# Same-origin scanning missed docs links on 202 YC homepages

Canonical: https://mudpie.ai/agent-readiness/same-origin-scan-missed-yc-docs-links/
Breadcrumb: [Home](https://mudpie.ai/) / [Agent readiness](https://mudpie.ai/agent-readiness/) / [Same-origin scanning missed docs links on 202 YC homepages](https://mudpie.ai/agent-readiness/same-origin-scan-missed-yc-docs-links/)
Author: Ali Abouelatta (https://mudpie.ai/authors/ali-abouelatta/)
Published: 2026-09-19
Updated: 2026-09-19
Research type: original analysis
Method: Fresh guarded credential-free homepage GETs for all nine YC cohorts; record every eligible matching anchor without following reference links. Compare same-origin, any-origin and first-20 rules on each same response, with a conservative docs URL-pattern sensitivity subset and current primary-industry strata. Non-observed HTML signals remain unknown.

A same-origin filter missed more than half the YC homepages advertising a docs or developer link in this scan.

Across 1,515 freshly parsed homepages, 390 contained a docs/developer keyword match in an eligible link. Keep only links on the homepage’s exact origin and that count falls to 188. The other 202 companies advertised matching links only on other origins: 51.8% of the companies we detected.

The first draft of this analysis would have blamed company websites for a limitation in the collector. Checking the collection rule changed the story: it discarded links to docs subdomains and other hosts. That boundary was ours, not the companies’ choice to hide documentation.

So the follow-up question was straightforward: how much did that filter miss?

| Homepage link match | Same-origin only | Any origin | Companies recovered |
| --- | ---: | ---: | ---: |
| Docs/developer | 188 | 390 | 202 |
| Pricing | 374 | 381 | 7 |

Each number counts company rows among the same 1,515 responses. I applied both rules to each freshly downloaded page. Changes between the earlier collection and this one cannot explain the difference in this table.

Pricing barely moves. Docs more than doubles. The choice of origin boundary changes the apparent balance between two kinds of buyer information.

There is overlap, too. Of the 390 docs/developer matches, 141 companies linked only within their origin, 202 only elsewhere, and 47 in both places. A scanner that already detects a company can still discard additional reference links that its homepage advertises.

The individual pages make the problem obvious. [Riveter’s homepage](https://www.riveterhq.com/), listed in its [YC profile](https://www.ycombinator.com/companies/riveter), linked to `docs.riveterhq.com`. [Theseus’s homepage](https://www.theseus.us/), from the [Industrials roster entry](https://www.ycombinator.com/companies/theseus), linked to `docs.theseus.us`. Both links were present in the HTML. Neither survived a same-origin-only inventory.

The industry comparisons change accordingly. B2B goes from 148 to 312 docs/developer matches among 936 parsed homepages. Fintech goes from six to 21 among 103. Industrials goes from eight to 16 among 195. Those are current YC primary-industry classifications; I did not assign companies to categories from their names.

Pricing still differs by industry after including all origins: 303 of 936 B2B homepages matched, compared with seven of 195 Industrial homepages. That describes the links exposed to this collector. It gives me no basis to say why a founder chose a sales process.

The original 20-reference cap was a smaller problem here. Holding the original keyword and path rules fixed, lifting that cap recovered one additional docs/developer company and one pricing company. The origin filter accounted for the larger gap.

I also tightened the docs check. A broad “docs” match can include a Google Form. A narrower URL-pattern subset—documentation-style hosts or paths, excluding Google and YouTube—finds 350 companies across all origins versus 147 within the homepage origin. The gap survives the stricter definition.

For anyone building a buyer-research collector, this is a concrete design choice: inventory eligible advertised destinations before deciding which ones to fetch. Keep the request budget separate. A docs subdomain can be recorded as a useful next destination without granting a crawler permission to follow it.

> **Methods:** Fresh credential-free homepage GETs ran September 19, 2026, 16:14–16:20 UTC, with robots policy and concurrency three. All nine cohorts contributed 1,586 rows: 1,584 valid public hosts, one missing URL, and Brumby’s single-label host. We parsed 1,515 HTML responses; 54 rows had no homepage HTTP status and 17 returned non-2xx responses. Those 71 remain unknown for HTML signals. Matches use anchor labels/hrefs; destinations were not fetched. Query/hash and private or credential links are excluded from published URLs. Observations describe raw HTTP HTML, not rendered visibility or verified documentation contents. The earlier HTTP snapshot remains separate.

Sources: [YC company directory](https://www.ycombinator.com/companies), [Summer 2024 roster](https://www.ycombinator.com/companies?batch=Summer%202024), [Summer 2026 roster](https://www.ycombinator.com/companies?batch=Summer%202026). The [follow-up dataset](/research/yc-homepage-links-2026-09-19.json) contains all nine cohorts and the per-company link observations. The [methods package](/research/yc-homepage-methods/README.md) includes the parser, guarded collector and no-network recount, with historical replay limits stated.


## Author disclosure

I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.