# Reading a topic

Every topic answers four questions: is it worth bidding on, which way is it
moving, how crowded is it, and what should you bid on.

## Is it worth bidding on

**`attention_index`** (0–100) is how hard a topic's videos are pulling. It is a
speed, not a rank: we measure how fast each video earns views per day, compare it
against comparable YouTube videos of the same format and the same age, and publish
the topic's median. Higher is faster.

The same number means the same thing in pet care as it does in DIY, and it moves
when the category moves — so you can track one topic across editions, and compare
topics across categories you have never bought in before.

| Score | Read it as |
|---|---|
| 95–100 | **Exceptional** — rare, and worth reallocating for |
| 80–94 | **Commanding** — the category's centre of gravity |
| 65–79 | **Strong** — dependable working inventory |
| Below 65 | **Proven** — real audience, narrower reach |

**There is no long tail in the response.** A topic only reaches an edition after it
has cleared the video-count, distinct-creator and category-share gates, and then
placed in the top ten by attention. Every row has a live audience and biddable terms
behind it; nothing is padded in to fill the list.

**`scale`** — `niche`, `moderate` or `broad` — is how much of the category's
attention the topic carries, against an even share of it. A `broad` topic holds
well over its share; a `niche` one well under. Deliberately not our sample size,
which would inflate as we watch more.

## Which way is it moving

**`momentum_index`** (0–100) is the same scale applied to acceleration. It reads
the topic's newest third of videos against the rest of its own back catalogue, so
it is movement *within* the topic rather than its standing against the category.
A topic can be commanding and slowing at once, and the two fields will say so.

**`trajectory`** is the band that index falls in:

| State | Meaning | What to do |
|---|---|---|
| `surging` | Accelerating hard, from a position that already holds real attention | Move budget now — this is the lead time you subscribed for |
| `building` | Accelerating, but not yet at commanding attention | Take a position early, while it is still cheap |
| `established` | Publishing and pulling at the pace it already was | Dependable base inventory, not a timing play |
| `softening` | The newest videos are under-performing the topic's own back catalogue | Harvest what is left, plan the exit |
| `single_channel` | Moving fast, but on one channel's output rather than an audience | Do not bid — one creator stopping ends it |

The thresholds are fixed, so `established` means a topic genuinely holding its
ground. A quiet category returns quiet numbers rather than being flattered into
looking like a busy one, and a category where everything is accelerating is allowed
to say so.

`single_channel` is the one guard sitting on top of the bands. The topic is moving,
but on one channel's output rather than an audience shifting, and reported as
`building` it would be lying to you. It is kept out of `highlights.fast_moving` for
the same reason.

## How crowded is it

**`brand_presence`** — `open`, `light`, `active`, `saturated` — is how much
branded content already sits in the topic. `open` means nobody is there yet.
It describes creative competition, not auction pressure, which we cannot see.

**`creator_spread`** — `concentrated`, `moderate`, `diverse` — is how many
different channels carry it. A `concentrated` topic can be lost when one creator
stops posting.

**`brands_present`** lists up to ten brands already appearing.

## What to bid on

**`terms`** are biddable search terms drawn from the topic's own videos.

**`evidence`** is three video URLs that carry the topic. In Google Ads these are
directly targetable placements; everywhere else they are proof of the trend.

**`iab`** gives both the taxonomy IDs and the human path, so it can be mapped to
whichever taxonomy your platform speaks.

**`audience_mood`** comes from what viewers actually write. `commercial_signal` is
the field to optimise against — `tone` is context, not a target. `top_questions`
is empty when viewers are not asking anything, rather than padded with remarks.

**`suitability`** rates the topic against [GARM's Brand Safety Floor +
Suitability Framework](https://www.isba.org.uk/system/files/media/documents/2022-01/GARM%20Brand%20Safety%20Floor%20Suitability%20Framework%2023%20Sept.pdf) — the vocabulary DSPs and verification vendors
already speak, so it maps onto controls you have rather than asking you to learn
ours.

- **`risk`** — `low`, `medium`, `high` or `floor`. The first three are all
  sellable, to advertisers of differing appetite. `floor` is not the top of that
  scale: it is content GARM places beyond any advertising support, which no risk
  appetite reaches. We publish the **worse** of two readings — how the videos
  look, and what viewers write underneath them — so a topic whose comments turn to
  alleged harm is rated down even when the footage itself is clean.
- **`categories`** — the GARM categories the topic trips, commonest first. The
  linked framework defines each one; we do not restate them here.
- **`risk_mix`** — the distribution across the topic's videos, for when you want to
  set your own cutoff rather than take ours.

GARM rates **treatment, not subject.** A vet explaining a procedure is `low`
however clinical it looks; the same footage dwelt on gratuitously is `high`.

## Reading the set as a whole

**`highlights`** shortlists topic names three ways:

- **`big`** — carrying well over an even share of the category's attention
- **`fast_moving`** — the three topics with the most momentum behind them
- **`uncontested`** — above the category's median attention, with brands not yet there

`uncontested` is usually where the lead time is: attention is already there and
the competition has not arrived.
