---
name: "convention-deck"
type: convention
license: CC BY-NC-SA 4.0
description: The house slide-deck design system — the slide archetype library and its weight tiers, the treatment library harvested from the TMS files (named treatments inside the archetypes, with build patterns), the fixed grid and type scale, the rhythm rules that carry a deck across a run of slides, the per-format mechanics for pptx and HTML output, and the deck-specific anti-patterns. Read before designing, producing, or reviewing any deck, so quality comes from a small set of opinionated constraints rather than the generic defaults a format skill falls back on. Owned by Pixel; inherits convention-aesthetic (Chronicle is the layout reference) and the venture's convention-brand instance. Read by skill-deck-narrative for archetype tagging, skill-visual-production for the build, and skill-deck-parity for the review. Surfaced by APP-556; treatments added by APP-662; grown as decks are produced.
---

# Deck design system

The house system for slides. `convention-aesthetic` holds the cross-venture sensibility and `convention-brand` holds each venture's livery; neither covers decks, and that gap is where deck quality has been leaking. This convention closes it: the archetypes, the grid, the type scale, and the rhythm rules a deck is built and judged against. Owned by Pixel; read by `skill-deck-narrative` (which tags the outline with archetypes), `skill-visual-production` (which builds), and `skill-deck-parity` (which scores).

## Why this exists — the constraint argument

Decks produced through the pptx skill come out generic: flat visual hierarchy, every slide carrying similar weight and layout, no rhythm across the run. The root cause is not the skill's competence but its remit — the pptx skill is a *format and mechanics* skill carrying generic design heuristics, and there was no deck-specific canon above it to constrain what it produced. Anything unconstrained regresses to the template default, which is title-and-bullets (APP-556).

The target reference is **Chronicle** (chroniclehq.com), recorded in `convention-aesthetic` as a layout and document-craft reference — explicitly not a brand-identity one. What is worth taking from it is *where its quality comes from*: a small set of opinionated blocks, a fixed type system, and narrative-first authoring. The quality is a product of constraint, not of effort per slide. So this convention is written as a short, closed set of choices rather than a menu of possibilities — a deck that can be assembled from ten archetypes on one grid at one type scale is a deck that cannot drift into flatness.

## Inheritance

Three layers stack, and each states only what the layer below does not:

* **`convention-aesthetic`** — the cross-venture sensibility (quiet over loud, editorial over commercial) and the Chronicle layout entry. Inherited whole.
* **`convention-brand`, the venture's instance** — palette, type faces, logo, imagery. Supplies the *values* this convention's slots take.
* **This convention** — the deck-specific system: archetypes, grid, scale, rhythm. Venture-agnostic structure.

Where a venture's brand instance and this convention disagree on a *value* (a face, an accent), the brand instance wins — it is the layer that owns livery. Where they disagree on *structure* (a layout, a rhythm rule), this convention wins. A brand instance does not get to reintroduce title-and-bullets.

## Output formats — the system is format-agnostic, the mechanics are not

A deck is not a pptx file; it is an argument with a shape. **The archetypes, the grid, the type scale, the colour rule and the rhythm rules are the system, and they hold whoever renders them** — they ported to HTML/CSS on first contact without amendment (a1.Muse summary deck, 2026-07-16, APP-571). What is format-specific is the *material*: how a typeface is resolved, what a template is, and where the review happens. Those are named per format below; everything else in this convention is read as canon regardless of output.

| | **pptx** | **HTML** |
| -- | -- | -- |
| **Type resolution** | The font-substitution rule below — the renderer silently substitutes a missing face. | A web-font stack stated explicitly in the deck's CSS, with deliberate fallbacks. No silent substitution, but an unloaded webfont degrades to the fallback the same way, so the stack is chosen, not defaulted. |
| **Type scale** | The five steps, in points, at 13.33in × 7.5in. | The same five steps in container-query units (`cqw`), so the scale holds at any viewport. **The ratios are the canon; the point sizes are one format's expression of them.** |
| **Template** | A `.potx` whose masters carry the archetypes (see *Templates*). | A stylesheet carrying the archetypes as classes. Same argument, same current gap. |
| **Review** | `skill-deck-parity` renders the slides to images. | `skill-deck-parity` screenshots the rendered deck — same review, same rhythm check across the sequence. |

**Choose the format from how the deck will be used, not from what is easiest to generate.** pptx where the client will open, edit, or present it themselves. HTML where it is reviewed in-browser, linked, or embedded — which is what A1 client work has wanted so far, avoiding a PowerPoint round-trip nobody needed. One instance is not yet a default; if HTML keeps winning for client review, that is the sentence to sharpen next.

## The archetype library

Ten archetypes, three weight tiers. Weight is what the slide asks of the audience, not how much ink is on it — a full-bleed statement is ink-light and attention-heavy, and that is the point.

**Tier 1 — Statement.** One idea, set large, carried by negative space. The job is to land something or reset the room.

| Archetype | What it is | When |
| -- | -- | -- |
| **Full-bleed statement** | One sentence at display size on a full-bleed ground, heavy negative space. Doubles as the cover. | The deck's opening claim; any moment the argument turns. |
| **Big-number** | A single figure at display size, with a caption naming what it is and why it matters. | One number is the argument. |
| **Quote** | A verbatim quote, attributed, at headline size. | Borrowed authority, or a user's own words. |
| **Section break** | Section title plus its one-line job. Near-empty by design. | Between sections; the breather. |
| **Closer** | The ask, the decision needed, or the hero's new normal — one line. | The last slide, always. |

**Tier 2 — Development.** One idea plus the structure that makes it legible. The job is to advance the argument.

| Archetype | What it is | When |
| -- | -- | -- |
| **Two-column** | Claim on one side, its support on the other, on the asymmetric grid. | A point that needs one piece of scaffolding. |
| **Comparison** | Two to four items against shared criteria, ideally a table. | Options, trade-offs, before-and-after. |
| **Timeline** | A sequence along one axis, with the stated beats. | A plan, a roadmap, a history. |
| **Chart figure** | One chart as a flat, single-accent figure in a soft card, titled with its finding. | Data that carries a point. |

**Tier 3 — Evidence.** The proof, at density. The job is to substantiate.

| Archetype | What it is | When |
| -- | -- | -- |
| **Dense evidence** | Multiple findings, a full table, or a detail grid — the one place density is licensed. | The audience needs to see the working. |

Two rules on the library itself. **The set is closed** — a slide that does not fit an archetype is a signal the idea is not resolved, not a licence to invent an eleventh layout; take it back to the outline. And **an archetype is a layout, not a content type** — a chart can be a big-number if one figure is the point, and a finding can be a full-bleed statement if it is the argument's hinge. Pick by the job the slide does.

## Treatment library — the TMS harvest

A **treatment** is a named layout *inside* an archetype. The archetype owns weight, rhythm, scale and colour; the treatment names a structure that has already proved itself and the pattern that builds it, so production picks a treatment rather than inventing a layout. The set was harvested on 2026-07-23 from Aled's three TMS Figma files — 16 Strategy layouts, 21 Planning layouts and the 64-page Core 2.0 base system — and every entry maps to one of the ten archetypes above. **The archetype set stays closed at ten.** Where a treatment and this convention disagree on structure, this convention wins; where a harvested layout turned out to *be* an archetype (the cover, the section header, the thank-you, the full-page quote), it is recorded as a confirmation of the Tier 1 set, not added.

**Geometry the harvest shares.** Canvas 1920 × 1080 with a 1832px content field; A1 house HTML keeps its own 5% margins and borrows the structures, not the margins. Header block top-left — eyebrow, then title at the headline step, then a muted subtitle — often paired with an **intro paragraph right-aligned in the top-right quadrant**, which is the house alternative to a bulleted agenda. Split geometry aside + figure = 588 + 1200 with a 44px gutter; thirds at 3 × 594.7 with 24px gutters; quarters and fifths divide the same field. Body region from y≈188, ~1000px tall; a ~78px footer strip where present. Category **chips** (pill labels) mark taxonomy on content-heavy pages. One idea per slide holds throughout — even the dense pages are a single matrix, list or grid.

**Strategy treatments (16).**

| Treatment | Structure | Archetype · tier | The job |
| -- | -- | -- | -- |
| Propositions | 5 equal vertical cards with bullet lists, one highlighted | Comparison · T2 | Option sets, proposition shortlists |
| Strategy map | 2×2 quadrant (Consumer / Culture / Category / Company), centre pill = the proposition | Chart figure · T2 | Positioning synthesis — "where we sit" |
| Outlook | Stepped horizon boxes rising left→right (Today → 2030 → 2040 → 2050), intro top-right | Timeline · T2 | Long-horizon ambition staircase |
| Objectives | 3 year-columns, heading + bullets each | Comparison · T2 | Objectives by year or phase |
| Funnel | Matrix: 4 stages across (Awareness → Loyalty), content rows down, hairline rules | Dense evidence · T3 | Funnel content and measurement mapping |
| Framework | Full-width stacked table: Mission / Positioning / 4-col typology / Objectives / Topics | Dense evidence · T3 | One-page strategy framework |
| Allocation | 588 aside (legend + narrative) + 1200 line chart, phase bands, dashed and solid series | Chart figure · T2 | Budget or resource allocation over time |
| Golden circle | 3 concentric rings left (Why / How / What), text ladder right | Chart figure · T2 | Why / how / what narrative |
| Attributes | 8 semantic-differential sliders, aside left | Comparison · T2 | Brand attribute positioning |
| Onion | 4 nested circles, text right | Chart figure · T2 | Layered model, core → outer |
| Collaborators | Orbit rings around "You" (Team / Community / Collaborators / Researchers), aside left | Chart figure · T2 | Ecosystem or proximity map |
| Positioning | 3-circle Venn, corners labelled, centre = proposition | Chart figure · T2 | Sweet-spot positioning |
| Methodology | Circular loop of 4 nodes on a dotted orbit (Learn → Envision → Plan → Create) | Chart figure · T2 | Operating loop or method |
| Strategy pyramid | 6-tier pyramid, text right | Chart figure · T2 | Hierarchy of intent |
| Projection | Single growth curve, milestone ticks, month axis, legend top-right | Chart figure · T2 | Growth projection with beats |
| Targets | Stair-step card grid ascending to top-right on a date axis | Timeline · T2 | Targets laddering up over time |

**Planning treatments (21).**

| Treatment | Structure | Archetype · tier | The job |
| -- | -- | -- | -- |
| Roadmap 1 | One horizontal track, week and Today nodes, bullets under the active node, intro top-right | Timeline · T2 | Light roadmap, few beats |
| Roadmap 2 | Kick-off + Phase 1–4 stations on a line, month + bullets per station | Timeline · T2 | Phase roadmap with detail |
| Roadmap 3 | Ascending gantt-style bar steps on a date axis | Timeline · T2 | Workstream ramp, stacked start dates |
| Phases 1 | 4 outlined phase cards in a row, intro top-right | Comparison · T2 | Phase overview |
| Phases 2 | 4 dotted semicircle arcs numbered 01–04, description under each | Timeline · T2 | Phases with a softer rhythm |
| Phases 3 | Pill segments on a month scale, Today marker | Timeline · T2 | Engagement-stage bar |
| Sprints | 588 aside + 1200 dark panel: sprint groups, phase pills per week row, iteration brackets | Dense evidence · T3 | Sprint or iteration plan |
| Milestones 1 / 2 (+ highlight states) | Central timeline, media cards above and below at ticks; highlight dims all but the current | Timeline · T2 | Milestone walk-through — the highlight is a presentation device: duplicate the slide per milestone |
| Timeline 1 | Month-column grid (~3 months), row labels, descriptions footer | Dense evidence · T3 | Detailed near-term schedule |
| Timeline 2 | 12-month fine grid, many rows | Dense evidence · T3 | Annual programme view |
| Output | 6 week-columns, bullet lists in two banded rows | Dense evidence · T3 | Weekly output or deliverables |
| Content plan | Matrix: KPI rows × 6 channel columns, bullets per cell | Dense evidence · T3 | Channel or content planning |
| Q1–Q4 | 4 quarter columns × 3 white month cards, intro top-right | Timeline · T2 | Year at a glance |
| Organisation | Org chart: client departments flanking a centre core-team column with role pills | Chart figure · T2 | Engagement org, who does what |
| Team 1 | 3 leadership rows (photo + name/role) + 4-col thumb grid right | Media grid · T2 | Team with hierarchy |
| Team 2 / Team 3 | 4-up or 5-up portrait cards, name + title (+ bio on 4-up) | Media grid · T2 | Team gallery |
| Propositions | As the Strategy treatment | Comparison · T2 | — |

**Core 2.0 — the base system, as families.** *Structure pages* — cover; numbered section index; section header; TOC with a content card, or as six numbered groups in two columns; header-only text slide; full-bleed media; thank-you with contact block. These are the Tier 1 set — statement, section break, closer — confirmed, not extended. *Media grids* — the deep bench: 1-up hero (plain, offset, or hero + side text); 2-up; 3-up (including asymmetric); 4-up including mixed-height; 5-up; 2×3 and 2×4 grids; 8–12 thumbnail walls; **bento / masonry** mixed-size compositions; a logo wall with per-cell descriptions. Caption pattern is `Name / Subtitle` for people or `Title + description` for work. These are the media expression of two-column, comparison or dense evidence depending on count; a 1-up hero can carry a Tier 1 moment. *Text systems* — definition ladders (label + paragraph rows at 1- to 5-column density); statement + supporting references; category matrices with chips. Two-column at low density, dense evidence at high. *Quotes* — full-page quote, or quote + portrait card, attribution bottom-left: the Quote archetype, T1.

**Media grid** is used above as a treatment family, not an eleventh archetype: an n-up grid resolves to two-column (1–2 up), comparison (3–5 up) or dense evidence (6 and above) for weight and rhythm purposes.

**Build patterns (HTML).** The a1 deck stylesheet already carries kicker and furniture, `.rows`, `.fun`, `.tbl`, `.card`, `.dash` lists and the dark and paper grounds. Treatments are added in the same idiom — flat, hairline, one accent:

- `.media-grid` — CSS grid, n-up; each media slot an `aspect-ratio` box with `object-fit: cover`; caption block beneath, title at 600 weight plus muted description.
- `.bento` — `grid-template-areas` per composition; gutters stay 24px.
- `.ladder` — label (600) + paragraph rows, hairline between rows; 1–5 columns via `grid-template-columns`.
- `.matrix` — a table with an uppercase caption header row, hairline rules, chips for taxonomy (Funnel, Framework, Content plan, Output, Timelines).
- `.track` — the roadmap line: `border-top` plus absolutely positioned nodes; stations as flex columns (Roadmaps, Phases 3, the Projection axis).
- `.stairs` — ascending steps: a grid with `nth-child` row offsets (Targets, Roadmap 3, Outlook).
- `.quarters` — 4 columns of stacked white month cards.
- Diagrams (golden circle, onion, orbit, venn, pyramid, loop, quadrant) — inline SVG, stroke in the hairline tone, fills from the neutral ramp, one accent for the active element; **labels as an HTML overlay, not SVG text**, so type stays on the scale.
- `.org` / `.team` — org chart as a 3-column grid with pill nodes; team as a media-grid variant.
- Charts (Allocation, Projection) — flat single-accent lines per *Colour*; no renderer palettes.

## Section libraries as a layout source

A **section library** — a catalogue of finished, production-grade web sections — is a sanctioned third source of layout, beside the TMS harvest above and Chronicle in `convention-aesthetic`. It earns its place for the same reason the harvest does: a layout that has already proved itself beats one re-derived per deck, and the quality comes from the constraint, not from the effort per slide.

**The named library is BYQ Supply, where its connector is attached to the surface in hand, and within it the `Stringer` collection.** Stringer transfers to decks better than most because its visual language is already the house one: a warm off-white ground (`#e2ddd4` on the tile variant), near-black ink (`#121212`) with a muted companion, hairline rules at the lightest tint, a label-above-heading header block, display headings at the top of the scale, and exactly one accent used for meaning. (Verified first-hand 2026-10-05: `search_byq_components` returns the collection by the `stringer-*` id prefix; `get_byq_component` returns a rendered screenshot, a React/TypeScript + Tailwind reference implementation, and a *Section details* breakdown of layout, type, colour, interactions and responsive behaviour. The sections set **General Sans**. Every Stringer item is marked Pro, which did not block retrieval on that surface — where a retrieval fails for access, that is an Owner account matter, not a deck decision.)

**What carries over is layout structure and styling language — never code.** The sections are React and Tailwind; a Slides artifact is a closed inline-CSS subset, and a pptx is further still. Nothing is pasted. Read a section exactly the way a TMS treatment is read: take the geometry, the hierarchy and the rhythm, recolour to the venture's `convention-brand` instance per *Colour*, and resolve the face deliberately per *Type scale* — on the first deck built this way, DM Sans stood in for General Sans and the substitution was recorded rather than left to the renderer.

**The archetype set stays closed at ten, and a section maps to one before it is used** — the same rule the treatment library takes, and the same reason: a library that sets its own layouts quietly becomes an eleventh archetype. Catalogue instances verified 2026-10-05:

| Section shape | Catalogue instance | Archetype · tier |
| -- | -- | -- |
| Stat tile / count-and-meter | `stringer-ecommerce-10` | Big-number · T1 where one figure is the argument |
| Labelled spec grid | `stringer-ecommerce-16` | Dense evidence · T3 where it is a grid of figures |
| Flat tile grid, three up | `stringer-cta-3` | Comparison · T2 — themes, fix groups, "three ways in" |
| Comparison spec table | `stringer-ecommerce-6` | Comparison · T2 — prioritisation, options against shared criteria |
| Two-column, aside plus figure | `stringer-ecommerce-17` | Two-column · T2 — the plan |
| Dark statement section | `stringer-cta-1` | Full-bleed statement · T1 — the statement moment |
| Bento / masonry | `stringer-bento-4` | Media grid family — resolves by count, per *Treatment library* |

**A library is a layout source, not a rhythm source.** The limit is the one the treatment library already carries and it binds harder here, because a section catalogue is almost entirely development and evidence weight: Tier 1 opens, closes and breathers still come from the core set — statement, quote, section break, closer — and rules 2, 3 and 5 are checked across the run as ever. Never let the library set the rhythm.

(APP-1989, a1.Muse findings-and-plan deck, 2026-10-02 — the Owner's judgement on the deck built from these layouts was that it was "much better than what you've created before from scratch", and he asked for it to be codified here. Layout-reference precedent: the Chronicle entry in `convention-aesthetic`, APP-561. One instance so far — a second library is added here only when one is actually used, never speculatively.)

## Grid

* **Canvas** — 16:9, 13.33in × 7.5in. This is the house default; 4:3 only where a client's own projection demands it.
* **Twelve columns**, with a margin of one column's width on each side. The safe area is columns 2–11 for anything that is not full-bleed.
* **Asymmetric thirds is the default placement.** Content sits in columns 2–7 or 6–11, not centred. Chronicle's grid is asymmetric, and it is what stops a deck reading as a template — centred-everything is the default that generic output falls into.
* **Hairline structure, not decoration.** Where a rule is needed to separate or align, it is 0.5–0.75pt in a muted tone from the brand's neutral ramp. Rules divide; they never ornament. A bar that exists to add colour is an anti-pattern, named below.
* **Negative space is a component.** A Tier 1 slide is majority empty. This is the constraint most likely to be quietly eroded during production, so it is stated as a rule rather than an aspiration.

## Type scale

One scale, five steps, fixed. The whole deck is set from it; a size that is not on the scale is not available.

| Step | Size (16:9) | Use |
| -- | -- | -- |
| **Display** | 72pt | Full-bleed statement, big-number, closer. |
| **Headline** | 40pt | Slide titles — which are action titles (`skill-deck-narrative`), never labels. |
| **Subhead** | 28pt | Section breaks, quote attribution, chart finding lines. |
| **Body** | 18pt | The floor. Nothing on a slide goes below body size. |
| **Caption / eyebrow** | 12pt | Small-caps, tracked, in the single accent. Sources, units, the eyebrow. |

Sizes are the pptx expression of the scale, at the 16:9 canvas above. In HTML the same five steps are set in `cqw` so they hold at any viewport — the **ratios and the five-step closure are what this table fixes**, not the points (*Output formats*).

**One family, two weights.** The house discipline is a single type family across the deck, in two weights — a display weight and a regular. Pairing two families is available only where the venture's brand instance specifies a pairing; it is never a per-deck choice.

**The font-substitution rule (pptx), and why it is here.** The pptx renderer does not embed an arbitrary face — a font absent from the rendering host is substituted *silently*, and the deck that arrives on the client's machine is not the deck that was reviewed. So: where the venture's brand face is available to the renderer, use it. Where it is not, choose the substitute **deliberately** from the cross-platform core set (Arial, Georgia, Verdana, Trebuchet MS, Times New Roman, Courier New), record the substitution on the hand-off, and never let the renderer pick. A recorded substitution is a known compromise; a silent one is a defect that reaches the audience and nobody else. This is also why a `.potx` template beats from-scratch generation — the masters carry the resolved face.

## Colour

* **Near-monochrome ground, one accent per deck.** The ground comes from the venture's neutral ramp; exactly one accent from the brand palette carries emphasis across the whole deck. Not one accent per section, and not a palette per chart.
* **The accent is for meaning, not interest.** It marks the eyebrow, the one figure that matters, the single series in a chart. A second accent is a request to be re-read as a second idea, which is a rhythm problem, not a colour one.
* Charts are **flat, single-accent figures in soft cards** — no gradients, no drop shadows, no default renderer palettes.
* **Harvested treatments are livery-agnostic — recolour, never ship the source.** The TMS files are drawn dark, structure-only, with grey gradient placeholder cards: a gradient card is a **media slot** (image or screenshot), a flat panel is a **content card**. Ground, ink and the single accent always come from the venture's `convention-brand`instance (A1: paper`#ddd9ce`, ink `#070100`, accent `#ff2a00`); the TMS grey is never shipped.

## Rhythm rules

Rhythm is the property a deck has and a slide cannot. It is checked across the run, before rendering, and it is the thing generic production has no concept of.

1. **One idea per slide.** If the title needs an "and", it is two slides.
2. **Open heavy, close heavy.** The first and last slides are Tier 1, always.
3. **A breather every 4–6 slides.** A Tier 1 slide — usually a section break — at least that often. A run longer than six without one is where an audience stops tracking the argument.
4. **Never two consecutive slides on the same archetype.** Two comparisons back to back read as one slide the audience has already seen.
5. **Never a run of three or more slides at the same tier.** This is the equal-weight run, stated positively: weight must vary.
6. **Never more than two consecutive Tier 3 slides.** Density is licensed, but it is not a mode the deck can live in.

**The treatment library is almost entirely Tier 2 and Tier 3.** A deck assembled only from it breaks rules 2, 3 and 5 by construction: Tier 1 open and close, and the breathers, still come from the core set — statement, quote, section break, closer. Pick treatments for the development and evidence slides; never let the library set the rhythm.

Rhythm is set at the **narrative stage**, not discovered at the render stage — `skill-deck-narrative` tags every outline line with an archetype and tier precisely so these rules can be checked against an outline, when fixing them is free. `skill-deck-parity` checks them again against the rendered sequence, because that is where they actually break.

## Anti-patterns

Inherits every anti-pattern in `convention-aesthetic` (couture gloss, the generic AI look, cultural cliché, trend-pastiche). Deck-specific additions, each observed in the generic output APP-556 was raised against:

* **The equal-weight run.** Ten slides that all ask the same thing of the audience. The deck has no shape, so nothing in it is emphasised, so nothing is remembered. The single most common defect and the reason rhythm rules exist.
* **Title-and-bullets by default.** The layout an unconstrained renderer reaches for. A bulleted list is a slide whose idea has not been resolved into a claim — take it back to the outline rather than styling it.
* **Decorative accent bars.** A coloured bar, block, or swash placed to make a slide "look designed". It adds ink and no meaning, and it is the tell of generated-not-designed. Hairlines divide; bars decorate; only the first is allowed.
* **The label title.** "Q3 Update", "Our Approach", "Findings". A title that names a topic instead of stating a claim (`skill-deck-narrative`; the same failure `convention-deck-narrative` names under Inverted Pyramid).
* **Chart-as-dump.** A chart pasted at renderer defaults, titled with its topic rather than its finding, in a palette the deck does not otherwise use.

## Naming

A deck is named for its **purpose**, not its sequence: `{venture}-{purpose}-deck-vNN.{ext}`, in a per-purpose folder under the engagement's `40-deliverables/`. The purpose vocabulary is short and closed — `diagnostic` (the findings/plan deck), `checkpoint` (a dated progress review), `proposal` (a scoping or commercial deck) — extended only when a genuinely new deck job appears, the same closed-set discipline the archetype library uses. `vNN` is the iteration, `-v01` upward; a new purpose starts its own version line. This replaces sequence-only names (`…-summary-deck-0N`), which say nothing about the deck's job and let two different jobs share one name.

The failure this prevents is a real one: a1.Muse's standing diagnosis deck and a one-off progress review both lived under `summary-deck/…-06`, so "summary" quietly blurred two different jobs into one version line. Named by purpose, a recurring format versions independently without colliding with the deck beside it. (Applied 2026-07-21: `summary-deck/a1-muse-summary-deck-06` → `diagnostic-deck/a1-muse-diagnostic-deck-v06`, and a new progress deck filed as `checkpoint-deck/a1-muse-checkpoint-deck-v01`.)

## Templates

**Template-fill beats from-scratch generation**, and the reason is the substitution rule above plus the constraint argument: a `.potx` whose masters carry the archetypes, the grid, the scale, and the resolved face makes the house system the path of least resistance, where from-scratch generation makes it a thing production has to hold in mind slide by slide. Constraint has to be built into the material, not remembered.

**Current state: no venture has a `.potx` yet, and none has a deck stylesheet.** Until one exists, production builds against this convention directly and `skill-visual-production` raises a `human`/Pixel ticket to build the template for that venture (APP-558). When a template lands, record its pointer here, per venture, alongside the masters it carries. This is the same shape as the Figma reference-board gap in `convention-aesthetic` — the pointer lives in canon, the artefact is a human/Pixel action.

**The argument holds for HTML, and so does the gap.** A deck stylesheet carrying the archetypes as classes is the HTML equivalent of the masters, and it earns its place for the same reason: constraint built into the material rather than remembered per slide. The a1.Muse deck's CSS was written per-deck and is the first candidate to harvest one from (APP-571). The a1.Muse checkpoint deck (2026-07-21) then reused that same CSS directly rather than rebuilding — a second data point, so the case for pulling the a1 deck stylesheet out as a shared asset is now stronger than one instance. Record its pointer here when it exists.

**Treatment sources, for re-render.** TMS Strategy — Figma `iVjh1zIwhvUK5kndA5S623`, dark strip node `80:7707` (16 slides). TMS Planning — `FklBVzGzoTf5NSGNlLQe4k`, strip `87:32627` (21 slides). TMS Core 2.0 — `OdtdeqbdreKChVtT89hVZj`, strip `82:18704` (64 pages). Light-mode duplicates exist in each file; per-slide nodes are in the file page metadata. The full harvest with screenshots is on APP-662. Deck Studio (the A1 presentation tool) absorbs these as templates in its Phase 2; the treatment library above is the spec for that work.

## Quick checklist before a deck ships

- [ ] Every slide maps to one of the ten archetypes — nothing invented?
- [ ] Where a slide uses a harvested treatment, is it named from the treatment library, recoloured to the venture's livery, and not carrying the deck's rhythm on its own?
- [ ] Every slide carries one idea, titled as a claim rather than a topic?
- [ ] Opens and closes on Tier 1?
- [ ] A Tier 1 breather at least every 4–6 slides?
- [ ] No two consecutive slides on the same archetype; no run of three at one tier; no more than two consecutive Tier 3?
- [ ] All type on the five-step scale, nothing below body size, one family in two weights?
- [ ] Output format chosen from how the deck will be used, and its mechanics read from *Output formats*?
- [ ] Type resolved deliberately for that format — pptx: brand face available or the substitution chosen and recorded; HTML: the web-font stack stated in the CSS?
- [ ] Near-monochrome ground, exactly one accent, used for meaning?
- [ ] Hairlines only — no decorative bars, no bulleted defaults?
- [ ] Named for purpose, not sequence — `{venture}-{purpose}-deck-vNN`, purpose from the closed set?
- [ ] Built from the venture's `.potx` or deck stylesheet where one exists, or a template ticket raised where one does not?
- [ ] Tone clean per `cos.tov`?