@officexapp/vidfarm-devcli 0.21.58 → 0.21.59

package/experimental/sticker-slideshow-tips.md

@@ -20,14 +20,22 @@ checks:
   # upper bound: 9 slides x (1 bg + 3 cutouts).
   scenes: 5-40
   max_scene_sec: 3.2        # 3.0 is the number. The 0.2 is render tolerance, not licence
-  # A tip slide carries TWO text runs (headline + subtext) and there are up to 9 slides, so the
-  # whole-composition count is ~18. This is a READ format, not a subtitle format — the
-  # 1-3 ceilings in the sibling harnesses are for videos where a voice is doing the work.
-  max_text_cards: 20
-  max_simultaneous_text: 2  # headline + subtext. A third text run on a slide is clutter
+  # A tip slide carries THREE text runs (topic rail + headline + subtext) and there are up to
+  # 9 slides, so the whole-composition count is ~26. This is a READ format, not a subtitle
+  # format — the 1-3 ceilings in the sibling harnesses are for videos where a voice is doing
+  # the work.
+  max_text_cards: 28
+  max_simultaneous_text: 3  # topic rail + headline + subtext. A FOURTH text run is clutter.
+                            # The rail is the same string on every slide, so it costs the
+                            # reader one read, not N — see "The topic rail"
   max_words_per_cue: 16     # the headline is ONE read at arm's length, 3 lines maximum
   max_dead_air_sec: 1.0     # there is never a frame without text. A gap means a broken slide
   max_tail_sec: 0.5
+  # THE DECK HAS A CTA — it is the close on slide N, and it is an ADDRESS
+  # (`dishcover.io — every dish in NYC, searchable`). None of the phrases below
+  # are that. Each is an INSTRUCTION the reader did not need, because the address
+  # is already on the frame, and each is phrase-matched for suppression on a
+  # carousel. Keeping them banned is what makes the CTA shippable. See "The close".
   forbid_text:
     - link in bio
     - sign up
@@ -71,7 +79,7 @@ Three questions. **All three must be yes** or you have not got a tip slideshow y
 | # | Question | If no |
 |---|---|---|
 | 1 | **Actionable** — could a stranger do this tonight, without the product, and be better off? | You wrote a fact, or a platitude. Both die |
-| 2 | **The shill is a tip too** — delete the brand name from slide 3. Is what's left still one of the best tips in the deck? | You wrote an ad and put tips around it |
+| 2 | **Both brand touches survive deletion** — delete the brand from the shill and delete the close's last line. Is the shill still one of the best tips in the deck, and is slide N still a complete tip? | You wrote an ad and put tips around it |
 | 3 | **Still-complete** — freeze any slide as a PNG with no context. Does it stand alone? | You built a video. This format ships as pictures |
 
 Test 3 is the one that gets skipped, and it is structural, not cosmetic: **the slides are the
@@ -112,12 +120,12 @@ floor (fewer reads as a single post) and nine is the ceiling (nobody swipes to s
 
 | Slot | Slide | Job |
 |---|---|---|
-| **1** | **The cover** | Name the list literally + a hero picture. The thumbnail and the promise |
+| **1** | **The cover** | Name the list literally + a hero picture. The thumbnail and the promise. The only slide with **no topic rail** — its headline is the title |
 | **2** | Best tip | Pay the reader immediately. The single strongest thing you know |
 | **3** | Second-best tip | Pay them again, before you ask for anything |
 | **⌈N/2⌉** | **The shill** | The tip that happens to be what the product does. Dead centre |
 | **…** | Tips | Keep the average high. A weak slide here is where people leave |
-| **N** | The tip you'd tell a friend last | The one that earns the save. Not a CTA, not a recap |
+| **N** | **The close** — the tip you'd tell a friend last, plus the destination | The one that earns the save. Still a tip, never a recap. Its subtext carries the deck's ONE call to action, on a third line |
 
 **The shill sits in the MIDDLE — slide 4 of 7, slide 3 of 5, slide 5 of 9.** Not a fixed number: a
 fixed number breaks the moment the deck changes length. What is fixed is the *condition* —
@@ -133,14 +141,79 @@ set-up. The middle is the only slot where the reader is committed and not yet su
 earliest legal shill slot is **slide 4**. A 7-slide deck is cover + 6 tips with the ad at 4: two
 tips in front of it, three behind.
 
-**Exactly one shill in the deck.** Two brand mentions and it stops being a list with a product in
-it and becomes an ad with a list on it. There is no second-mention exception and no "but the last
-slide is subtle".
+**Exactly TWO brand touches in the deck, and they do different jobs.** Not one, and never three:
+
+| | Slide | What it is | What it does |
+|---|---|---|---|
+| **The shill** | the middle | The **mechanism** — the tip that happens to be what the product does | *Hints* the offer. The reader learns the product exists by watching it be useful |
+| **The close** | **N** | The **destination** — the bare address, on the last slide's third subtext line | *Asks.* The one call to action in the deck |
+
+A third mention is a brochure. Two mentions of the *same kind* — two mechanisms, or a destination
+in the middle as well as the end — is also three too many: **the deck may name the mechanism once
+and the address once.** There is no "but this one is subtle" exception.
 
 **Slides do not build.** No "1 of 6" counters, no "and finally", no "but here's the thing" carrying
 into the next slide, no numbered sequence where slide 4 depends on slide 3. Every slide is a
 standalone poster. Shuffle the middle slides at random; if the deck breaks, it was a video.
 
+### The topic rail — the title, quietly, on every slide
+
+**Slide 1 names the list. Slides 2…N carry that name again, small, in the same place, unchanged.**
+
+The reference deck does not do this and it is the one thing it is missing. The reason is the same
+reason test 3 exists: **a slide travels alone.** Somebody screenshots slide 4 and sends it to a
+friend. Somebody's swipe drops them on slide 5 with no cover. Somebody opens a save three weeks
+later. In every one of those cases the cover — the only thing that said what the list was about — is
+not on screen, and a great tip with no topic is a fortune cookie.
+
+The rail costs one line and fixes all three.
+
+| | |
+|---|---|
+| **What it says** | The deck's title, compressed to **≤5 words**, no article, no verb needed |
+| **Where** | A band at **14%**, directly above the headline, centred. Same y on every slide |
+| **How it looks** | Uppercase, 32px, weight **500**, letter-spacing `+0.16em`, the mode's ink at **55% opacity on `paper`, 78% on any `photo-*` mode** |
+| **Which slides** | **2 through N.** Never slide 1 — the cover's headline already IS the title |
+| **If the deck has no cover** | Then **all N**, slide 1 included. A coverless deck has nothing else naming the list, so the rail is the only thing doing that job |
+| **How much it varies** | **Not at all.** Byte for byte identical on every slide that has it |
+
+```
+cover headline:   6 rules for eating well in NYC
+topic rail:       EATING WELL IN NYC
+```
+
+- **It is the topic, never the brand.** A rail that reads `DISHCOVER.IO` is not a rail, it is a
+  watermark, and it turns one shill into N. See Rule 11.
+- **It is not a counter.** No `3/6`, no `TIP 4`, no progress dots. A counter is a build, and
+  *slides do not build* — it also tells a reader who landed mid-deck that they missed something,
+  which is the opposite of what the rail is for.
+- **It never comments on its slide.** `EATING WELL IN NYC` on all of them. The moment it becomes
+  `EATING WELL IN NYC — TIMING` it is a second headline competing with the first.
+- **Repetition is the mechanism, not a defect.** After slide 2 the reader stops seeing it, which is
+  correct: it is not addressed to the person swiping the deck. It is addressed to the person who
+  arrives at one slide with no deck around it.
+- **The 55% opacity is deliberate and it is the one place opacity is allowed here.** Elsewhere this
+  file bans faking a font weight with opacity, because that yields grey text where you wanted light
+  text. Here grey *is* the intent: the rail must be readable and must never be read first. Set it on
+  the rail layer, not on the type inside it.
+- **On a photo background the opacity goes UP to 78%, not down.** Translucent ink over a picture
+  loses contrast far faster than over a flat page, because the background it is compositing with is
+  neither flat nor known. `paper` is the only mode that can afford 55%. Carrying 55% into a photo
+  mode is the deck's first unreadable element and the last one you would look at — which is why
+  `slides-qa.py` grades the rail band on its own, against the ink's *composited* value.
+- **It sits on the shill slide too, identical.** Anything the rail does differently on slide 3 is a
+  free tell that slide 3 is the ad.
+
+**Why 32px and not smaller.** `vidfarm qa`'s `font-size` rule errors under 2.6% of canvas width
+(28px at 1080). A 24px rail would be both flagged and genuinely unreadable at feed scale. 32px is
+the floor with headroom.
+
+> **Note on the cover.** The main body of this harness requires slide 1 to be a cover; Appendix D
+> and Appendix E predate that and describe a coverless deck. Both shapes are buildable — pass
+> `--no-cover` to `build-deck.py` and the rail goes on all N slides instead of 2…N. **Decide which
+> shape you are building before you write the manifest**, because the shill slot moves with it: a
+> cover pushes the earliest legal shill from slide 3 to slide 4.
+
 ### Length
 
 `N × 3.0s`, exactly. Set `data-duration` on the root to `N * 3.0` and every slide clip to `3.0`.
@@ -176,10 +249,11 @@ Walk two more blocks.
 **The subtext is the REASON, and it is optional.** One or two lines, smaller, lighter, directly
 under the pictures. It answers "why" and never repeats the headline in other words.
 
-**The shill slide, and only the shill slide, may run to three lines.** It has one job the others do
-not — naming the offer — and squeezing that into two lines is what tempts you into cutting the
-reason down to a slogan. Two lines of tip, one line naming the tool. Every other slide stays at two:
-a three-line subtext elsewhere means the tip needed a paragraph, which means it was not a tip.
+**Two slides may run to a three-line subtext: the shill and the close.** Both have a job the others
+do not — naming the offer — and squeezing that into two lines is what tempts you into cutting the
+reason down to a slogan. Two lines of tip, one line naming the tool or the address. **Every other
+slide stays at two:** a three-line subtext anywhere else means the tip needed a paragraph, which
+means it was not a tip.
 
 ```
 headline:  Trust a 30-item menu over a 200-item one.
@@ -208,7 +282,8 @@ Rule 6 exists: a rung-4 number has to be a number you'd defend, not one you roun
 Each of these fails in a way that is invisible while you're writing it:
 
 - **The tip that is really the product.** "Use an app that lets you search by dish" on slide 5 is a
-  second shill wearing a hat. One shill, and it is slide 3.
+  second shill wearing a hat. One mechanism, in the middle — and the close names an address, not a
+  mechanism, which is exactly why the two do not collide.
 - **The tip nobody disputes.** "Read reviews before you go", "book ahead on weekends". Being true
   is not the bar; being *told* is.
 - **The tip that needs the reader to already be an expert.** If the instruction assumes they can
@@ -222,9 +297,11 @@ Each of these fails in a way that is invisible while you're writing it:
 
 ### The cold-viewer test
 
-Cover the brand name on slide 3. Read all N slides. **Does the deck still read as a good list?**
-If slide 3 is now visibly the weakest one, you did not write a tip — you wrote a placement and
-dressed it. Rewrite slide 3, not the deck.
+Cover the brand name on the shill slide and cover the close's last line. Read all N slides.
+**Does the deck still read as a good list?** If the shill is now visibly the weakest one, you did
+not write a tip — you wrote a placement and dressed it. Rewrite the shill, not the deck. And if
+slide N now reads as unfinished, the close was carrying it: the address was doing the work its
+headline should have done.
 
 ---
 
@@ -261,6 +338,72 @@ product does not fit this format — use `meme-recaption` or `ugc-reaction-green
 
 ---
 
+## The close — where the call to action goes
+
+The shill hints. **The close asks.** It is the last slide, it is still a tip, and its subtext runs
+to a third line that names the destination.
+
+```
+headline:  Ask what the kitchen is proud of.
+subtext:   Not "what's good" — everything is good. "Proud of" gets you the one
+           dish the chef would put their name on.
+           dishcover.io — every dish in NYC, searchable
+```
+
+Why slide N and not anywhere else:
+
+- **The reader has been paid N−1 times.** An address on slide 2 is a toll booth; an address on the
+  last slide is a business card handed over at the end of a conversation that went well.
+- **It cannot retroactively turn the deck into an ad**, which is the failure the middle-shill rule
+  exists to prevent — because slide N is still a *tip*. The deck resolves on the tip; the address
+  rides underneath it. Delete the third line and slide N is unchanged. That is the test.
+- **It is the slide with the lowest reach and the highest intent.** Everyone sees slide 1. The
+  people who reach slide N swiped past five tips on purpose. That is who the ask is for.
+
+### The grammar of the close
+
+**It is an address, not an instruction.** This is the whole rule and it is where decks go wrong.
+
+| | |
+|---|---|
+| ✅ | `dishcover.io — every dish in NYC, searchable` |
+| ✅ | `dishcover.io` (bare, no clause — always legal) |
+| ✅ | `built this: dishcover.io` |
+| ❌ | `Link in bio 👇` — an instruction, and a phrase the platforms down-rank |
+| ❌ | `Sign up free at dishcover.io` — an offer, not an address. Now the slide is an ad |
+| ❌ | `Find better restaurants tonight` — an outcome the deck cannot keep, and no address at all |
+
+- **≤6 words after the address**, one clause, describing what it *is* — never what it does for them.
+- **Lowercase, weight 500, same size as the rest of the subtext.** No 800, no colour, no arrow, no
+  emoji, no box. A styled CTA is the single loudest "this was an ad" tell in the format.
+- **No verb aimed at the reader.** Not "visit", "try", "check out", "grab", "get". The address is
+  the verb.
+- **No urgency, no price, no "free".** Those belong to a landing page, and on a slide they cost
+  more reach than they buy clicks.
+- **It never repeats the shill's mechanism.** The middle slide already said *how* it works. Saying
+  it twice is the third brand touch wearing a different hat.
+
+### What stays in the caption
+
+**The engagement bait does not move.** `save this`, `follow for more`, `comment below`, `link in
+bio` are still in `forbid_text` and still belong in the post caption — not because a CTA is
+forbidden, but because *those* CTAs are platform-gamed phrases that read as spam on a frame and
+get matched by suppression. The deck's on-frame ask is an address. The caption's ask is engagement.
+They are different asks and they live in different places.
+
+> **If you want `link in bio` on the frame anyway**, delete it from `forbid_text` in the front
+> matter — it is your deck. Know the trade: you gain an instruction the reader did not need (the
+> address is already on screen) and you take a reach penalty on every slide the phrase appears on.
+
+### When the product has no address worth naming
+
+Some offers are a person, a newsletter, a shop inside another platform. The close still works —
+name the thing a reader would type. If there is genuinely nothing to type, **drop the close** and
+end on a plain tip. A deck with a good last tip and no CTA outperforms a deck with a CTA nobody
+can act on. `slides-qa.py --no-close` grades that shape without complaining.
+
+---
+
 ## Casting the pictures
 
 ### The art class — pick one, hold it across the whole deck
@@ -419,9 +562,12 @@ slide you opened.** They live in `MODES` in Appendix A.
   the photograph on the `<img>`; carry the dim on a separate flat colour layer above it. Identical
   result, and it is not modal staging.
 - **The wash is a BANDED GRADIENT, not a flat tint.** A flat wash has to be deep enough for the
-  worst band, which flattens the whole picture into mud. A gradient that is strong across the
-  headline band (0–32%) and the subtext band (72–100%) and weak through the cutout band (44–64%)
+  worst band, which flattens the whole picture into mud. A gradient that is strong across the rail
+  + headline band (0–36%) and the subtext band (72–100%) and weak through the cutout band (48–64%)
   buys the contrast exactly where the text is and lets the photograph breathe where it doesn't.
+  **The rail lives inside the headline's half of the gradient**, so it inherits the headline's
+  contrast for free — that is why it was put above the headline rather than in the air below the
+  cutouts, where it would have needed a fourth wash stop of its own.
 - **The photo is scaled to 110% and offset −5%.** A blurred image at exactly 100% shows a pale
   vignette on all four edges — the blur sampling past the frame — and it reads as a rendering bug.
 - **Rule 5 inverts on a photo.** On `paper`, a text shadow is decoration and banned. On a photo it
@@ -441,7 +587,9 @@ only find out on the slide you did not open.
 
 `slides-qa.py --ink light|dark` (Appendix C) recovers each text band's background by **blurring the
 band hard enough to erase the glyphs**, takes the worst-case patch of what is left, and computes the
-WCAG contrast ratio against the mode's known ink colour.
+WCAG contrast ratio against the mode's known ink colour. It measures **three** bands — rail,
+headline, subtext — and the rail is graded against its *effective* ink, ink colour composited at the
+rail's own opacity, because a 78%-opacity white is not white.
 
 | Ratio | Verdict |
 |---|---|
@@ -449,6 +597,10 @@ WCAG contrast ratio against the mode's known ink colour.
 | 4.5–7:1 | Passes AA. Fine for 70px display type; tighten if you can |
 | **≥ 7:1** | Ship it |
 
+**The rail is graded on a different bar: fail under 3:1, warn under 3.5:1.** It is 32px, which
+clears WCAG's large-text threshold, and it is deliberately quieter than the headline — the
+documented `paper` default measures about 3.9:1. Grading it at 4.5 would fail every correct deck.
+
 Measured on the reference build: `photo-darken` and `photo-lighten` clear 7:1 on every band of every
 slide. **`photo-gray` lands at 6.1–6.9:1 and is the mode with the least headroom** — greyscale
 throws away the colour separation the other two lean on, so it is squeezed between "picture still
@@ -462,17 +614,21 @@ full-bleed photograph has ink everywhere). On a photo deck the contrast gate rep
 
 ### The layout grid
 
-Three horizontal bands, held identically on every slide. Read off the reference and expressed as
-percentages of the 1080×1920 canvas:
+Four horizontal bands, held identically on every slide. Read off the reference (which has three;
+the rail is the addition) and expressed as percentages of the 1080×1920 canvas:
 
 ```
   0% ─────────────────────────────────────
         (air — nothing lives here)
  14% ┌───────────────────────────────────┐
+     │  TOPIC RAIL   1 line, slides 2..N │
+16.5%└───────────────────────────────────┘
+        (air)
+ 19% ┌───────────────────────────────────┐
      │  HEADLINE   centred, 1-3 lines    │
- 30% └───────────────────────────────────┘
+ 35% └───────────────────────────────────┘
         (air)
- 34% ┌───────────────────────────────────┐
+ 37% ┌───────────────────────────────────┐
      │                                   │
      │  CUTOUTS   1-3 objects + marks    │
      │                                   │
@@ -485,9 +641,15 @@ percentages of the 1080×1920 canvas:
 100% ─────────────────────────────────────
 ```
 
+- **The headline starts at 19%, not 14%.** The rail took the top band, so the headline moved down
+  and the cutout band lost 3% off its top. Nothing else changed. If you are reading an older build
+  whose headlines start at 14%, it predates the rail — move them, do not add the rail above them
+  and let it collide with the reserved zone.
 - **The bands do not move between slides.** A headline that is 2 lines on slide 1 and 3 lines on
-  slide 2 still starts at 14%. Vertically centring each slide's content individually is the single
+  slide 2 still starts at 19%. Vertically centring each slide's content individually is the single
   most common way a deck acquires a "wobble" nobody can name — and it is invisible slide by slide.
+- **Slide 1 leaves the rail band EMPTY.** It does not move its headline up to fill the gap. The
+  cover's headline sits at 19% like every other headline, or the deck wobbles on its first cut.
 - **The bottom 20% and the top 14% stay empty.** That is where the platform puts the caption, the
   username, the buttons, and the swipe affordance.
 - **Everything is centred.** The headline block centres, the cutout group centres, the subtext
@@ -534,10 +696,15 @@ looking like itself.
 
 | Role | Face | Weight | Size | Line height |
 |---|---|---|---|---|
+| Topic rail | same family | **500** | 32px, uppercase, `+0.16em`, 55% opacity | 1.2 |
 | Headline — emphasis | TikTok Sans *or* Montserrat | **800** | 76px | 1.22 |
 | Headline — body | same family | **500** | 76px | 1.22 |
 | Subtext | same family | **500** | 40px | 1.34 |
 
+**Three sizes now, not two.** The old count was two and the contact-sheet review asked you to count
+them; the answer is three, and the third is the rail at one fixed size on every slide. A fourth size
+anywhere is the defect that check is looking for.
+
 `vidfarm qa`'s `font-regime` rule errors on Inter / Roboto / Arial / system-ui and warns on any
 family outside the imported regime (Montserrat, TikTok Sans, Abel, Source Code Pro, Yesteryear).
 **Montserrat is the closest match to the reference** and is in the regime.
@@ -717,12 +884,14 @@ fraction of the reach the format is for.
 Every rule has its reason attached. A rule without its reason gets argued away by the next agent
 that reads this file.
 
-### Rule 1 — one shill, on slide 3, written as a tip
+### Rule 1 — two brand touches: the mechanism in the middle, the address at the end
 
-Any other placement or count changes what the deck *is*. Slide 1 makes it an ad, slide N makes it a
-pitch, two mentions make it a brochure. The shill is in the middle because that is where the reader
-is already paid and not yet suspicious. **Enforced:** the cold-viewer test, and `slides-qa.py`
-checks that exactly one slide contains the brand token.
+Any other placement or count changes what the deck *is*. A mechanism on slide 1 makes it an ad; a
+mechanism on slide N makes the whole list retroactive set-up for a pitch; a third touch anywhere
+makes it a brochure. The shill is in the middle because that is where the reader is already paid
+and not yet suspicious. The close is at the end because that is where they have been paid N−1
+times and are choosing to still be there. **Enforced:** the cold-viewer test, and `slides-qa.py`
+checks the brand token appears on exactly the shill slide and the close slide.
 
 ### Rule 2 — every slide is a complete still
 
@@ -732,9 +901,10 @@ on, because the carousel is the deliverable and the video is not.
 
 ### Rule 3 — the layout bands never move
 
-Headline at 14%, cutouts 34–66%, subtext at 73%, on every slide, whatever the content length. Per
-slide it looks fine either way; across the deck, individually-centred content produces a vertical
-wobble that reads as sloppy and that nobody can name while looking at one slide.
+Rail at 14%, headline at 19%, cutouts 37–66%, subtext at 73%, on every slide, whatever the content
+length. Per slide it looks fine either way; across the deck, individually-centred content produces a
+vertical wobble that reads as sloppy and that nobody can name while looking at one slide. The cover
+has no rail and still starts its headline at 19% — an empty band is not a gap to close.
 
 ### Rule 4 — one art class, one light direction, one canvas colour
 
@@ -764,17 +934,30 @@ An attribution line on the frame is brand chrome, and this format bans brand chr
 reason the others do. The carousel always has a caption; use it. If a licence forces on-image
 credit, change the asset.
 
-### Rule 9 — the bait is in the post caption
+### Rule 9 — the frame carries an address, the caption carries the bait
 
-The deck ends on a tip. The comment prompt, the save prompt and the follow prompt all live in the
-caption where the platform expects them. A slide that says "save this" spends the slot that could
-have held the tip that made them save it.
+The deck ends on a tip **with the destination under it** — one line, lowercase, no verb (see *The
+close*). Everything else that asks the reader for something stays in the caption: the comment
+prompt, the save prompt, the follow prompt. A slide that says "save this" spends the slot that
+could have held the tip that made them save it, and the phrase is down-ranked besides. An address
+is information; "link in bio" is an instruction the reader did not need, because the address is
+already on screen.
 
 ### Rule 10 — production floor
 
 1080×1920, 30fps, `N × 3.0s` exactly, hard cuts, one canvas, no letterboxing, text at frame 0 of
 every slide, `data-duration` on the root equal to the sum.
 
+### Rule 11 — the topic rail names the TOPIC, on slides 2…N, unchanged
+
+One string, ≤5 words, identical byte for byte, at 14% on every slide except the cover. It exists
+because a slide travels alone — screenshotted, reposted, landed on mid-swipe, reopened from a save
+three weeks later — and a tip with no topic attached is a fortune cookie. **It is never the brand**
+(that is N shills, and Rule 1 says one), **never a counter** (that is a build, and Rule 2 says
+none), and never a comment on the slide it sits on. If it varies at all, it has stopped being a rail
+and become a second headline. **Enforced:** `slides-qa.py` measures its band on every slide and
+rejects the brand token appearing in it.
+
 ---
 
 ## Cost-saving mode — the whole deck at $0
@@ -827,7 +1010,7 @@ Sizing a round, capacity, and the ledger belong to `vidfarm experiment`, not to
 | Gate | Tool | What it can actually see |
 |---|---|---|
 | **1. Composition** | `vidfarm qa ./work --harness ./sticker-slideshow-tips.md` | The DOM: font regime, slop, safe zone, frame 0, the `checks:` block above |
-| **2. The stills** | `slides-qa.py ./slides [--ink light\|dark]` *(Appendix C)* | The exported PNGs: blank slides, canvas drift, band alignment, cutout alpha, the one-shill count |
+| **2. The stills** | `slides-qa.py ./slides [--ink light\|dark]` *(Appendix C)* | The exported PNGs: blank slides, canvas drift, band alignment, cutout alpha, the two-brand-touch placement, the topic rail on 2…N, the close's grammar |
 | **3. The sequence** | **You**, on `slides/contact-sheet.png` | Whether it reads as one deck. Nothing else can see this |
 | **4. The tips** | **You**, cold | Whether any of it is worth saving. No tool has an opinion here |
 
@@ -876,6 +1059,9 @@ in the edit.
 - [ ] N is between 5 and 9 (cover included), and every slide is exactly 3.0s.
 - [ ] Slide 2 is the strongest tip in the deck, not the easiest one to illustrate.
 - [ ] Shuffling two middle tips breaks nothing.
+- [ ] The topic rail is written once, ≤5 words, and is the TOPIC — not the brand, not a counter.
+- [ ] It appears on slides 2…N, identical byte for byte, and NOT on the cover.
+- [ ] Screenshot slide 4 alone and show it to someone: they can say what list it came from.
 
 **The tips**
 
@@ -889,11 +1075,23 @@ in the edit.
 **The shill**
 
 - [ ] It sits in the MIDDLE: at least two real tips before it and two after it, cover excluded.
-- [ ] It is the only slide naming the offer.
+- [ ] It is the only slide naming the offer's MECHANISM. The close names the address, and nothing
+      else in the deck names either.
 - [ ] With the brand name covered, it is still one of the two best tips in the deck.
 - [ ] It states a MECHANISM the reader could follow with no product at all.
 - [ ] The brand is lowercase, in the subtext, in the 500 weight, named once.
-- [ ] Slide 3 is visually indistinguishable from the others — same layout, type, colour, art class.
+- [ ] The shill slide is visually indistinguishable from the others — same layout, type, colour,
+      art class.
+
+**The close**
+
+- [ ] Slide N is still a TIP. Delete its last line and the slide is unchanged.
+- [ ] The last line is an ADDRESS, plus at most 6 words saying what it is.
+- [ ] No verb aimed at the reader, no "free", no urgency, no emoji, no arrow, no box.
+- [ ] It does not restate the shill's mechanism.
+- [ ] Lowercase, weight 500, same size as the subtext. Nothing about it is styled differently.
+- [ ] Exactly two brand touches in the whole deck: the shill and the close.
+- [ ] If there is nothing worth typing, there is no close — and slide N is a plain tip.
 
 **The pictures**
 
@@ -914,8 +1112,11 @@ in the edit.
 - [ ] Photo mode: hand-drawn marks use `{ink}`; photographic cutouts are never recoloured.
 - [ ] Photo mode: the blur is on the image and the dim is a separate flat layer (no `modal-scrim`).
 - [ ] Paper mode: one canvas colour across the deck. No footage, gradient, texture, border or card.
-- [ ] Headline band starts at 14% on every slide, whatever its line count.
+- [ ] Rail band at 14%, headline band at 19% on every slide, whatever its line count.
+- [ ] The cover leaves the rail band empty and does NOT move its headline up to fill it.
 - [ ] Nothing in the top 14% or bottom 20%.
+- [ ] Rail measured at ≥3:1 (its own bar, not the headline's 4.5): 0.55 opacity on `paper`,
+      0.78 on any photo mode.
 - [ ] Two weights of ONE self-hosted family, 800 and 500, no fallback chain in the stack.
 - [ ] 2–4 emphasised runs per headline, weight-only, verb and noun.
 - [ ] No stroke, no text shadow, no plate behind any text.
@@ -938,15 +1139,22 @@ actually ships broken. **Open `slides/contact-sheet.png` and read it as one imag
 
 - [ ] **Does the headline sit at the same height on all N slides?** The wobble is invisible one
       slide at a time and obvious on the sheet.
-- [ ] **How many type sizes are there?** There should be two. Count them.
+- [ ] **How many type sizes are there?** There should be three — rail, headline, subtext. Count
+      them.
+- [ ] **Cover the headline and the subtext on every slide. Does the rail still read the same on
+      all of them, at the same height?** This is the check the pixel gate cannot do properly at
+      32px, and a rail that drifted or changed wording is obvious here and nowhere else.
+- [ ] **Did your eye go to the rail first on any slide?** Then it is too loud. Drop the opacity.
 - [ ] **Do the cutouts look like one set?** Same class, same rendering, same light, plausible
       relative scale.
 - [ ] **Is the canvas the same white on all N?** A cutout with a near-white background rectangle
       makes its slide read a different colour and only shows up beside its neighbours.
 - [ ] **Which slide is the weakest?** There is one. Either rewrite it or delete it — a five-slide
       deck with no weak slide beats a six-slide deck with one.
-- [ ] **Can you tell which slide is the ad?** If your eye goes to slide 3 before you read it, it is
-      styled differently and Rule 1 is broken.
+- [ ] **Can you tell which slide is the ad?** If your eye goes to the shill before you read it, it
+      is styled differently and Rule 1 is broken.
+- [ ] **Does the last slide read as a tip or as an ending card?** Cover its last line. If what is
+      left feels like it is missing something, the address was carrying the slide.
 - [ ] **Would you save this?** Not "is it correct". Would you save it.
 
 Then watch the MP4 end to end at 1× with sound off, which is how it will be seen.
@@ -963,8 +1171,10 @@ Diagnose by slot, never by declaring the deck bad.
 | Views, drop-off at slide 2–3 | **The shill** | It reads as an ad. Cover the brand: if the slide is now weak, rewrite it as a tip |
 | Swiped to the end, no saves | **The tips** | They are on rung 1–2. Add a mechanism and a number to each |
 | Saves, no comments | **The caption** | The bait is missing or generic. It lives in the caption, so fix the caption, not the deck |
-| Good numbers, no clicks to the offer | **Nothing is broken** | This format seeds a mechanism; it does not close. Judge it on saves and branded search, not clicks |
+| Good numbers, no clicks to the offer | **The close** | Check it exists and is an address, not a promise. Then check expectations: this format seeds a mechanism and hands over an address — most of its return arrives as branded search, not as clicks. Judge it on saves and branded search first |
+| Reach collapsed on an otherwise fine deck | **The close** | An instruction phrase reached the frame — "link in bio", "sign up free". Gate 2 check 8 catches it; a suppressed deck looks exactly like a bad deck |
 | "Looks AI-made" in the comments | **The sequence** | Almost always the band wobble, mixed art classes, or a stroke on the type. Contact sheet, gate 3 |
+| Reshares of one slide, no traffic back to the deck | **The topic rail** | It is missing, or it names the brand instead of the topic. A lifted slide has to say what list it came from |
 
 ---
 
@@ -996,7 +1206,13 @@ Four things this exists to get right:
     first_frame_text then fail on a composition whose frame 0 is actually fine.
   * THE BANDS DO NOT MOVE. Every headline is pinned to the same top, whatever
     its line count. Vertically centring per slide is the "wobble" that only
-    shows up on the contact sheet (Rule 3).
+    shows up on the contact sheet (Rule 3). The cover leaves the rail band
+    empty and STILL starts its headline at head_top — do not close the gap.
+  * THE TOPIC RAIL IS EMITTED FROM ONE STRING. --topic is written to slides
+    2..N verbatim, so it cannot drift, cannot pick up a counter, and cannot
+    say something different on the shill slide (Rule 11). It is deliberately
+    NOT a TSV column: a per-slide column is a per-slide value, and a rail with
+    per-slide values is a second headline.
   * GEOMETRY IS INLINE, TYPOGRAPHY IS IN THE CLASS. `.clip{inset:0}` has to be
     reset per layer with `inset:auto`, that reset lives in the inline style
     attribute, and inline BEATS the stylesheet — so a `top:` in a CSS class
@@ -1015,15 +1231,19 @@ Background modes (--bg-mode):
 
 Any photo mode needs a 4th TSV column naming that slide's background image.
 
-usage: build-deck.py <slides.tsv> <out.html> [--title=…] [--canvas=#FDFCFA]
+usage: build-deck.py <slides.tsv> <out.html> --topic="EATING WELL IN NYC"
+                     [--title=…] [--canvas=#FDFCFA] [--no-cover]
                      [--bg-mode=paper|photo-darken|photo-lighten|photo-gray]
-                     [--head-px=70] [--sub-px=40]
+                     [--head-px=70] [--sub-px=40] [--rail-px=32]
+
+--no-cover says slide 1 is a TIP, not a cover — then the rail goes on all N
+slides, because nothing else in the deck names the list.
 """
 import html, sys
 
 SLIDE_SEC = 3.0
 W, H = 1080, 1920
-BANDS = dict(head_top=14.0, cut_top=34.0, cut_bot=66.0, sub_top=73.0)
+BANDS = dict(rail_top=14.0, head_top=19.0, cut_top=37.0, cut_bot=66.0, sub_top=73.0)
 
 # Each mode is a complete legibility contract: what happens to the photo, what
 # flat wash sits over it, what colour the text is, and what that text needs to
@@ -1034,36 +1254,61 @@ MODES = {
         filter=None, wash=None, ink="#111111",
         # On a flat page the contrast is already maximal and a shadow is the
         # thing that makes type read as machine-made (Rule 5).
-        shadow="none", cut_shadow="none"),
+        shadow="none", cut_shadow="none",
+        # The rail must recede without disappearing. A flat page can carry 0.55;
+        # a photograph cannot — translucent ink over a picture loses contrast far
+        # faster than over paper, so every photo mode runs the rail at 0.78.
+        rail_opacity=0.55),
     "photo-darken": dict(
         filter="blur(18px) brightness(0.38) saturate(0.90)",
-        wash="linear-gradient(180deg,rgba(0,0,0,0.62) 0%,rgba(0,0,0,0.62) 32%,rgba(0,0,0,0.22) 44%,rgba(0,0,0,0.22) 64%,rgba(0,0,0,0.62) 72%,rgba(0,0,0,0.62) 100%)", ink="#FFFFFF",
+        wash="linear-gradient(180deg,rgba(0,0,0,0.62) 0%,rgba(0,0,0,0.62) 36%,rgba(0,0,0,0.22) 48%,rgba(0,0,0,0.22) 64%,rgba(0,0,0,0.62) 72%,rgba(0,0,0,0.62) 100%)", ink="#FFFFFF",
         shadow="0 2px 18px rgba(0,0,0,0.55)",
-        cut_shadow="drop-shadow(0 14px 30px rgba(0,0,0,0.55))"),
+        cut_shadow="drop-shadow(0 14px 30px rgba(0,0,0,0.55))",
+        rail_opacity=0.78),
     "photo-lighten": dict(
         filter="blur(18px) brightness(1.62) saturate(0.42)",
-        wash="linear-gradient(180deg,rgba(255,255,255,0.78) 0%,rgba(255,255,255,0.78) 32%,rgba(255,255,255,0.46) 44%,rgba(255,255,255,0.46) 64%,rgba(255,255,255,0.78) 72%,rgba(255,255,255,0.78) 100%)", ink="#111111",
+        wash="linear-gradient(180deg,rgba(255,255,255,0.78) 0%,rgba(255,255,255,0.78) 36%,rgba(255,255,255,0.46) 48%,rgba(255,255,255,0.46) 64%,rgba(255,255,255,0.78) 72%,rgba(255,255,255,0.78) 100%)", ink="#111111",
         shadow="0 2px 18px rgba(255,255,255,0.75)",
-        cut_shadow="drop-shadow(0 14px 30px rgba(0,0,0,0.30))"),
+        cut_shadow="drop-shadow(0 14px 30px rgba(0,0,0,0.30))",
+        rail_opacity=0.78),
     "photo-gray": dict(
         filter="blur(14px) grayscale(1) brightness(0.48)",
-        wash="linear-gradient(180deg,rgba(18,18,20,0.64) 0%,rgba(18,18,20,0.64) 32%,rgba(18,18,20,0.18) 44%,rgba(18,18,20,0.18) 64%,rgba(18,18,20,0.64) 72%,rgba(18,18,20,0.64) 100%)", ink="#FFFFFF",
+        wash="linear-gradient(180deg,rgba(18,18,20,0.64) 0%,rgba(18,18,20,0.64) 36%,rgba(18,18,20,0.18) 48%,rgba(18,18,20,0.18) 64%,rgba(18,18,20,0.64) 72%,rgba(18,18,20,0.64) 100%)", ink="#FFFFFF",
         shadow="0 2px 18px rgba(0,0,0,0.50)",
-        cut_shadow="drop-shadow(0 14px 30px rgba(0,0,0,0.50))"),
+        cut_shadow="drop-shadow(0 14px 30px rgba(0,0,0,0.50))",
+        rail_opacity=0.78),
 }
 
 args = [a for a in sys.argv[1:] if not a.startswith("--")]
 opts = dict(a.split("=", 1) for a in sys.argv[1:] if a.startswith("--") and "=" in a)
+flags = {a for a in sys.argv[1:] if a.startswith("--") and "=" not in a}
+has_cover = "--no-cover" not in flags
 manifest, out = args[0], args[1]
 canvas = opts.get("--canvas", "#FDFCFA")
 title = opts.get("--title", "tip deck")
 head_px = int(opts.get("--head-px", 70))
 sub_px = int(opts.get("--sub-px", 40))
+rail_px = int(opts.get("--rail-px", 32))
 bg_mode = opts.get("--bg-mode", "paper")
 if bg_mode not in MODES:
     sys.exit(f"unknown --bg-mode {bg_mode!r}; pick one of {', '.join(MODES)}")
 M = MODES[bg_mode]
 
+# The topic rail (Rule 11). Required, because "the deck with no rail" is not a
+# variant of this format — it is the failure mode the rail was added to fix.
+topic = opts.get("--topic", "").strip().upper()
+if not topic:
+    sys.exit('--topic is required: the deck title in <=5 words, e.g. '
+             '--topic="EATING WELL IN NYC". It goes on slides 2..N unchanged.')
+if len(topic.split()) > 5:
+    sys.exit(f"--topic is {len(topic.split())} words; the rail holds 5. "
+             f"Compress it — the cover carries the full title.")
+if any(ch.isdigit() for ch in topic) and "/" in topic:
+    sys.exit("--topic looks like a counter. The rail is not a progress indicator (Rule 11).")
+# `vidfarm qa`'s font-size rule errors below 2.6% of canvas width.
+if rail_px < W * 0.026:
+    sys.exit(f"--rail-px {rail_px} is below the {W * 0.026:.0f}px floor; qa will error on it.")
+
 rows = []
 for raw in open(manifest):
     raw = raw.rstrip("\n")
@@ -1101,6 +1346,14 @@ CSS = f"""
 
  /* THE BANDS are set INLINE, not here — see the module docstring. Geometry
     inline, typography in the class. */
+
+ /* The topic rail. Uppercase + tracked + small is what makes it subordinate;
+    the opacity is what stops it being read FIRST. This is the one place in the
+    deck where opacity is legitimate — everywhere else it is a way of faking a
+    font weight, which yields grey text where you wanted light text. */
+ .rail{{font-family:'Deck';font-weight:500;font-size:{rail_px}px;line-height:1.2;
+   letter-spacing:0.16em;text-transform:uppercase;color:{M['ink']};
+   opacity:{M['rail_opacity']};text-align:center;text-shadow:{M['shadow']}}}
  .head{{font-family:'Deck';font-weight:500;font-size:{head_px}px;line-height:1.22;
    letter-spacing:-0.01em;color:{M['ink']};text-align:center;
    text-shadow:{M['shadow']}}}
@@ -1141,6 +1394,14 @@ for i, (head, sub, cuts, bg) in enumerate(rows):
         if M["wash"]:
             layers.append(layer("shape", f"s{n:02d}-wash", t, "wash",
                                 "inset:0;z-index:1", label=f"slide {n} wash"))
+    # The rail: slides 2..N when there is a cover. Slide 1's headline IS the
+    # title, so a rail above it would repeat itself — but the cover still leaves
+    # the band EMPTY rather than moving its headline up, or the deck wobbles on
+    # its first cut. With --no-cover nothing else names the list, so all N.
+    if n > 1 or not has_cover:
+        layers.append(layer("text", f"s{n:02d}-rail", t, "rail",
+                            f"inset:auto;left:8%;width:84%;top:{BANDS['rail_top']}%;z-index:3",
+                            html.escape(topic), label=f"slide {n} topic rail"))
     layers.append(layer("text", f"s{n:02d}-head", t, "head",
                         f"inset:auto;left:8%;width:84%;top:{BANDS['head_top']}%;z-index:3",
                         rich(head), label=f"slide {n} headline"))
@@ -1174,7 +1435,8 @@ open(out, "w").write(f"""<!doctype html>
 </div></body></html>
 """)
 print(f"{len(rows)} slides · {total}s · {len(layers)} flat layers · "
-      f"bg-mode {bg_mode} · ink {M['ink']} · {out}")
+      f"bg-mode {bg_mode} · ink {M['ink']} · rail “{topic}” "
+      f"on slides {2 if has_cover else 1}-{len(rows)} · {out}")
 print("stills at: " + ",".join(str(round(i * SLIDE_SEC + 1.5, 1)) for i in range(len(rows))))
 ```
 
@@ -1222,7 +1484,8 @@ echo "preview : $DIR/final.mp4"
 runs on the pixels that ship.
 
 ```bash
-python3 slides-qa.py ./work/slides --brand dishcover.io --shill-slide 3
+python3 slides-qa.py ./work/slides --brand dishcover.io --shill-slide 3 \
+        --topic "EATING WELL IN NYC"
 ```
 
 ```python
@@ -1230,7 +1493,7 @@ python3 slides-qa.py ./work/slides --brand dishcover.io --shill-slide 3
 """
 Grade an exported tip deck on the PNGs, not on the markup.
 
-Six checks, each one a defect that passes every DOM check and every per-slide
+Seven checks, each one a defect that passes every DOM check and every per-slide
 eyeball, and shows up only across the set:
 
   1. no slide is blank            — an entrance that had not settled at +1.5s
@@ -1243,22 +1506,44 @@ eyeball, and shows up only across the set:
      brand                          manifest rather than by OCR
   6. every cutout's alpha is
      mostly 0-or-255              — run separately on media/cutouts/*.png
+  7. the topic rail is on slides  — Rule 11. A missing rail is invisible while
+     2..N and NOT on slide 1        you read the deck in order and is the whole
+                                    point of the deck when a slide travels alone
+  8. the close is an ADDRESS, not  — Rule 9. "Sign up free" and "link in bio"
+     an instruction                  are what a CTA turns into when nobody checks
 
 usage: slides-qa.py <slides-dir> [--brand <token>] [--shill-slide <n>] [--ink light|dark]
                                  [--manifest <slides.tsv>] [--cutouts <dir>]
+                                 [--topic <rail text>] [--rail-opacity 0.55] [--no-cover]
+                                 [--no-close]
 """
 import glob, os, subprocess, sys
 
 d = sys.argv[1]
-opt = dict(zip(sys.argv[2::2], sys.argv[3::2]))
+# `opt` pairs positionally, so a bare flag would shift every option after it by
+# one and silently give you --brand="--shill-slide". Strip the bare flags first.
+BARE = {"--no-cover", "--no-close"}
+argv = [a for a in sys.argv[2:] if a not in BARE]
+has_cover = "--no-cover" not in sys.argv
+no_close = "--no-close" in sys.argv
+opt = dict(zip(argv[0::2], argv[1::2]))
 brand = opt.get("--brand")
 shill = int(opt.get("--shill-slide", 3))
 manifest = opt.get("--manifest")
 cutdir = opt.get("--cutouts")
 ink = opt.get("--ink")   # "light" | "dark" — the mode's text colour
+topic = (opt.get("--topic") or "").strip()
+rail_alpha = float(opt.get("--rail-opacity", 0.78 if ink else 0.55))
 
 fails, warns = [], []
 
+# The band geometry, in pixels on a 1080x1920 canvas. Kept here rather than
+# inlined so the rail's arrival could not silently leave the headline scan
+# looking at the rail — which is exactly what happened on the first attempt.
+RAIL_Y, RAIL_H = 262, 52       # 14.0% .. 16.5%
+HEAD_Y, HEAD_H = 355, 300      # 19.0% .. 34.6% (3 lines at 76/1.22)
+SUB_Y, SUB_H = 1395, 175       # 73.0% .. 82.1%
+
 
 def probe(png, crop):
     """Mean RGB of a crop, via ffmpeg — no PIL dependency."""
@@ -1306,10 +1591,13 @@ if not ink:
             fails.append(f"{os.path.basename(s)}: canvas {c} drifts from {base}")
 
 # --------------------------------------------------- 3. the headline band is level
+# The scan starts BELOW the rail band. Starting at 200px (as this did before the
+# rail existed) locks onto the rail on every slide except the cover, reports a
+# perfectly level 269px on 5 of 6 slides, and hides a real headline wobble.
 tops = []
 for s in (slides if not ink else []):
     top = None
-    for y in range(200, 700, 12):        # scan the headline band, 10%-36%
+    for y in range(HEAD_Y - 20, 760, 12):    # scan the headline band, 17%-40%
         if rowdark(s, y, 12) > 0.06:
             top = y
             break
@@ -1328,14 +1616,92 @@ for s in (slides if not ink else []):
     if rowdark(s, 1536, 384) > 0.02:
         warns.append(f"{os.path.basename(s)}: ink in the bottom 20% (caption/buttons)")
 
-# ------------------------------------------------------------ 5. exactly ONE shill
+# ------------------------------------------- 5. exactly TWO brand touches, placed
+# The shill (mechanism, middle) and the close (address, slide N). A third is a
+# brochure; a mechanism on slide N turns every tip before it into set-up.
 if brand and manifest:
-    hits = [i + 1 for i, line in enumerate(
-        l for l in open(manifest) if l.strip() and not l.lstrip().startswith("#"))
-        if brand.lower() in line.lower()]
-    if hits != [shill]:
+    lines = [l for l in open(manifest) if l.strip() and not l.lstrip().startswith("#")]
+    hits = [i + 1 for i, line in enumerate(lines) if brand.lower() in line.lower()]
+    want = [shill] if no_close else sorted({shill, len(lines)})
+    if hits != want:
         fails.append(f"brand '{brand}' appears on slides {hits or 'none'} — "
-                     f"Rule 1 requires exactly [{shill}]")
+                     f"Rule 1 requires exactly {want} "
+                     f"({'shill only, --no-close' if no_close else 'shill + close'})")
+    # The close is an ADDRESS. A mechanism restated on slide N is the third touch
+    # wearing a hat, and it is the one the eye forgives and the reader does not.
+    if not no_close and len(lines) >= 1 and brand.lower() in lines[-1].lower():
+        close_sub = (lines[-1].split("\t") + ["", ""])[1]
+        if close_sub.count("|") > 2:
+            warns.append(f"the close's subtext runs to {close_sub.count('|') + 1} lines — "
+                         f"two lines of tip, one line of address, and no more")
+
+# -------------------------------------------- 8. the close is an address, not an ad
+# The CTA is the one line in the deck with gravity: every rewrite pulls it toward
+# "sign up free today". These are the phrases it lands on, and each one costs
+# reach on a carousel while adding nothing the address did not already say.
+INSTRUCTIONS = ("link in bio", "sign up", "free trial", "get started", "book a demo",
+                "download now", "download the app", "learn more", "swipe up",
+                "visit our", "check out", "click", "tap the", "try it", "try now",
+                "grab your", "don't miss", "limited time", "act now", "save this",
+                "follow for more", "comment below")
+if manifest and not no_close:
+    lines = [l for l in open(manifest) if l.strip() and not l.lstrip().startswith("#")]
+    if lines:
+        close = lines[-1].lower()
+        for phrase in INSTRUCTIONS:
+            if phrase in close:
+                fails.append(f"the close contains '{phrase}' — the CTA is an ADDRESS, not "
+                             f"an instruction (Rule 9). The reader can already see the "
+                             f"address; the phrase only costs reach")
+        # An outcome promise is the other direction the close drifts in, and it
+        # only ever appears in the subtext — the headline is still a plain tip.
+        close_sub = (lines[-1].split("\t") + ["", ""])[1].lower()
+        for phrase in ("faster", "easier", "never again", "in seconds", "in minutes"):
+            if phrase in close_sub:
+                warns.append(f"the close may be promising an outcome ('{phrase}') — it "
+                             f"names what the product IS, never what it does for them")
+
+# --------------------------------------------------------- 7. the topic rail
+# The rail is the one element that does nothing for the reader who swipes the
+# deck in order — so nothing about reading the deck tells you it is missing.
+# It earns its place when ONE slide travels alone, which is the moment nobody
+# is testing. Hence a machine check.
+if topic and brand and brand.lower() in topic.lower():
+    fails.append(f"the topic rail contains the brand '{brand}' — that is N shills on "
+                 f"N slides, not one (Rule 11). The rail names the TOPIC")
+if topic and (len(topic.split()) > 5 or "/" in topic):
+    fails.append(f"topic rail '{topic}' is a counter or over 5 words (Rule 11)")
+
+
+def bandtexture(png, y, h):
+    """How much of a band deviates from its own median.
+
+    Glyphs deviate hard; a flat page does not, and neither does a
+    hard-blurred photo under a wash. This is a SMOKE TEST — 32px type is
+    below what a 256x8 downscale can characterise properly, so a missing
+    rail is a warn, not a fail, and the contact sheet is the real gate.
+    """
+    raw = subprocess.run(
+        ["ffmpeg", "-v", "error", "-i", png, "-vf",
+         f"crop=1080:{h}:0:{y},format=gray,scale=256:8", "-frames:v", "1",
+         "-f", "rawvideo", "-pix_fmt", "gray", "-"], capture_output=True).stdout
+    if not raw:
+        return 0.0
+    vals = sorted(raw)
+    med = vals[len(vals) // 2]
+    return sum(1 for b in raw if abs(b - med) > 25) / len(raw)
+
+
+rail_tex = [bandtexture(s, RAIL_Y, RAIL_H) for s in slides]
+railed = list(zip(slides, rail_tex))[1:] if has_cover else list(zip(slides, rail_tex))
+if has_cover and rail_tex[0] > 0.03:
+    warns.append(f"{os.path.basename(slides[0])}: the COVER has something in the rail "
+                 f"band. Slide 1's headline is the title; it carries no rail (Rule 11)")
+for s, tex in railed:
+    if tex <= 0.03:
+        warns.append(f"{os.path.basename(s)}: no topic rail found at 14%. Every slide "
+                     f"but the cover carries it — a slide that travels alone has "
+                     f"nothing else naming the list (Rule 11)")
 
 # ------------------------------------------------- 5b. TEXT CONTRAST (readability)
 # The one thing a photo background can silently destroy. Measuring beats looking:
@@ -1351,7 +1717,15 @@ def _lin(v):
     return c / 12.92 if c <= 0.03928 else ((c + 0.055) / 1.055) ** 2.4
 
 
-def band_contrast(png, y, h, ink_is_light):
+def band_contrast(png, y, h, ink_is_light, alpha=1.0):
+    """Contrast of a band's text against its own background.
+
+    `alpha` is the text layer's opacity. It is NOT cosmetic: a 55%-opacity
+    white is not white, and grading the topic rail as if it were is how a rail
+    that measures 3.1:1 gets signed off as "the headline is 9:1, we're fine".
+    The composite is done in gray against the SAME worst-case patch the
+    background is graded on, which is what the eye sees.
+    """
     raw = subprocess.run(
         ["ffmpeg", "-v", "error", "-i", png, "-vf",
          f"crop=1080:{h}:0:{y},format=gray,boxblur=24:2,scale=48:8",
@@ -1359,20 +1733,44 @@ def band_contrast(png, y, h, ink_is_light):
         capture_output=True).stdout
     if not raw:
         return None
-    vals = sorted(_lin(b) for b in raw)
-    L_ink = 1.0 if ink_is_light else _lin(0x11)
+    vals = sorted(raw)
     # 90th / 10th percentile rather than the absolute extreme: one stray blurred
     # highlight is not where the text sits, and grading against it fails a deck
     # that is genuinely fine.
-    L_bg = vals[int(len(vals) * 0.90)] if ink_is_light else vals[int(len(vals) * 0.10)]
+    g_bg = vals[int(len(vals) * 0.90)] if ink_is_light else vals[int(len(vals) * 0.10)]
+    g_ink = alpha * (255 if ink_is_light else 0x11) + (1 - alpha) * g_bg
+    L_ink, L_bg = _lin(g_ink), _lin(g_bg)
     hi, lo = max(L_ink, L_bg), min(L_ink, L_bg)
     return (hi + 0.05) / (lo + 0.05)
 
 
+# The rail is graded in EVERY mode, including `paper` — its opacity means it is
+# the only element whose contrast is a choice rather than a consequence.
+# Thresholds are lower than the headline's on purpose: at 32px it clears WCAG's
+# large-text bar (>=24px), and it is SUPPOSED to be quieter than the headline.
+# A rail as loud as its headline is a second headline.
+#
+# The documented default — #111 at 0.55 on a #FDFCFA page — measures ~3.9:1
+# through this same blur, which is why the warn sits at 3.5 and not at 4.5: a
+# rail graded like a headline fires on every correct paper deck. If you DO see
+# the warn on `paper`, the canvas is not the canvas — something is sitting under
+# the rail band.
+light = ink == "light"
+for s, _ in railed:
+    r = band_contrast(s, RAIL_Y, RAIL_H, light, rail_alpha)
+    if r is None:
+        continue
+    if r < 3.0:
+        fails.append(f"{os.path.basename(s)}: topic rail contrast {r:.1f}:1 — below the "
+                     f"3:1 large-text floor. Raise --rail-opacity (0.78 on any photo "
+                     f"mode) or deepen the wash")
+    elif r < 3.5:
+        warns.append(f"{os.path.basename(s)}: topic rail contrast {r:.1f}:1 — legal but "
+                     f"thin. A rail nobody can read is a layer nobody needed")
+
 if ink:
-    light = ink == "light"
     for s in slides:
-        for name, y, h in (("headline", 260, 340), ("subtext", 1395, 175)):
+        for name, y, h in (("headline", HEAD_Y, HEAD_H), ("subtext", SUB_Y, SUB_H)):
             r = band_contrast(s, y, h, light)
             if r is None:
                 continue
@@ -1422,9 +1820,13 @@ READER:    Lives here, eats out twice a week, keeps ending up somewhere fine
 OFFER:     dishcover.io — search restaurants by DISH instead of by restaurant
 ART CLASS: photographic cutout, one sheet, contact shadows off
 SHILL:     slide 3
+RAIL:      EATING WELL IN NYC   (all 6 slides — this deck is coverless, never the brand)
+SHAPE:     coverless (--no-cover). See the note under "The topic rail"
+CLOSE:     slide 6, third subtext line — "dishcover.io — every dish in NYC, searchable"
 ```
 
-**Post caption** (this is where the topic, the bait and the credits live)
+**Post caption** (this is where the topic, the bait and the credits live — the *address* is on
+slide 6, so the caption never repeats it)
 
 ```
 6 rules for eating well in NYC 🗽
@@ -1439,9 +1841,13 @@ Go at *5:30* or go at *9:45*.	The kitchen is calmest at the edges|of service. Sa
 *Search the dish*, not|the *restaurant*.	"Best birria in Queens" finds you dinner.|"Best Mexican near me" finds you a parking lot.|dishcover.io is built to search that way.	media/cutouts/query-scribble.png@14,42,26,-3 ; media/cutouts/arrow.png@42,46,14,0 ; media/cutouts/birria.png@58,38,28,5
 *One dish* decides the place.	Every good spot does one thing|better than anyone. Order that.	media/cutouts/signature-dish.png@36,37,30,-6
 Trust a *30-item* menu|over a *200-item* one.	A short menu means a kitchen that repeats.|A long one means a freezer.	media/cutouts/menu-short.png@20,38,26,-5 ; media/cutouts/menu-long.png@54,36,26,7
-*Walk two more blocks.*	Every train exit has a rent-funded|trap in the first 200 feet.	media/cutouts/subway-exit.png@34,39,32,-4
+*Walk two more blocks.*	Every train exit has a rent-funded|trap in the first 200 feet.|dishcover.io — every dish in NYC, searchable	media/cutouts/subway-exit.png@34,39,32,-4
 ```
 
+Slide 6 is **the close**: a real tip, with the address on a third subtext line. Delete that line and
+slide 6 is unchanged — which is the test. Note what the line does *not* say: no verb, no "free", no
+"link in bio", and no repeat of slide 3's mechanism.
+
 **Slide 3, read against the lock**
 
 - *Actionable?* Yes — search a dish name in any search box tonight.
@@ -1453,10 +1859,12 @@ Trust a *30-item* menu|over a *200-item* one.	A short menu means a kitchen that
 
 ```bash
 python3 build-deck.py slides.tsv work/index.html --title="dishcover — NYC eats deck" \
-                                                 --canvas="#FDFCFA"
+                                                 --canvas="#FDFCFA" --no-cover \
+                                                 --topic="EATING WELL IN NYC"
 vidfarm qa ./work --harness ./experimental/sticker-slideshow-tips.md
 bash export-deck.sh ./work 6 nyc-eats
 python3 slides-qa.py ./work/slides --brand dishcover.io --shill-slide 3 \
+        --topic "EATING WELL IN NYC" --no-cover \
         --manifest slides.tsv --cutouts ./work/media/cutouts
 open ./work/slides/contact-sheet.png     # gate 3 — the pass that actually matters
 ```
@@ -1471,16 +1879,33 @@ open ./work/slides/contact-sheet.png     # gate 3 — the pass that actually mat
    40px high. Fixed by pinning the band, not by padding the headline.
 4. `dishcover.io` was originally in the 800 weight. It read as a banner. Dropped to 500 with the
    rest of the subtext and the slide stopped looking like the ad slide.
+5. The topic rail was first written as `DISHCOVER.IO` — which looked tidy, and was six brand
+   mentions in a deck allowed one. Replaced with `EATING WELL IN NYC` (Rule 11).
+6. The close was first written `Find better food tonight → dishcover.io (free!)`. Three failures in
+   one line: an outcome the deck cannot keep, an instruction the reader did not need, and a price.
+   Rewritten as the address plus five words saying what it is.
+7. The rail is the one element whose contrast is set by a number you chose rather than by the
+   mode, so carrying `paper`'s 0.55 opacity into a `photo-darken` variant is the easy mistake:
+   the headline still measures 7:1+ and nothing looks wrong on any slide you open. Run
+   `slides-qa.py` with `--ink light --rail-opacity` set to whatever you actually shipped, and let
+   the rail band be graded on its own.
 
 ## Appendix E — the harness in one paragraph, for a handoff
 
 > Build a 6-slide tip carousel, 9:16, 1080×1920, on a flat `#FDFCFA` page — no footage, no
-> gradients, no cards. Each slide: one imperative tip as a centred headline (TikTok Sans or
-> Montserrat, two weights 800/500 mixed inside the sentence, 76px, pinned at 14% from the top,
-> hand-broken lines) plus an optional reason line at 73% (40px, 500) and 1–3 die-cut photographic
-> cutouts in the 34–66% band, rotated ±3–8°. No stroke, no shadow, no plate on any text. Slide 3
-> names dishcover.io once, lowercase, in the subtext, and is written so it is still a good tip with
-> the brand removed. No title slide; the topic and the comment bait are the post caption. Ship the
-> six PNGs as the deliverable (durable My Files, ordered in `slides.json`) and an MP4 at exactly
-> 3.0s per slide with hard cuts as the preview. Full rules:
+> gradients, no cards. Each slide: the deck's topic as a small rail at 14% (`EATING WELL IN NYC`,
+> uppercase, 32px, weight 500, `+0.16em`, 55% opacity, IDENTICAL on every slide — it is what tells
+> a screenshotted slide what list it came from, and it is never the brand and never a counter), one
+> imperative tip as a centred headline (TikTok Sans or Montserrat, two weights 800/500 mixed inside
+> the sentence, 76px, pinned at 19% from the top, hand-broken lines), an optional reason line at 73%
+> (40px, 500), and 1–3 die-cut photographic cutouts in the 37–66% band, rotated ±3–8°. No stroke, no
+> shadow, no plate on any text. Slide 3 names dishcover.io once, lowercase, in the subtext, as the
+> MECHANISM the tip describes, and is written so it is still a good tip with the brand removed.
+> Slide 6 is the close: a real tip whose subtext runs to a third line carrying the ADDRESS
+> (`dishcover.io — every dish in NYC, searchable`) — lowercase, 500 weight, same size as the rest of
+> the subtext, no verb aimed at the reader, no "free", no "link in bio", and delete-safe (cut that
+> line and slide 6 is unchanged). Exactly two brand touches in the deck, no more. No title slide;
+> the full topic and the comment/save bait are the post caption. Ship the six PNGs as the
+> deliverable (durable My Files, ordered
+> in `slides.json`) and an MP4 at exactly 3.0s per slide with hard cuts as the preview. Full rules:
 > `vidfarm.cc/experimental/sticker-slideshow-tips.md`.