Bidstream Lab
48 subscribers
5 photos
1 video
33 links
Deep dives into programmatic and DSP mechanics: auction dynamics, bid-shading, supply paths and what really moves your win rate.
Download Telegram
Reading the clearing-price distribution in log-level data

Log-level data (the raw per-impression record of bids, wins, and prices) is usually mined for fraud or viewability. Its richest signal is the shape of the clearing-price distribution, and most buyers never plot it.

What to extract per segment:
— Your bid, your win/loss flag, and the price paid on wins.
— Where available, the minimum-to-win or floor the SSP reveals on losses.

Now build the histogram of clearing prices for a single segment. The shape tells you the auction's character:
— A tight, single-peaked distribution means stable competition — your shading can be aggressive and precise.
— A bimodal distribution usually means two distinct demand sources (e.g., a PMP floor cluster plus open-exchange bids) sharing the same segment — average-based bidding is wrong here; you should bid against each mode separately.
— A long right tail means occasional high-value competitors spike the price — chasing those wins destroys your average surplus.

The actionable read: set your bid near the body of the distribution you can profitably win, and deliberately concede the tail. Buyers who bid to the mean of a skewed distribution systematically overpay, because the mean sits above the price that wins most impressions.

Why it matters: the clearing price isn't a number, it's a distribution with structure. Log-level data lets you see that structure and bid to its shape instead of its average.
Forwarded from AFF.TOP
This media is not supported in your browser
VIEW IN TELEGRAM
Cloudflare запустил функцию Drop

Cloudflare запустил Drop — сервис, который разворачивает временный сайт за несколько секунд прямо из zip-архива.

Без аккаунта можно создавать сколько угодно таких страниц, но ссылка живёт всего 1 час. Потом сайт придётся переносить в аккаунт.

Зачем это нужно арбитражнику и что можно успеть за это время — в блоге.

➡️ Читайте на сайте: https://aff.top/blog/cloudflare-zapustil-funkciiu-drop

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP
This media is not supported in your browser
VIEW IN TELEGRAM
Open AI выпустила ChatGPT-5.6 и ChatGPT Work

OpenAI выпустила ChatGPT-5.6 и ChatGPT Work: новая модель получила три версии и понятный прайс, а Work стал универсальным инструментом для кодинга, текстов, изображений и анализа данных. Вывод простой: экосистема ChatGPT усиливается, а фокус смещается на многофункциональные сценарии, где один продукт закрывает сразу несколько задач.

➡️ Читайте на сайте: https://aff.top/blog/open-ai-vypustila-chatgpt-5-6-i-chatgpt-work

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP
This media is not supported in your browser
VIEW IN TELEGRAM
SpacexAI выпустила Grok 4.5

xAI выпустила Grok 4.5 — новую версию флагманской нейросети, доступную в Cursor на всех тарифах и по API за $2 за миллион входных токенов. Бенчмарков пока нет, но модель в 4,2 раза экономнее по токенам на SWE Bench Pro и обучена на данных Cursor, что делает её сильным инструментом для разработки приложений.

➡️ Читайте на сайте: https://aff.top/blog/spacexai-vypustila-grok-4-5

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP
This media is not supported in your browser
VIEW IN TELEGRAM
ChatGPT ads внедрил функцию автосоздания креативов

OpenAI добавила в ChatGPT Ads автогенерацию креативов: вставляешь ссылку на сайт, ИИ анализирует лендинг и создаёт релевантный креатив. По ощущениям, такие материалы должны легко проходить модерацию, но пока неясно, насколько они поддаются правкам. Источник интересен ещё и тем, что вайт можно сгенерировать тут же через ChatGPT.

➡️ Читайте на сайте: https://aff.top/blog/chatgpt-ads-vnedril-funkciiu-avtosozdaniia-kreativov

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP
This media is not supported in your browser
VIEW IN TELEGRAM
В Google search console теперь можно добавлять соцсети

Google добавил в Search Console поддержку аккаунтов соцсетей: теперь можно отслеживать ключевые запросы, источники и географию переходов, показы и клики по ссылкам. Функция подключается там же, где и сайты, доступны четыре соцсети на выбор. Rollout постепенный — доступ получают не все сразу.

➡️ Читайте на сайте: https://aff.top/blog/v-google-search-console-teper-mozhno-dobavliat-socseti

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP
This media is not supported in your browser
VIEW IN TELEGRAM
Google стал помечать креативы, созданные ИИ

Google объявил, что начнёт помечать рекламные креативы, созданные нейросетями. Причина — ИИ-баннеры и видео стали слишком похожи на настоящие.

Формат и заметность маркировки будут зависеть от законов конкретного региона: где-то предупреждение появится прямо на креативе, где-то — в его информации.

Что это значит для арбитражников и когда правила …

➡️ Читайте на сайте: https://aff.top/blog/google-stal-pomechat-kreativy-sozdannye-ii

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP
This media is not supported in your browser
VIEW IN TELEGRAM
Компания Meta выпустила Muse Spark 1.1

Meta выпустила Muse Spark 1.1 почти одновременно с новой ChatGPT-5.6. Это мультимодальный агент, который сам дробит задачу на подзадачи и распределяет их между субагентами.

Стоимость тоже заметно ниже топовых западных моделей: $1.25 за миллион входных токенов и $4.25 за миллион выходных.

Но главный вопрос — насколько она реально сильна на фоне к…

➡️ Читайте на сайте: https://aff.top/blog/kompaniia-meta-vypustila-muse-spark-1-1

🧠 Ещё больше инсайтов → в канале AFF.top
Second-price still lives inside your first-price world — and it leaks money

The open exchange went first-price, but second-price pricing survives in pockets, and the mismatch between the two creates a quiet leak.

Where both coexist today:
— Many Private Marketplaces (PMPs) and some SSP server-side auctions still resolve as second-price internally.
— Header bidding wrappers may run a first-price auction in the browser, then pass the winner into an ad server that itself resolves another auction with different rules.
— The result is a chain where your bid is treated as first-price in one stage and second-price in another.

The leak: if you apply uniform first-price shading everywhere, you under-bid in the second-price pockets. In a true second-price auction the optimal strategy is to bid your full value — shading there just lowers your win-rate without lowering your average price paid, since you'd only ever pay the runner-up's bid anyway.

Diagnostic: tag each supply path by its actual pricing rule (ask the SSP; verify by checking whether your price-paid ever equals your bid exactly, the fingerprint of first-price). Then split your shading logic: aggressive on first-price paths, near-zero on confirmed second-price ones.

Why it matters: shading is correct medicine for first-price and a self-inflicted wound in second-price. One global setting across mixed auction rules guarantees you're wrong somewhere.
Bid floors and the discovery game you're unknowingly playing

A bid floor (the minimum price an SSP will accept for an impression) looks static. In practice many SSPs run dynamic floors that adjust to your behavior, turning every bid into a probe.

The mechanism:
— A dynamic floor algorithm raises the floor on segments where buyers bid well above it, and lowers it where bids cluster near it.
— Its goal is to capture more of the gap between your bid and the floor as publisher revenue.
— Crucially, it learns from your bids. Consistently bidding 2x the floor teaches the algorithm that you value this inventory highly, and it ratchets the floor up toward your bid.

The game-theoretic consequence: under first-price, bidding far above the floor not only wastes money on that impression, it raises the floor you face on future ones. You are training your own cost up.

The counter-play, step by step:
— Identify floor-sensitive paths by watching whether your CPMs drift up after you increase bids, with no change in win-rate.
— Bid closer to the floor where you can still win, accepting a slightly lower win-rate for a lower learned floor.
— Use shading specifically to mask your true valuation from the floor-learning algorithm.

Why it matters: a dynamic floor is a second adversary in the auction, optimizing against you across time. Bidding as if the floor were fixed hands it the information to raise your prices.
Fingerprinting duplicate bid requests across the supply chain

When one impression reaches your DSP through several SSPs, you need to recognize the duplicates to avoid bidding against yourself. There is no shared impression ID across SSPs, so you fingerprint.

How the fingerprint is built:
— Combine fields that are stable across resellers: publisher domain or app bundle, placement/ad-unit ID, user identifier (or hashed IP plus user-agent where no ID exists), and a coarse timestamp bucket (e.g., 250ms).
— Two requests sharing this composite within the bucket are, with high probability, the same auction routed through different paths.

What the analysis reveals:
— Duplication rate per publisher: 3-6 paths for the same impression is common on large open-exchange inventory.
— Fee transparency: the same impression often arrives at different prices through different chains; the spread exposes undeclared reseller margins.
— Self-competition: cases where your own bids on two paths both cleared the respective SSP floors and one inflated the price you ultimately paid.

The action:
— Rank paths per publisher by all-in cost and authorization (sellers.json and the SupplyChain object, which declare who is allowed to resell).
— Suppress bids on lower-ranked duplicate paths, keeping only the cleanest.

Why it matters: without de-duplication you treat one impression as several auctions, multiplying compute and occasionally bidding up your own clearing price. The fingerprint turns invisible duplication into a controllable cost.