A spectrum, not a binary
I had Marketing pinned to strict Porcelain. You're describing something wider — waves, navy, luminous glass, dark bands, editorial treatments, all available on the marketing side. That's a better model, and it fixes the seam problem from 05c almost by itself: if marketing has range, the walk from a warm social post to the site is a gradient rather than a step.
Luminous glass means backdrop-filter: blur() and translucent fills — which the absolutely-flat lock forbade. Bringing it back on marketing is a genuine reversal of that decision, not an extension of it. I've scoped it to Campaign, Marketing and Editorial only, and left Product, Console and Video absolutely flat. Say if you'd rather keep glass out entirely and use navy + waves for the range instead.
The ordering isn't taste, it's consequence of error. A campaign card that over-reaches costs you nothing; a dashboard that over-reaches makes someone trust a stale number. Expression is cheap on the left and expensive on the right, so permission narrows in exactly that order.
What still holds on marketing: Cinzel titles the page and Iowan carries the reading — the range is in surface and texture, not in the type hierarchy. Teal remains the only action colour, and the CTA keeps its gold ring. Gold still never fills a button on marketing; that stays a Campaign-only permission, because a gold button on the site would collide with the teal action rule the product depends on.
Five registers on one page and it still reads as one thing: waves kept to .2 low-right behind a hero with a lifted readiness card, a navy band carrying three glass step-cards, a warm #F5EFE6 testimonial in Playfair with the gold italic, and a flat porcelain stat row. What holds it together is that Cinzel titles every section and teal is the only thing you can click — the range moves underneath a fixed hierarchy.
Two materials, one brand
Your call, and I think it's the right one: v1's warm editorial voice keeps social and campaign; Porcelain takes the product and the website. It also resolves the one real loss I flagged in the comparison — the gold-italic quote card was the most ownable thing in v1, and under this split it doesn't die, it just stops being asked to run a dashboard.
The reason it works is that the two materials fail in opposite places. Warm editorial is persuasive — soft shadows, gold italic, generous radii, a hospitable page. That's exactly right for something a stranger meets in a feed, and exactly wrong for a dashboard, where it makes every number look equally trustworthy. Porcelain is precise — flat, engraved, one action colour, trust states. Right for instruments, a little cold for a first impression.
These are corrected against ui_kits/website/components.jsx rather than the guide's prose — the real kit uses Playfair 500 (not 400), ink #0D1B2A (not #142033), and an aina-chip kicker at rgba(201,169,110,.18). Earlier drafts had those wrong.
AI NATIVE ACADEMYOne real cost, worth naming before you commit: somebody clicks the warm social post and lands on the cool website. That's the highest-traffic transition the brand has, and under a two-material system it is a visible jump. It only works if enough carries across that the jump reads as range rather than as two companies.
Never mix materials inside a single surface. A porcelain page does not get one warm-editorial card, and a social post does not get a Cinzel headline with a Playfair subhead. The two systems are allowed to sit next to each other in a funnel; they are not allowed to sit next to each other in a layout. Every hybrid I've seen attempted reads as a mistake rather than a range — and it is the failure mode that will actually happen, because someone will want "just one" gold italic on the homepage.
Mostly it makes it cheaper. The social pack no longer needs rebuilding — wave 03 shrinks to the website and email, and the four canvases in 07 become the Porcelain option for product-led posts (score reveals, cohort results) rather than the replacement for the whole pack. The thirty lesson videos are the interesting case: they're taught content, so they're Product — which means Iowan, not Cinzel, and the re-typeset stays small. And the H1 question narrows to a single surface, the website, because social already has its answer.
AINA Porcelain
A complete parallel design system, forked from the AI Native Academy system rather than edited into it. Same brand, same mark, same voice — a different material. Where v1 is warm editorial with soft shadows and a Playfair serif, v2 is cool porcelain: engraved Cinzel titling, one action colour, and a data-trust contract v1 has no equivalent for. It is absolutely flat where it counts — Product, Console and Video — and gives Marketing a wide expressive band above that floor. It also carries the two things v1 is missing outright — a social system and a motion system.
v1 stays frozen and shippable. Thirty rendered lesson videos, the live site, and the packaging artwork all resolve against it. v2 lives at its own path with its own token file. Nothing under _ds/ is touched — a surface migrates by swapping which token file it imports, and can migrate back.
v1 is organized by product — five silos, each with its own rules. That's why social and video keep drifting: a new format has no silo to belong to. v2 organizes by how loud a surface is allowed to be. Six classes, one matrix. A new format inherits its rules on day one instead of inventing them.
Four waves, cheapest first. Tokens and flat surfaces are a stylesheet change. Type is a re-typeset. The thirty videos are a re-render and go last — and only after the H1 question is settled, because that decides whether video needs Cinzel at all.
You were right that the dark didn't line up. There were three navies in play and I'd picked one without accounting for the others. Laid out honestly, then resolved into a single ladder that keeps all three doing a job.
The brand's existing navy. Also v1's primary text and primary button colour, so it is load-bearing in three places at once. Relative luminance .0105 — the lightest of the three by a full stop.
The only one running in production code. Luminance .0049 — deep enough that gold #E6C883 reads as luminous rather than merely bright.
The most neutral — nearly a true blue-black. Its contribution was the warm cream ink, not the base. That insight survives; the base value doesn't.
Rather than pick a winner and retire the others, the three values stack. Luminance ascends cleanly — .0049 → .0075 → .0105 — so each layer reads as one step above the last, and v1's Deep Navy survives as a real token instead of being dropped. Nothing anyone already shipped gets abandoned.
The cost, stated plainly: #121D30 from the earlier drafts is retired, and v1's Deep Navy changes role — it is no longer a page background or a button fill, it is the top surface layer. Anything in v1 using it as a base maps to #07101D; anything using it as a raised element keeps it.
What survives untouched: the wing mark and full logo lineage, Inter as chrome, the icon sprite geometry, the 12-column grid with 24px gutters, the 620px reading cap, the voice and copy patterns, the Do/Don't vocabulary, Teal 600 #1AA79D, Teal 800 #0E4D4A, Gold 600 #B88E4C, and now Deep Navy #0D1B2A in a new role.
The honest conflict: v1's SKILL.md explicitly rejects this direction — "Cinzel lowercase reads as small-caps." That's correct, and v2 doesn't argue with it: Cinzel is never set lowercase. It is uppercase-only titling, and Iowan takes every job needing mixed case. v2 is a deliberate departure from v1's stated position, not an extension of it.
Porcelain Light
Exact carry-overs in teal. Action is the one genuinely new value — v1's Teal 500 #249487 reaches only 3.4:1 with white so it cannot carry a filled button. #1C6B65 reaches 5.1:1.
Porcelain Dark
Two exact carry-overs on the dark side now: action #1AA79D is v1's Teal 600, and raised #0D1B2A is v1's Deep Navy. Teal never fills on dark — cream does, with the gold ring.
Data-trust states — new in v2
A number must declare whether it can be trusted. Mandatory in product, optional elsewhere, and never colour alone — every state carries a word. These sit outside the teal/gold system and never borrow from it, which is why they don't break the flat lock.
Dark equivalents lift for contrast: #5DD5A8 · #E5BC6F · #E08A7A · .28 ink.
Two open verdicts
v1's one non-system accent, used for the APPLY THIS NOW chip. It's semantically load-bearing, so I'd keep it scoped to that single chip and forbid it spreading — but it is the one colour in v2 belonging to no family.
v1 has three golds; v2 needs two. Gold 500 is the one that goes — it sits between porcelain's light and dark gold and does the job of neither. Anything using it maps to #B88E4C.
AINA helps ambitious professionals turn AI knowledge into practical skills.
AINA helps ambitious professionals turn AI knowledge into practical skills.
Three things to notice. The kicker moves from Inter to Cinzel — v1 forbids the serif in kickers because Playfair at 12px with .14em tracking falls apart; Cinzel is engraved capitals and is at its best there. The headline drops ~18% in size for the same optical weight. And the accent phrase keeps its gold but loses its italic, because gold italic serif is the single strongest tell of the aesthetic this direction moves away from.
Role assignment
Loading & fallback
None of the four faces is a system font everywhere, so every surface declares its loading path and its fallback — the fallback must hold the layout, not just the genre.
The four variants survive as a taxonomy — base light, dark focus, framed editorial, framework diagram. What changes inside them is the material: no shadow, tighter radius, hairline borders, and the type roles above.
Short, practical lessons that explain AI with real examples.
accounts.
Buttons
v1's primary was navy with a deepening shadow on hover. v2's is teal with a rotating 1.5px gold ring — the system's only gradient, and the only place motion appears in product. One per screen, on the single commitment: never on a chat send button, where every turn is primary and the ring stops signalling. The dark secondary is where v1's Deep Navy now earns its keep.
Chips & states
Pills keep 999; the square chip variant moves 10→8. The lavender Apply This Now chip survives as v1's one legacy exception — darkened to #4B51C4 to pass on porcelain.
Read the Campaign column as history. Since 05a assigned Campaign to warm editorial, that column now describes what Porcelain would permit there — useful for product-led posts, but the live campaign rules are v1's. The five columns Porcelain actually governs are Marketing, Editorial, Product, Console and Video.
Invariant across all five: Cinzel titles uppercase, Inter chrome, Plex Mono data, Iowan for anything read in sentences. Teal is the only action and state colour. Gold never on an icon, never on a data series, never as body text.
The flat floor — Product, Console and Video only: no shadow inside a surface, no glass, no italic, and the CTA ring is the one gradient. Marketing and Editorial sit above that floor; see 06b for what they get.
v1 ships a Social & Content Pack as flat reference images with no spec behind them. v2 makes it a format system: fixed canvases, four registers, and one register per post. Canvases below are true aspect at .32. Since 05a gave Campaign to warm editorial, these are the product-led set — score reveals, cohort results, anything that is really a view of the product. Product rules therefore apply: no gold fills and no waves.
in seven weeks.
AINAainative.academyBand 3 of 5. Judgement is where the score is being held down.
AINAainative.academyv1 has motion guidance — "subtle", parallax, an easing curve. v2 inherits a built system from the lesson-video project: six named directions, nine beat templates, a real type scale, and a deterministic render contract. Frames are 1920×1080 at .32 with true type. Video is its own class — the sixth column in 06. A rendered frame is read at a distance and has no interaction, so it takes Product's restraint on colour and texture but Campaign's licence on scale and motion.
AI NATIVE ACADEMY1:301.1AI doesn't fail.
Your brief does.
The model fills every gap you leave with a guess — and a confident guess looks exactly like an answer.
AI NATIVE ACADEMY1:301.1AI doesn't fail.
Your brief does.
The model fills every gap you leave with a guess — and a confident guess looks exactly like an answer.
Source values are the literals in const T. Spec off this table, not off the frames above.
The 18% reduction applies to only the two Cinzel roles — head 78→64 and mega 124→102. Everything else keeps its number and changes face only. Just elName moves families outright, because Cinzel at 56px cannot hold "Constraints & must-nots" on one line.
Warm Editorial, Night Studio, Editorial Focus and the three others are a tempo-and-surface-sequence system, not a palette system. They compose with porcelain rather than competing with it. Only the values inside pal() change.
Base #0a0a0a→#07101D, ink #F7F6F2→#F5F1E6, gold #D4AF37→#E6C883. Glass cards at .045 flatten to #0C1626.
Glass now lives on Marketing, Campaign and Editorial per 06b — but not here. A lesson frame is watched, not browsed: blur costs render time, softens type at 1920×1080, and competes with the timeline for attention. Video stays on the flat floor.
Both URLs are behind the dashboard login so I couldn't load them, but the routes are in the repo — src/pages/Systems.tsx and src/pages/Readiness.tsx, both registered in app.tsx and nav.ts. Reading them changes my answer. Mission Control isn't just a palette — it's a built instrument vocabulary, and it carries three primitives v2 doesn't have. Option D is: adopt that grammar wholesale for ops and founder surfaces.
LivenessDot encodes state in the glyph shape — filled, half, hollow — then colour, then a title attribute. Three redundant channels. My v2 contract used a coloured dot plus a word; this is strictly better and colour-blind-safe. Adopting it.
Every readiness claim carries kind + label pills that link to the issue, PR, or plan doc it came from. This is a governance primitive v2 has nothing equivalent to — and it's arguably worth more than the trust states, because it answers "who says so" rather than just "how fresh".
The speedometer needle and the reticle's brackets and centre dot are all --gold. Not premium cue, not attention hairline — a pointer that indicates a position. That's a fourth semantic role, and it doesn't conflict with "gold never fills" because a needle is a line.
A fifth class — and that's the useful answer. Its colours are already 01's ladder, so it isn't a new palette at all. What makes /systems and /readiness feel different from the learner product is density and instrumentation, not hue. I'd add Console to the permission matrix as a fifth column: gauges and dials allowed, provenance mandatory, Iowan dropped for Inter, density roughly double Product, and no marketing texture at all. Everything else in v2 stays as written.
The material split settled two of the three open questions. Social keeps warm editorial, so it needs no rebuild. Lesson video is Product, so it takes Iowan and the thirty re-renders shrink to title cards. What remains is the website H1 alone: does it run in Cinzel uppercase, or does Cinzel sign the page as an eyebrow while Iowan carries the headline? That gates wave 03 and nothing else — a single-surface call now, not a system-wide one.