aadharsh ✓

@oddhash

காலை வணக்கம் enjoyoor of technical systems @archetypevc prev econ + data @uchicago

1 129Followers

1 981Following

2 047Posts total

21.9KViews on collected posts

Against accounts of the same size

13 posts from the last 90 days, next to the under 10K follower range. reaches fewer people than peers, but engages them much harder.

Median views301this account2 126median for under 10K

Reach, %26.66%this account156.19%median for under 10K

Engagement, %2.10%this account1.63%median for under 10K

Metric	This account	Median for under 10K	Ratio
Median views per post	301	2 126	0.14×
Reach (views ÷ followers)	26.66%	156.19%	0.17×
Engagement rate	2.10%	1.63%	1.28×

Others in this range → Compare with another account → How these benchmarks are built →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

73526 Aug

18.3K

562

394

308

312

301

222

188

166

150

143

189

Last 13 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

1.22%26 Aug

0.47%

1.07%

1.27%

1.62%

1.60%

2.66%

2.70%

2.13%

2.41%

2.67%

2.10%

2.65%

Reactions — likes, reposts, replies and quotes — divided by views. Median for under 10K accounts is 1.63%.

What the audience does

Likes58.3%116 in total

Reposts4.5%9 in total

Replies10.6%21 in total

Quotes2.0%4 in total

Bookmarks24.6%49 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Latest posts

@oddhash oh youre a fan of windows xp? name every window 189 views · 4 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

As an aside, I sent 16 different bot identities at 11 sites, 3 times each. Two made up user-agent strings got served where GPTBot and ClaudeBot got a 403. A denylist has to explicitly name who or what it stops, so sometimes, being honest gets you blocked https://t.co/9NxpesALM3 143 views · 3 likes · 0 reposts · 0 replies 26 Aug 2026 · Open on X →

None of this is a knock on any of these sites; almost nothing is built around bots as a desired reader, because until recently, there was no reason to be. But as we see personal bots make up larger swathes of web traffic, hamstringing them hurts everyone https://t.co/6koUhaXkWD 150 views · 3 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

The delta view flips each missing infrastructure piece on one at a time so you can watch the bot route evolve, and see which verbs were actually blocked by which missing link or set of links and what could be done to unblock a site. https://t.co/90DF8d1JHF 166 views · 3 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

We can run head to head matchups: Stripe scores a 20 with no agent doors and nearly perfect retrieval, while more native sites score higher on net across an agent’s web lifecycle. Retrieval is largely a solved problem waiting for implementation. https://t.co/xDqsQmgkAs 188 views · 3 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

NYTimes scores 13 out of 100 on publishing for agents, but an agent can read today's paper anyways Both things can be true at once. The paper is reachable, but the agent pays 446x to reach it, and nobody at the Times chose that trade, nor do they benefit https://t.co/vyfvDRRMIu 222 views · 5 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

The imperative has gone from reading to acting with agents. Can sites hand agents tools, identity and payment instead of static content? Yeah! There are 10,000+ MCP servers with tools, there's a push to durably ID bots, and there's a neat WebMCP hackathon happening right now htt 301 views · 5 likes · 0 reposts · 3 replies 26 Aug 2026 · Open on X →

Reading a site is the easy part! The agent trace on Stripe ends pre-Action, so bots burn tokens to read content and can’t do much else without paying proxies and third parties to LARP as a human By adopting accessible protocols, pages allow bots to interact with data and tools 312 views · 4 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

Protocol solutions survive when they help publishers; https://t.co/gDuljobpWO, which powers the preview you see below, drives clicks Serving LLMs and the sheer quantity of bot traffic similarly can pay publishers back as a carrot, with SEO becoming opaque AEO/GEO as a stick 308 views · 4 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

The Semantic Web has been a decades long effort: 1995's Meta tags, 2008's RDFa, 2009's microdata, 2010's Open Graph, 2011's JSON-LD, etc. I pointed the lens at Wikipedia's entry on the Semantic Web. A bot finds none of the above and must infer structure from the prose it sees. h 394 views · 4 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

Reading the same page can be priced many ways based on how it’s formatted and served. Raw HTML costs 87.4k tokens, md is 1.9k, and header outlines are just 366 A naive read costs 239x an outline, and this stacks up if you read a page on a scheduled cadence or when it updates. ht 562 views · 5 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

Stripe’s homepage has 637 KB of HTML for 539 words. Humans read it for free, but a bot needs $0.26 to read the same words, because it wades 87,400 tokens of markup, if it even gets there. With this in mind, I built a lens that shows you any webpage the way machines see them http 18.3K views · 66 likes · 8 reposts · 8 replies 26 Aug 2026 · Open on X →

I had the chance to present https://t.co/6koUhaXkWD at @newinterfaces as an opinionated object expressing an otherwise invisible view, that of the ~60% of web traffic represented by bots and the monetary, resource, and accessibility costs they must pay to LARP as a human today 735 views · 7 likes · 1 reposts · 1 replies 26 Aug 2026 · Open on X →