On the wire, an idle stick doesn't brake, so momentum swings the frame round the
    82: anchor, and letting go keeps the velocity. Rock blocks frames and shots, and a hard hit costs hp.
    83: 
    84: ## Open questions
    85: 
    86: - Light at 6 m (knightmare scale, a different kind of fight) or about 9 m (closer to the striker)?
    87: - Do the classes share every battlefield, or do some contracts suit one class (boarding for lights)?
    88: - Heavy at 24 m, or smaller so it keeps more room in the tube?
  • read_file{"Path":"~/work/sortie/docs/lore.md"}

    input

    {
      "Path": "~/work/sortie/docs/lore.md"
    }

    result, 154 lines

    Showing lines 1-153 of 153
    1: # Lore
    2: 
    3: The setting as agreed with the owner, 2026-10-25. The wiki page (docs/wiki.html, published at
    4: [redacted: a link to a private page]) carries the same content with colours and
    5: logo slots. Update both together. Names marked (working) are placeholders, candidates live in
    6: the wiki.
    7: 
    8: Influences: Lancer for how powers and makers are arranged, Gundam 00 for blocs in a cold war,
    9: Ace Combat for named aces, Code Geass for knights and their frames. We take structure, not names.
    10: 
    11: ## Scale
    12: 
    13: Galactic. Many star systems, joined by wormhole gates. Earth is open: leaning towards lost,
    14: a star nobody can find, kept for later arcs.
    15: 
    16: ## The Federation and the gates
    17: 
    18: The Federation is the union of the great powers. It runs the gate network. Every great power
    19: sits on the gate council. Gate stations are Federation ground and armed. On paper the guns are
    20: for peacekeeping against raiders, insurgents and alien incursions. Nobody doubts a power would
    21: turn a station's guns on a rival if it came to that.
    22: 
    23: Tension runs between the powers and inside each of them.
    24: 
    25: ## Glass
    26: 
    27: Gate cores are built from glass, the silver material of the aliens. A new gate needs glass,
    28: and glass comes from contact zones on the frontier. Whoever holds the frontier decides where
    29: the next gates go.
    30: 
    31: The aliens act unpredictably because we harvest them. They are a late stage enemy. The alien
    32: contact contract in the slice is there for testing.
    33: 
    34: ## Great powers
    35: 
    36: The groups inside the Federation. Four or five, depending on whether there are one or two
    37: military powers.
    38: 
    39: 1. **The Administration.** Runs the gates and the council's day to day. Gate security, customs,
    40:    the common pattern frames every power licenses.
    41: 2. **Freight and commerce** (working). Shipping and mining lines that became a state. Owns
    42:    more stations than anyone.
    43: 3. **Hardline military** (working). A militarist state with a standing fleet. Believes the
    44:    Federation is too soft on the frontier and on the aliens.
    45: 4. **The houses** (working, the second military power if kept). A peerage of noble houses with
    46:    sworn knights. Personal machines and callsigns. Most factioned aces come from here.
    47: 5. **Science** (working). Gate science and glass research. Built the gate cores, and resents
    48:    that the Administration runs what it built.
    49: 
    50: ## Lesser factions
    51: 
    52: Smaller players on the frontier, each tied to a sector. They fly refits and whatever a patron
    53: slips them, and they are often what an arc is fought over. A handful at launch, more as arcs
    54: open sectors. Proposals:
    55: 
    56: - **Independence movement.** Wants out of the Federation. Insurgent cells, stencilled slogans.
    57: - **Glass miners' cooperative.** Small scale harvesters sitting on contact zones. Everyone
    58:   courts or squeezes them. Work frames with tools.
    59: - **Free port.** A neutral station where mercs hire on, smugglers trade and raiders fence
    60:   loot. Loud mixed liveries.
    61: - **Glass order.** A habitat order that reveres the aliens and opposes harvesting. Wears glass
    62:   shards as ornament. Sometimes sides with the aliens.
    63: - **Exodus fleet.** Refugees from a world cut off by an old council vote. Patched habitat
    64:   plating. Proof of what a gate suspension costs.
    65: - **Company town.** A colony the commerce power abandoned, now running itself on old work
    66:   frames.
    67: 
    68: Raiders are criminals, not a lesser faction, though a lesser faction may fund or harbour a gang.
    69: 
    70: ## The cold war
    71: 
    72: Open war gets a power voted off the gate network, which cuts its fleets and trade off from
    73: every other system. That threat keeps the war cold. So the powers fight by proxy: mercenaries,
    74: gangs funded under the table, incidents at armed stations, insurgents flying suspiciously new
    75: parts.
    76: 
    77: ## Mercenaries
    78: 
    79: Mercs exist because the powers need someone to blame. Every company flies the same class of
    80: medium ship (see the project brief) and takes contracts from any side its standing allows.
    81: 
    82: ## Arcs
    83: 
    84: An arc is one frontier sector in contest, played slowly over real weeks.
    85: 
    86: - Contracts move the balance. Every company's results add to each power's hold on the sector:
    87:   escort a survey, raid a rival's glass haul, defend a station, sabotage a gate project.
    88: - An arc ends in a council vote: a gate opens into the sector for the winner, a power is
    89:   suspended from the network, a station changes hands.
    90: - Results stay. The map changes, contract lines close and open, makers' catalogue prices move
    91:   with their power's fortunes, lesser factions rise or fall.
    92: - Companies on opposite sides of one arc are the long term stakes.
    93: 
    94: ## Aces
    95: 
    96: Aces are shared by the whole playerbase. They never die for good.
    97: 
    98: - **Entrance.** An ace entering an instance gets a title card: callsign, name, power, unit,
    99:   frame. The camera cuts to them in their personal paint.
    100: - **Factioned aces** eject, get picked up by their side and come back soon, usually in the same
    101:   machine, repaired.
    102: - **Rogue aces** answer to no power. If their machine is seized after a fight (towed home by a
    103:   company), they vanish for a while and are sighted again later in a better or different
    104:   frame.
    105: - **Chronicle.** Every ace has a closed record of their story: sightings, fights, who downed
    106:   them, machines lost and seized, arcs they took part in. Players read it, nobody edits it.
    107: - **Players earn cards too.** A company with enough notoriety gets an entrance when it drops
    108:   into a shared instance or turns up on the other side of an arc.
    109: 
    110: The models are Char Aznable and Graham Aker: rivals who keep coming back, sometimes in a new
    111: machine, sometimes under a new name.
    112: 
    113: ## Automatons and their cores
    114: 
    115: Automatons carry a guidance core and a charge in the chest. Past a damage threshold they
    116: self-destruct: some at range, some by throwing themselves at a piloted frame or a defence
    117: objective. A company that seizes one in that state safely, by a means still to design, gets
    118: the core.
    119: 
    120: ## Research sources
    121: 
    122: Rare tiers come from what a company hauls home.
    123: 
    124: - **Automaton cores:** guidance and electronic warfare. Hacking, EMP, better homing, drones,
    125:   funnels, auto salvagers, an assistant pilot (chaff against missiles, funnel shields against
    126:   beams, the merc ship's guns, commands to the ship mid fight), and late stage whole
    127:   autonomous NPC pilots.
    128: - **Glass (late):** alien physics. Temporal tech such as a local slow field or a short blink,
    129:   field shields.
    130: - **The powers, by standing:** their own lines. Suppressed ballistics fit the proxy war:
    131:   deniable work where a hit is harder to trace back to the company.
    132: 
    133: The assistant and NPC pilots loosen an earlier rule (the player flies their own mech and
    134: nothing more). They stay late stage, research only.
    135: 
    136: ## Weapons in hand
    137: 
    138: Proposal, to pin down with the parts system: keys 1 to 4 pick the active mount (right hand,
    139: left hand, right shoulder, left shoulder) like an FPS, the mouse fires it. A utility (chaff,
    140: EMP burst, hack) sits on its own key. How many mounts a frame has may depend on its class and
    141: parts.
    142: 
    143: ## Open
    144: 
    145: - The game's name. Sortie is a working name.
    146: - Final names for the powers, lesser factions and makers.
    147: - One military power or two.
    148: - How many lesser factions at launch.
    149: - Where the Foundry (the automatons' maker) fits.
    150: - How a company safely seizes an automaton in its critical state.
    151: - Earth: lost or a protected heritage world.
    152: - Whether a rogue ace can return under a new callsign, with the chronicle linking the two once
    153:   enough players have seen both.
  • read_file{"Path":"~/work/sortie/docs/thrusters.md"}

    input

    {
      "Path": "~/work/sortie/docs/thrusters.md"
    }

    result, 69 lines

    Showing lines 1-68 of 68
    1: # Thruster plumes
    2: 
    3: What games and real rockets do with exhaust, and what sortie takes. Bench: `/?view=thrusters` (see `src/view/thrusters.ts`), sheets: `tools/thruster-sheets.sh`.
    4: 
    5: ## What reads in games
    6: 
    7: None of these studios has published a breakdown of its thruster effects, so this is from watching play, not from their tools.
    8: 
    9: - **Armored Core 6.** A boost is a short, hot flame off the back plus a lot of light on the frame. A quick boost is an event, not a state: a hard flash at the nozzles, a burst of flame that is gone in a fraction of a second, and the frame lurching. Assault boost has a distinct ignition (a bright bang and a ring of hot gas off the main booster) and then a long sustained flame. Booster parts change the flame colour, so colour is a part trait. ([booster parts](https://armoredcore6.wiki.fextralife.com/Booster))
    10: - **Gundam games (Battle Operation 2, Extreme Vs).** Blue white, wide and short flames from many small verniers rather than one big plume. Boost dashes and step dodges are sold by the burst at onset and by many nozzles firing at once in different directions. The exhaust reads because the inside of each bell glows.
    11: - **Titanfall jump jets.** Short bursts timed to inputs, a warm core and a trail. The trail is the readability tool for the enemy: the Low Profile kit removes it so the pilot is harder to track. ([Low Profile](https://titanfall.fandom.com/wiki/Low_Profile))
    12: - **Ace Combat afterburners.** A cone with a bright inner ring and shock diamonds, brighter when seen from behind. Modders build them from a volume emitter plus a ribbon for the diamonds. ([afterburner mod notes](https://combatace.com/files/file/12395-improved-afterburner-effects-for-strike-fighters-2))
    13: - **Space games (Elite Dangerous, Star Citizen, Everspace).** A bright nozzle, a short clean plume, small manoeuvring puffs from many directions, colour as a cosmetic per ship. ([Elite engine colours](https://elite.drinkybird.net/store/FORC_FDEV_V_ENGINE_COLOUR_01))
    14: 
    15: ## Real plumes
    16: 
    17: - Shock diamonds come from supersonic exhaust whose pressure does not match the air around it: shocks and expansion fans bounce across the jet and the bright discs (Mach discs) sit where the gas is reheated. They fade as the jet mixes. ([aerospaceweb](https://aerospaceweb.org/question/propulsion/q0224.shtml), [MIT notes](https://web.mit.edu/16.unified/www/SPRING/fluids/Spring2008/LectureNotes/f20.pdf), [F-15E footage](https://theaviationist.com/2020/03/30/watch-shock-diamonds-and-mach-disks-form-in-the-exhaust-plume-of-an-f-15es-engine/))
    18: - In thin air or vacuum the exhaust is underexpanded: it leaves at a higher pressure than outside and balloons out past the bell, much wider and fainter than at sea level, fastest along the axis. ([ESA](https://blogs.esa.int/cleanspace/?p=184), [FIU](https://web.eng.fiu.edu/allstar/rocket3.htm), [KSP 2 dev diary](https://forum.kerbalspaceprogram.com/topic/200605-developer-insights-2-%E2%80%93-visual-effects))
    19: - So true vacuum plumes have no diamonds and no smoke. We keep faint diamonds for boosts anyway because they sell power, and keep the vacuum flare and the lack of smoke.
    20: 
    21: ## Techniques
    22: 
    23: - Layers: a white hot core, a coloured mantle, a soft outer glow. Usually nested cones or cylinders with an additive shader whose brightness, length and noise are curves of throttle. ([Waterfall](https://github.com/ChrisAdderley/Waterfall/wiki/ModuleWaterfallFX))
    24: - Fresnel and end fades so a cone has no hard rim, and a scrolling noise for turbulence. A plain cone shader breaks seen end on, which is exactly how the chase camera sees the main thrusters. ([Godot thruster](https://godotshaders.com/shader/jet-engine-thrustler-shader/), [Godot trail](https://godotshaders.com/shader/futuristic-jet-thruster/))
    25: - Soft particles: fade where a sprite meets geometry, using scene depth. ([NVIDIA](https://developer.download.nvidia.com/whitepapers/2007/SDK10/SoftParticles_hi.pdf))
    26: - Ignition: a flash at the exit, a ring of gas thrown out, the flame overshooting its length then settling.
    27: - Light: a short lived light near the nozzles sells the heat more than the flame does. Bloom does the rest, so only the core should cross the bloom threshold. Heat shimmer needs a screen copy and a distortion pass, too dear for many frames on screen.
    28: 
    29: ## What sortie does
    30: 
    31: - **One small volume per nozzle** (`src/render/plume.ts`): a closed cylinder bounding the plume, and the fragment shader marches up to 10 steps through it summing an analytic density: a narrow core that is white at the exit and the exhaust colour by its tip, shock diamonds that pinch and brighten it in three cells, a mantle that flares out past the exit and breaks into streaks downstream (two value noises, one of them also in the radius), the ignition flash and a soft shock ring. The cylinder grows with the flare and reaches two ring widths past the shock ring, so neither is cut off flat by its wall. The throat glow is a gaussian integrated along the ray in closed form, so it never falls between steps end on. Because it is a volume it reads from any angle with no cone edges, so no fresnel or depth fade is needed.
    32: - **Plume state is pure logic** (`src/render/exhaust.ts`, tested): sustained level per nozzle kind from the flight's thrust (mains forward, binders strafe and lift, calves brake and lift), a pilot flame at rest, fast attack and slower release.
    33: - **Three distinct events.** A boost start flashes and surges the length. A quick boost is a punch: full in 8 ms, held to 45 ms, gone by about 150 ms, wider than a boost, outside the level smoothing so it never smears. An assault boost has its own ignition (the boost one is suppressed), then over 100 ms runs out to a heavy jet and holds it, with diamonds: 1.8 times boost length, 1.45 times the flare, a mantle that starts wider and thins later down the plume, a core 1.45 times as thick, a bigger and brighter throat glow and half again the light.
    34: - **Colour from the palette**: the `glow_exhaust` role sets the mantle, the core runs it to white, glow scales the whole plume. Defaults: blue white for mercs, cyan for the Administration, red for automatons, dirty orange for raiders. `glow_thruster` (the lit throat in the bell) follows the same hue.
    35: - **One light**: the player's frame gets one point light at the plumes' weighted centre, present from the start so shaders compile once. Enemies get none.
    36: - **No smoke** in space, no sparks yet.
    37: 
    38: ## Cost
    39: 
    40: - Draw calls: one per nozzle, as before (6 to 10 a frame). Materials: three per frame (one per nozzle kind), one shader program for all.
    41: - Fragment: up to 10 steps of two 3D value noises (16 hashes) plus a few exps. Past 400 nozzle radii from the camera (about 170 m for a striker) the march drops to 5 steps. A boosting striker at chase distance covers a few thousand pixels, so a screen of 20 frames costs well under a millisecond on a mid GPU. A plume filling a fifth of a 1080p screen (camera in the exhaust) is the worst case, about 8 million noise evaluations a frame.
    42: - Plumes frustum cull against a fixed bound in plume space. Hidden when a kind is not firing.
    43: - The light is one forward light for every lit pixel of the scene. Budget it: one for the player, a fixed pool for the nearest few if enemies ever get one, never one per nozzle.
    44: - No per frame allocation in the step or the uniform upload.
    45: 
    46: ## What is still weak
    47: 
    48: - Plumes are depth tested but the march does not stop at the mech, so where a plume runs through armour (the striker's mains fire down past the rear skirt) the glow shows over the plates. A scene depth texture would fix it, the forward pass has none.
    49: - The striker's mains point down between the legs, so from the chase camera much of the plume hides behind the frame. That is a kit choice, not a shader one.
    50: - No sparks or hot particles, no heat shimmer.
    51: - Turbulence at 10 steps is soft. It reads in motion, not in stills.
    52: 
    53: ## What the sim would need
    54: 
    55: - Quick boost: a time since the last quick boost began (or an impulse counter) on `FlightState`. Its direction comes from the thrust vector.
    56: - Assault boost: a flag on `FlightState`.
    57: - Until then the game plays the boost ignition when `boosting` turns on. The bench drives quick and assault boosts directly.
    58: 
    59: ## Bench
    60: 
    61:     /?view=thrusters&state=boost&t=0.5                 one state, a still at t seconds in
    62:     /?view=thrusters&state=quick&t=0.05&close=main     close on a main nozzle
    63:     /?view=thrusters&state=cycle&play=1                every state in turn, live
    64:     /?view=thrusters&cells=quick:0,quick:0.05,boost:0.5:main:178:4&cols=4&cw=480&ch=300
    65:                                                        a sheet, cells are state:t[:close[:yaw[:pitch]]]
    66:     &frame=heavy_C &palette=raider &yaw=150 &pitch=12 &dist=40 &light=0 &tier=low
    67: 
    68: States: idle, cruise, boost, quick, assault, brake, strafe. `tools/thruster-sheets.sh <dir> <tag>` renders the four review sheets headless.