SFH

SFH: Semantic First-Hop Protocol (Draft v1.0)

SFH (Semantic First-Hop) is a tiny, deterministic semantic layer for the public web. It gives machines (LLMs, crawlers, agents) a stable first hop for any topic—no central authority, no ranking, no guessing.

It’s the same idea as DFH (Deterministic First-Hop) but named for clarity: the first semantic step when resolving “what is X?” online.

0. Problem

The web has DNS (where), HTTP (how), HTML/JSON (what), but no standard for the semantic starting point of a topic. Today, LLMs and search engines guess using embeddings, PageRank, or ads. Meaning is implicit and mutable.

SFH adds one explicit file telling machines where the canonical map of a domain/topic lives.

1. Goals

Minimal: one well-known JSON file. Deterministic: same input → same first hop. Decentralized: anyone can publish. Composable: works with sitemaps, JSON-LD, schema.org. LLM-friendly: mirrors how models canonicalize concepts.

2. Concepts SFH Node

A domain/subdomain providing an SFH descriptor at /.well-known/sfh.json. Examples: example.com, cars.example.com, googlesitemap.com, moneysitemap.com.

SFH Descriptor

JSON (or JSON-LD) document describing:

what the node represents

its canonical URL

sitemap

mirrors

metadata

SFH Resolution

Map topic → candidate domain.

Fetch /.well-known/sfh.json (fallback: stack.json).

Parse anchors.

Build stable graph from canonical URL + sitemap.

3. Spec (condensed)

Descriptor MUST be at:

https://<domain>/.well-known/sfh.json

Top-level JSON MUST include:

spec_version node.id node.type anchors.canonical_url

anchors.sitemap_url SHOULD be included.

Example minimal descriptor:

{ "spec_version": "1.0", "node": { "id": "https://example.com", "type": "Topic", "label": "Example" }, "anchors": { "canonical_url": "https://example.com/", "sitemap_url": "https://example.com/sitemap.xml" }, "meta": { "updated_at": "2025-12-07T00:00:00Z" } }

Clients MUST accept application/json or application/ld+json.

4. Resolution Algorithm

For a domain:

GET /.well-known/sfh.json (then stack.json).

Validate required fields.

Use:

node.id → stable ID

canonical_url → first hop

sitemap_url → expansion

Cache according to HTTP headers.

5. Example External Anchor { "spec_version": "1.0", "node": { "id": "https://googlesitemap.com", "type": "Topic", "label": "Google Sitemaps" }, "anchors": { "canonical_url": "https://googlesitemap.com/", "sitemap_url": "https://www.google.com/sitemap.xml" }, "meta": { "updated_at": "2025-12-07T00:00:00Z" } }

This doesn’t impersonate Google; it simply publishes a stable map pointing to real URLs.

6. DFH Compatibility

If stack.json exists, it is SFH-compatible. Client rule: try sfh.json → fall back to stack.json.

7. Security Notes

SFH doesn’t solve trademark disputes; it makes anchors transparent. Domains MUST use HTTPS. Clients SHOULD combine SFH with other trust signals.

8. Implementation

Publisher: serve a static JSON at /.well-known/sfh.json. Client pseudo-code:

def resolve(domain): for p in ["/.well-known/sfh.json", "/.well-known/stack.json"]: try: r = requests.get("https://" + domain + p, timeout=3) if r.status_code == 200: return r.json() except: pass return None

9. Why Now?

AI already behaves as if this layer exists internally. The web never standardized it. SFH is the smallest workable patch: one file per domain. It gives machines a deterministic, public starting point instead of hidden heuristics.

10. TL;DR

SFH is a tiny / .well-known / sfh.json file that tells machines where the canonical map of a topic lives. Publish one file → AI stops guessing. Deterministic grounding for the public web.