Point 5: a deck's GENRE declares content it is not finished without

`design-by-purpose.md` carries nine purpose recipes — Type · Palette · Density ·
Layout · Icons · Signature, one per genre. Measured by grep, NOTHING consumed
that file: a build could ignore it line by line and no gate would know, so every
per-purpose rule in it was advisory by construction. And the four genres with the
most rigid conventions were absent from the list entirely:

  grant          aims · feasibility · risk
  committee      progress · plan · ASK
  journal_club   attribution · critique
  clinical_case  presentation · investigations · management · outcome

Three of those four are among the most common decks an academic makes, and one of
them is the deck this session built by hand with no entry to follow.

THE MECHANISM IS BORROWED, NOT INVENTED. `formats.py` + `check_surface.py`
already do exactly this for SURFACES — a registry declares the sections a surface
cannot be finished without — and that precedent exists because of a measured
finding: "the billboard style that reads best is also the style that drops the two
things a passer-by cannot reconstruct". The finding generalises. The shape of a
genre is precisely what an author under time pressure drops: a committee deck
loses the ask and silently becomes a status update.

scripts/purposes.py (registry) + scripts/check_purpose.py (checker), on BOTH gate
paths (parity: 21 sections, 0 problems), block-with-waiver like check_surface.

EFFECTIVENESS — five purpose-built decks, both directions:
  * complete committee deck               -> clean
  * same deck minus the ask               -> reports exactly `ask`
  * grant deck with aims only             -> reports exactly feasibility + risk
  * ask living only in SPEAKER NOTES      -> clean (notes count; excluding them
    would fire on every deck that moved its sentences where this skill says to)
  * 中文 committee deck                    -> clean via the 中文 terms

GENERALISATION:
  * multilingual term lists, with a test asserting EVERY section carries a
    non-Latin synonym — a monolingual list reports every CJK deck as missing
    every section
  * both record schemas (shared interview.picks, Codex design.purpose, and the
    audience brief when no purpose row was filled) — a floor read from one
    runtime stops applying to the other, which this repo has been through
  * a genre outside the registry binds to NOTHING and reports NOT CHECKED
  * a language the registry does not carry is cleared by EXTENDING the terms
    (design_plan.purpose_section_terms), which keeps the check alive, rather than
    by waiving it away — tested with a Dutch deck

NEGATIVE BINDINGS, PINNED. The first draft bound on a bare "committee" and a bare
"proposal". That would have pulled an INVESTMENT committee, an ETHICS committee, a
steering committee and a business proposal into academic section lists — firing
three or four times at the wrong deck, which is exactly how people learn to ignore
a check. Worse, `recorded_purpose` also reads the AUDIENCE BRIEF, so a loose term
reaches far more decks than a purpose row alone. Seven negative cases are now
permanent assertions so a future loosening fails the suite.

Honest about its limit, in the docstring and in design-by-purpose.md: it asks
whether the deck NAMES the section, never whether it does it well — judging that
stays the critic's — and the OTHER purposes in that file remain prose, reported as
NOT CHECKED rather than implied to be covered.

58 assertions wired into CI · gate-batching 40/0 · gate-ergonomics 26/0 ·
surface-contract 40/0 · step1-artifacts 57/0 · critic-waiver 101/0 · timidity 21/0
· codex delivery gate and parity ok · all structural guards clean · SKILL.md
lossless 2692/2692.

Author: dong845

.github/workflows/ci.yml

@@ -679,6 +679,17 @@ jobs:
           grep -qE "^[0-9]+ passed, 0 failed" /tmp/renderpar.log || {
             echo "::error::render-equivalence suite did not run to completion"; exit 1; }
 
+      # A deck's GENRE declares content it is not finished without — the formats.py/check_surface
+      # mechanism generalised from SURFACES to GENRES. design-by-purpose.md carried nine purpose
+      # recipes that NOTHING consumed, so every rule in it was advisory by construction.
+      - name: Purpose sections (a genre's required content, both record schemas)
+        run: |
+          set -o pipefail
+          python tests/test_purpose_sections.py | tee /tmp/purp.log
+          grep -qE "^[0-9]+ passed, 0 failed" /tmp/purp.log || {
+            echo "::error::purpose sections suite did not run to completion"; exit 1; }
+          python scripts/check_purpose.py --selftest
+
       # The clinical / evidence-synthesis forms. Each owns a rule that is wrong BY DEFAULT when
       # hand-rolled (a linear axis for ratios, a line instead of a step, a correlation instead of
       # limits of agreement) — and a wrong chart is a well-formed set of shapes, so no geometry

skills/slide-maker/SKILL.md

@@ -274,7 +274,7 @@ every **🔴 CHECKPOINT** is a hard stop.
 | Concern | Owner |
 |---|---|
 | The craft / the "why" (contrast · hierarchy · C.R.A.P. · layout safety) | `references/design-principles.md` |
-| Per-purpose look (defense vs exec vs lecture …) | `references/design-by-purpose.md` |
+| Per-purpose look (defense vs exec vs lecture …) | `references/design-by-purpose.md` — 🔴 four genres (grant · progress/guidance committee · journal club · clinical case) also DECLARE content they are not finished without, checked on both gate paths by `scripts/purposes.py` + `scripts/check_purpose.py` |
 | Per-TOPIC look (domain → apt presets → ANTI-PICK + cliché guard — the topic-adapted pick) | `references/design-by-topic.md` — 🔴 it adapts the LOOK only; a domain that needs its own FORM vocabulary is routed on the two rows below |
 | Bespoke registers invented from a subject's world (verified library to ADAPT + grow) | `references/bespoke-registers.md` |
 | Content — deep read + per-slide message (Step 1) | `agents/content-planner.md` |
@@ -516,6 +516,21 @@ source-material floor still gets asked);
 when in doubt, keep it. Never assume the **topic/content**, the **style**, or **which
 template** — confirm each.
 
+🔴 **Four purposes DECLARE required content, and the deck is held to it at hand-off.** A grant
+proposal (aims · feasibility · risk), a progress/guidance committee (progress · plan · **ask**), a
+journal club (attribution · critique) and a clinical case (presentation · investigations ·
+management · outcome) each carry sections their audience is asked to judge against — so a deck
+missing one is not a lean deck, it is a deck missing something the room needs. `scripts/purposes.py`
+declares them and `scripts/check_purpose.py` looks for them in the built file (both gate paths, CJK
+terms included, extend with `design_plan.purpose_section_terms`, waive in writing). It binds ONLY
+when the recorded purpose matches a registry entry — anything else is NOT CHECKED, never a guessed
+genre. **This is the `formats.py`/`check_surface.py` mechanism generalised from SURFACES to GENRES**,
+and for the same measured reason: the shape of a genre is exactly what an author under time pressure
+drops. Each also carries one fidelity rule beyond never-invent — a pilot stated as a pilot, open work
+that reads as open, their claim separable from yours, and on a clinical case **de-identification,
+which is the one irreversible item on this page**. The other purposes in `design-by-purpose.md`
+remain prose, and a deck in one of those genres reports NOT CHECKED rather than clean.
+
 **The `review:` question is NOT asked here — it moved to Step 5, after the first clean render.**
 It used to be a Step-0 axis, which forced the user to size a review of a deck nobody had seen;
 with the rendered deck in front of them the same choice is informed instead of blind, and that is

skills/slide-maker/references/design-by-purpose.md

@@ -313,6 +313,74 @@ tabs. Like a conference talk but built for a *shared-screen* medium, not a proje
   `references/animation.md`); explicit "ask in the chat" prompts; a visible agenda so late
   joiners orient. If it's recorded, ensure every slide reads as a still frame.
 
+## Grant proposal / funding pitch
+ERC · NWO · NIH · Horizon. The highest-stakes deck an academic makes, and the one with the most
+rigid conventions — reviewers score against *named* criteria, so a missing section is a lost point
+rather than a stylistic choice. Not a product pitch: the register is sober, and enthusiasm reads as
+a substitute for evidence.
+- **Type:** the academic default; equations only where the method genuinely turns on one.
+- **Palette:** institutional or sober; nothing that looks like a startup deck.
+- **Density:** high but structured — a reviewer is reading against a checklist, not being toured.
+- **Layout/chrome:** the scored sections visible as sections. Numbered aims.
+- **Signature:** the aims page is the deck. One sentence per aim, each falsifiable.
+- 🔴 **Declares (checked): aims · feasibility · risk.** Feasibility means *why this team, why now* —
+  pilot data, track record, access. Risk means named risks with named mitigations; "no significant
+  risks" is read as not having thought about it.
+- 🔴 **Fidelity:** preliminary data is the section most often overstated. A pilot is a pilot — say n,
+  say it is preliminary. A reviewer who catches one inflated claim discounts the rest.
+
+## Progress / guidance committee meeting
+A PhD guidance committee, a thesis advisory board, a stage-gate review. It differs from a lab
+meeting in exactly one way, and it is the way that matters: **the room has to decide something** —
+is this on track, does the thesis hold together, what should be scoped down. A committee deck with
+no ask has silently become a status update, and the meeting's most valuable output is lost.
+- **Type/palette:** the lab or institution's own template, unchanged. Nobody is being impressed.
+- **Density:** moderate-to-high; the deck is also the record of the year, read as well as presented.
+- **Layout/chrome:** status legible at a glance — what is done, what is out for review, what is
+  still on your desk. Bind one hue to each state and keep it deck-wide.
+- **Signature:** the closing slide names the decisions you want them to weigh in on.
+- 🔴 **Declares (checked): progress · plan · ask.**
+- 🔴 **Fidelity:** what is not done must READ as not done. A committee exists to advise on open
+  work, and a hypothesis promoted to a result removes the thing they are there to help with
+  (`content.open_ledger` is the artifact).
+
+## Journal club / presenting someone else's paper
+You are presenting work you did not do. Extremely common, and it carries a fidelity risk the
+universal never-invent rule does not reach: not fabrication but **misattribution** — stating their
+result more strongly than they did, or blurring their claims with your critique until the audience
+cannot tell which is which.
+- **Type/palette:** reuse the paper's own figures; let its visual language show through.
+- **Density:** moderate — the method matters, but the room has (in principle) read it.
+- **Layout/chrome:** a visible separation between *their claim* and *your reading*. A rail, a
+  colour, a consistent slot — anything, as long as it is the same on every slide.
+- **Signature:** your assessment, owned as yours.
+- 🔴 **Declares (checked): attribution · critique.**
+- 🔴 **Fidelity:** attribute on the slide that carries the result, not only on the title slide. An
+  audience that cannot tell which is which will remember your critique as the paper's conclusion.
+
+## Clinical case presentation / tumour board / M&M
+A fixed clinical narrative. The structure is not a style choice: colleagues are being asked to judge
+a decision, and they cannot do that without the information that was available when it was made.
+- **Type/palette:** the institution's template; sober. Never playful — somber veto.
+- **Density:** high on the facts, minimal on chrome. Imaging large.
+- **Layout/chrome:** one section per stage, in order; imaging shown at a size a clinician can read.
+- **Signature:** the learning point, stated plainly.
+- 🔴 **Declares (checked): presentation · investigations · management · outcome.**
+- 🔴 **Fidelity — DE-IDENTIFICATION IS LOAD-BEARING.** This is the genre PRE-FLIGHT 5's burned-in
+  identifier check exists for: name, MRN, accession, date of birth, study date, institution, read on
+  all four edges and in every overlay strip. Get a de-identified export — never crop or blur and
+  ship. Unlike everything else on this page, it is irreversible once the deck is sent.
+
+> 🔴 **These four are CHECKED, not just described.** `scripts/purposes.py` declares the sections each
+> genre is not finished without and `scripts/check_purpose.py` looks for them in the built deck, on
+> both gate paths — the same mechanism `formats.py`/`check_surface.py` already applies to surfaces.
+> The check asks whether the deck NAMES the section, not whether it does it well; judging the latter
+> is the critic's job. Terms are multilingual; extend them with `design_plan.purpose_section_terms`
+> rather than waiving, and waive in writing when a deck genuinely has none.
+>
+> **The other purposes on this page are still prose.** That is a real gap and not a claim of
+> coverage: a genre with no registry entry binds to nothing and reports NOT CHECKED.
+
 ## Preset fit by purpose — a fast sanity map (fit is a default, avoid is a flag, both overridable with a stated reason)
 | Purpose | Natural fits | Flag before using |
 |---|---|---|

skills/slide-maker/references/file-inventory.md

@@ -179,6 +179,24 @@ waiver once carried a whole deck through `all hand-off gates pass` with no indep
   the colour") stay prose and are REPORTED as unchecked. The register is resolved by
   `check_style_applied.declared_preset`, never a substring search: a bespoke deck whose pick read
   "beat blueprint-the-preset" was checked as `blueprint` on the first try.
+- `purposes.py` + `check_purpose.py` — a deck's GENRE declares content it is not finished without,
+  checked against the built file on both gate paths. `references/design-by-purpose.md` carried nine
+  purpose recipes and, measured by grep, NOTHING consumed it — so every per-purpose rule there was
+  advisory by construction — and the four genres with the most rigid conventions were missing from
+  the list entirely: grant proposal (aims · feasibility · risk), progress/guidance committee
+  (progress · plan · ask), journal club (attribution · critique), clinical case
+  (presentation · investigations · management · outcome). Three of those four are among the most
+  common decks an academic makes. 🔴 This is the `formats.py` / `check_surface.py` mechanism
+  generalised from SURFACES to GENRES, for the same measured reason — the shape of a genre is
+  exactly what an author under time pressure drops (a poster loses methods and limitations; a
+  committee deck loses the ask and becomes a status update). Binds by matching the RECORDED purpose
+  (read from both record schemas, including the audience brief) against `binds_on` terms; no match
+  is NOT CHECKED, never a guessed genre, because firing a clinical section list at a product pitch
+  teaches the author to ignore the check. Section terms are multilingual (a 中文 deck names the same
+  section in 中文 and a monolingual list would fire on every one), extendable with
+  `design_plan.purpose_section_terms`, waivable with `{"purpose": {"waived": …}}`. It asks whether
+  the deck NAMES the section, not whether it does it well — judging that stays the critic's. Each
+  purpose also carries one fidelity rule beyond never-invent, printed every run.
 - `check_template_profile.py` — a registered template's `profile.md` must be OBEYED, not merely
   available. The registry is the one artefact in this skill with a MEMORY across decks — layout
   indices, decorative furniture that cannot be deleted from a layout and has to be covered by the