{"blyg":"0.4","id":"3t4kfb63bsyqajphwz14ngz52t","kind":"thread","version":1,"at":"2026-11-05T23:35:24Z","note":"RFC-5 open for comment until 2026-12-03.","pinned":true,"origin":"https://blyg.blygger.org/","author":{"name":"Venkatesh Rao","url":"https://blyg.blygger.org/"},"content_md":"\n# RFC-5 — Blygger inside an organisation: one audience per origin, a boundary the publisher enforces\n\n**Request for comments · non-normative · OPEN FOR COMMENT until 2026-12-03 on\n[blyg.blygger.org](https://blyg.blygger.org/t/3t4kfb63bsyqajphwz14ngz52t/) (#66); without a blyg, comment on [blygger-spec#27](https://github.com/blygger/blygger-spec/issues/27) ·\nsession 59–60, 2026-11-04/05.** Drafted by Opus 5.5 (session 59) at Venkat Rao's request\nas research only: no code, no client change. Edited and ruled by Fable 5.1 (session 60),\nwhose rulings are section 10. Like a technical note, an RFC constrains nothing. When the\ncomment period closes it becomes a technical note, for people building enterprise\nproducts on Blygger (decision #90: the close is the promotion, and the first build is a\nrevision trigger). It is about the protocol and about what any client serving enterprise\nuse has to handle. An enterprise product is expected to be a client of its own, so it\ndescribes no particular client. Written against protocol 0.4 at its tenth revision, which\ncarries the sentences F1 and F5 add; no pre-1.0 version promises anything (#21).\n\nFive questions in this RFC needed Fable under the four-question test (F1–F5). Section 6\nstates each item, with the test as it was applied, and section 10 rules it.\n\n## Summary\n\n**The protocol has no fundamental blocker for enterprise use.** Nothing on the wire\nassumes public DNS or the public internet. An origin is a URL string. Resolution and\nverification are fetches. Every fetch failure already degrades safely: a poll failure\ndiscards nothing (§12.3), and an unreachable source is a failed mention (§15.4). The\n`blyg:` namespace URI is an opaque token that is never fetched (§7).\n\nThree things need care, and only one of them is a missing capability:\n\n1. **One audience per origin.** A whole origin behind SSO or a private network works\n   today. Different readers seeing different items *within one origin* is L3, which #7\n   defers. Enterprises get audiences by running one origin per audience: the N-origins\n   group shape, `tn-3` shape A. This is the missing capability (F1, F4).\n2. **References leak across a boundary.** A quote bakes the bytes. Every reference\n   carries an origin string, and `cited` can carry an excerpt. A sent mention POSTs the\n   internal URL to the outside receiver. The protocol permits all of this. An enterprise\n   client has to refuse it at publish time (F3).\n3. **Retention law runs against two promises** the medium makes: published documents\n   answer 200 forever, and pins are irrevocable (F5).\n\nEverything else is client work, and section 4 lists it: SSO and roles, reader\ncredentials, egress policy, audit and retention, hosting and residency, AI controls, and\nthe publish-time boundary.\n\n## 1. The question\n\nVenkat asked whether anything fundamental stops Blygger being used as an enterprise\ntool, deployed inside an organisation's boundary with the usual enterprise features:\nSSO, an internal DNS scope, private networks, audit, retention.\n\nThe answer matters beyond enterprises. The protocol's claim is that anything able to\nserve files can host a blyg (§1). An intranet file server can serve files. If the claim\nheld only on the public internet, it would be narrower than it says.\n\n`plans/private-items-proposal.md` §8 (same session) recorded Venkat's answer that\n\"enterprise deployments are under discussion\" and are the concrete case likely to open\nL3. This RFC is the enterprise half of that thread. It does not repeat the\nprivate-items design. It asks what an enterprise needs *before* L3, and whether L3 is on\nits critical path.\n\n## 2. Deployment shapes\n\n| Shape | What it is | Audience per origin | Protocol status |\n|---|---|---|---|\n| **A. Fully internal** | Every blyg sits on an internal DNS zone (for example under `.internal`, which ICANN reserved for private use in 2024) or behind a gateway. Nothing is reachable from outside. | uniform (employees) | conformant today, given F1 |\n| **B. Internal plus selectively public** | Some origins internal, some public: a public engineering blog next to internal team blygs. | uniform per origin; different origins, different audiences | conformant today; the boundary between origins is F3 |\n| **C. Mixed federation** | Internal blygs subscribe to and quote public blygs, and the reverse is attempted. | as B | inbound (public → internal) is safe; outbound is F3 |\n| **D. Vendor multi-tenant hosting** | A vendor hosts many organisations' blygs, each tenant one or more origins. | uniform per origin | conformant today: #14, #36, #61, #84 make N origins on one host safe |\n| **E. Per-item audiences in one origin** | One team blyg where some items are for the team, some for the company, some public. | **mixed** | **L3** (#7). Not available. |\n\nShapes A–D need no new protocol. Shape E is the only one that does, and the\nrecommendation in section 3.3 is to build A–D first and treat E as the measured need\nthat opens L3 (F4).\n\n**Shape B is \"promotion\" by publishing anew.** Withdrawal is the only exit from public\n(#8), and an origin's audience never changes (F1). So an internal item does not *become*\npublic. Its author publishes a new item in the public origin. Whether that new item may\nname its internal ancestor is F3: `forked_from` or a quote would carry the internal\norigin string outward.\n\n## 3. Protocol analysis: is anything blocking?\n\nEach claim below is checked against the 0.4 text, not assumed.\n\n### 3.1 Nothing assumes public DNS\n\n| Construct | What it needs | Internal origin? |\n|---|---|---|\n| Origin, identity origin (§2, §12.2, #14, #17) | an absolute base URL; identity is the final fetch URL minus its last segment | **Fine.** No rule requires a publicly resolvable name. §12.1 step 1 expects HTTPS; an internal CA is a transport matter. |\n| Reference shape (§5.9) | `{origin, id, version}` as strings | **Fine** as data. A reader outside cannot check staleness, which §12.3 already treats as an ordinary failure. |\n| `content_html` absolute URLs (§5.2, #53) | every URL in the content absolute | **Fine** inside. A baked copy taken outside carries internal hostnames with it (section 3.2). |\n| Namespace URI `https://blygger.org/ns/0.1` (§7) | \"an opaque wire token\" | **Fine.** Never fetched, so an air-gapped reader needs no access to blygger.org. |\n| `generator_url` (§6.1) | informative, MUST NOT gate | **Fine.** |\n| Resolution (§12.1) | at most six fetches from the reader's network position | **Fine** for any reader that can reach the origin. |\n| Verification (§15.4) | the receiver fetches the source from its asserted origin | **Fine** when the receiver can reach the source. Otherwise *failed*, which is honest (F2). |\n| Webmention endpoint (§15.1) | optional at every level | **Fine.** A static internal blyg omits it (§15.7). |\n| CORS (§4) | `Access-Control-Allow-Origin: *` is a SHOULD | **Fine.** A gated origin answers credentialed CORS instead; the SHOULD bends for a stated reason. |\n| Census (§3.2) | \"the manifest is a file the protocol requires to be public\" | **Wording only.** The census simply never sees internal blygs. \"Public\" here means the same bytes to every reader who can reach the origin (F1, ruled in section 10; §4 now says so). |\n\n**One hazard to rule out: split-horizon DNS for a blyg origin.** If one name resolves\nto an internal server inside the network and a different server outside, then \"the item\nat that origin\" depends on where the reader stands. That breaks §12.2's guarantee that\ntwo readers given the same URL arrive at the same identity, and it reproduces per-reader\ncontent under one origin, which is shape E by the back door. Internal origins should use\nnames that do not exist publicly. Section 8 lists this as a rejected design.\n\n### 3.2 References across the boundary\n\nEvery cross-origin construct carries the target's origin string, and two of them carry\nwords:\n\n- **A quote bakes bytes.** §10.2 bakes the publisher's local snapshot into\n  `content_html`. §10.4 says the snapshot survives anything the source does, and §14\n  says so plainly (\"anything published can be quoted into a thread whose snapshot its\n  publisher cannot recall\"). A quote of an internal item into a public blyg copies the\n  internal text, and its internal absolute links (#53), out of the boundary for good.\n- **A `cited` object** (§5.9) may carry an `excerpt` and a `url`, so even a bare `stub_of`\n  or `forked_from` can carry words.\n- **Origin strings alone leak** internal hostnames, mount paths and item ids. That is\n  true of `stub_of`, `forked_from`, `links[]` (§10.3a), `transclusions[]` and\n  `generated[].sources[]` (§5.7).\n- **A sent mention leaks in the other direction.** §15.2 says a publisher SHOULD send a\n  mention for every remote reference. From an internal blyg that quotes a public one, the\n  POST hands the public receiver the internal page URL as `source`. The receiver then\n  tries to fetch it: the fetch fails (good), but it has learned the hostname and the\n  fact of the quote (bad).\n- **Generation is a quote by model.** A TK span drawing on an internal item, published\n  publicly, carries the internal origin in `generated[].sources[]` and paraphrased\n  content in the text. The 0.4 rationale for the `source` relation, that \"paraphrase by\n  model must not be the one way to use a stranger's words without telling them\", cuts\n  the same way here.\n\nNone of this is a protocol defect. The protocol is a public-speech medium, and quoting is\nits primitive (#12). It does mean that **the boundary is enforced by the publisher at\npublish time, or not at all.** No reader-side or wire-side mechanism can recall a baked\nquote. That is the same conclusion `private-items-proposal.md` §3.4 reached for private\nitems, generalised from \"private versus public\" to \"narrower versus wider audience\".\n\nThe rule a boundary-aware client applies:\n\n> **A version published at an origin with audience W may carry no reference (of any\n> kind, `cited` included) to an origin whose audience is narrower than W, and sends no\n> mention to a receiver outside the audience of its own origin.**\n\nThe audiences are client configuration: a list of origin prefixes per audience. Both\nhalves are within what the spec already permits. Refusing to quote is a publish error\nthe client chooses. Not sending a mention is a SHOULD the client declines for a stated\nreason. The spec says nothing more (F3, ruled in section 10): §14 already carries the\nsentence that matters, \"anything published can be quoted into a thread whose snapshot its\npublisher cannot recall\", and the rule above is client guidance in this note.\n\n### 3.3 \"Readable only by members\": static, L1/L2, or L3?\n\nThe distinction that settles this is **where the audience boundary falls**.\n\n- **Origin-uniform audience.** The whole surface (manifest, feed, index, item documents,\n  pins, media, pages) sits behind one gate: a private network, a VPN, an identity-aware\n  proxy, Cloudflare Access. Everything inside the gate is plain static files, every\n  reader who passes the gate sees the same bytes, and §4's 404/200 rules hold for every\n  such reader. The gate is transport, like TLS. This is conformant at L1 and L2 and\n  needs nothing new (F1, confirmed in section 10). §4 now says it in one sentence: \"What\n  an origin serves does not depend on who asks: every reader who can reach it gets the\n  same bytes, and reachability is the deployment's.\"\n- **Per-item or per-reader audience in one origin.** Two readers asking for the same\n  `items/{id}.json` get different answers (200 versus 401/404, or full versus redacted\n  content). That is the gate `private-items-proposal.md` §5 analyses as C1–C3. Section\n  10 classifies them: C2 is refused, C3 is where an L3 round starts, and C1 is client\n  territory under one invariant. The proposal's \"not an option\", a gated page over an\n  open JSON, applies here too.\n\nSo an enterprise gets members-only reading **today** by running one origin per audience\nbehind a gate. A company-wide blyg, a team blyg and a public blyg are three origins, on\none host if it likes (`tn-3` shape A). The cost is the one `tn-3` names: N origins, N\narchives, N sets of pins.\n\n**What this does not give** is an item that is team-only today and company-wide\ntomorrow: the publish-anew path (section 2). That is where L3, or a studio-side\ncross-origin \"promote\" with honest lineage, would earn its cost. Section 10 rules it the\ntrigger for L3 (F4), measured when a deployment hits that cost with a named case.\n\n### 3.4 Reader side\n\n- **Readers must reach the origin.** A subscribing studio behind the gate needs network\n  reach and, when the gate is identity-aware, a machine credential (a service token,\n  mTLS, workload identity) for each gated origin. That credential is transport. The\n  protocol has no opinion on request headers, and §12.3's polling rules are unchanged.\n  It is client work (section 4.2).\n- **Failures are already safe.** A reader that loses access keeps its stored state and\n  backs off (§12.3). It never treats 401 as withdrawal, because only an endcap is\n  withdrawal (§13.4).\n- **Sanitize-at-render is unchanged** (§14). The internal-content risk is egress, not\n  rendering.\n- **Retention follows the origin** (§13.4). A records system that keeps withdrawn\n  content because the law requires it acts outside the protocol, and keeps the archive\n  off every surface the spec governs (F5, ruled in section 10). §14 now says: \"Law may\n  compel a publisher to remove what it promised to serve, or a reader to retain what an\n  origin withdrew. Both are acts outside the protocol: a compelled removal is a 404 with\n  no marker, and a compelled retention never reaches any surface this document governs.\"\n\n### 3.5 Identity and SSO\n\nInvariant 4 and #11 say that identity is never in the protocol, and that the only\nauthenticated entity is the origin. This **helps** enterprise use rather than blocking\nit. The protocol takes no position on how an owner, an editor or a reader proves who\nthey are, so SSO is entirely the client's:\n\n- **Write side.** §14's last bullet and §16.6 put authoring auth out of scope, and #31\n  rules out a normative write surface. A client's authoring auth is its own; an IdP\n  session can be its root credential. The pattern #31 recommends (scoped,\n  owner-revocable tokens for tools, a root credential no tool holds) carries over\n  unchanged.\n- **Bylines.** An SSO-backed studio can assert verified employee bylines (`author.name`,\n  `author.url` pointing to an internal directory profile). #35's `rel=\"me\"` proof works\n  inside the network exactly as outside. RFC-1's `ids` and address book apply unchanged.\n  Bylines are still the origin's claim (§5.5), which inside one organisation is the\n  right trust model.\n- **Multi-author.** The invariant is single *publisher* per origin, not single author\n  (#11). A team blyg with many SSO users is `tn-3` shape B, the masthead. #36's test,\n  \"who can withdraw\", decides between shape A and shape B per team.\n- **Reader identity.** This is where SSO touches the protocol, and only in shape E,\n  where a group needs named readers. That is F4, and #11 calls it out (\"keys imply\n  identity\").\n\n### 3.6 Verdict\n\n| Item | Kind | Where |\n|---|---|---|\n| Public DNS or internet assumed anywhere | **none found** | 3.1 |\n| Whole-origin gating (SSO proxy, private network) | interpretation; one §4 sentence | **F1**, ruled, §10 |\n| Verification when the source is unreachable | interpretation, small | **F2**, ruled, §10 |\n| Cross-boundary quoting and reference leakage | client rule, no spec text | **F3**, ruled, §10 |\n| Per-item audiences in one origin | **missing capability (L3)** | **F4**, ruled, §10 |\n| Compelled deletion; compliance archives of others' content | out of protocol; one §14 sentence | **F5**, ruled, §10 |\n| SSO, roles, audit, residency, egress, AI controls, DLP | client and product work | 4 |\n\n## 4. Concerns for clients serving enterprise use\n\nThe protocol governs only the published surface (#3). Everything below is the client's\nbusiness. It is listed so that implementers see the whole job, and so that two enterprise\nclients can make compatible choices where they touch.\n\n### 4.1 Authoring identity: SSO and roles\n\n- The organisation's IdP (OIDC, or SAML through a broker) is the root credential. No\n  tool holds it; tools get scoped, revocable tokens (the pattern of #31).\n- Many people may author in one origin. The protocol needs one *publisher* per origin,\n  not one author (#11), so a team blyg is `tn-3` shape B with bylines, or shape A with\n  one origin per member. #36's test, \"who can withdraw\", decides between them.\n- IdP groups map onto the client's roles. Deprovisioning (SCIM or equivalent) revokes a\n  departing user's sessions and tokens.\n- Verified bylines: `author.url` can point at the internal directory profile, and #35's\n  `rel=\"me\"` proof works inside the network as outside.\n\n### 4.2 Reading gated origins\n\n- **Gate the whole surface** (manifest, feed, index, item documents, pins, media, pages)\n  with one gate, so that every reader past it gets the same bytes (F1). Gating only the\n  HTML leaves the full text in the JSON.\n- **A subscribing client needs a machine credential** per gated origin (service token,\n  mTLS, workload identity). Attach it only on an exact origin-prefix match, and never\n  follow a redirect with it to another origin: §12.2 allows cross-host redirects, and a\n  credential that follows one leaks.\n- **Caches** must be keyed so that nothing cached for one audience is served to another.\n\n### 4.3 Network reach and egress\n\n- **Internal names.** A client fetches manifests, feeds and documents (§12), and a\n  receiver fetches mention sources (§15.4). Inside the network those names often exist\n  only in internal DNS, and a resolver that consults public DNS will not find them.\n- **The probe risk.** §14 asks receivers to refuse private-network addresses where they\n  could be used to probe them. The Webmention endpoint is the dangerous one: anyone who\n  can POST to it chooses a URL for the receiver to fetch. An internal deployment needs an\n  allowlist (host suffixes or origins, plus address ranges), not a switch that turns the\n  check off, and the endpoint should fetch allowlisted sources only for claims that\n  arrived from inside (F2).\n- **Air gaps.** In an air-gapped deployment, no feature may reach a public host: CDN\n  scripts, update checks, embeds, hosted AI providers.\n- **Rebinding.** A resolve-then-fetch check does not pin the connection. Enforce egress\n  at the network boundary as well.\n\n### 4.4 Audit, retention and legal hold\n\n- **Audit.** An append-only, exportable log of sign-ins, token grants and revocations,\n  publish, withdraw and pin, visibility and settings changes, subscriptions, outbound\n  mentions, and AI calls (model and provider; not prompt text by default).\n- **Legal hold** mostly comes free on the published side: withdraw-not-delete (#8) keeps\n  every published version, and pins keep their bytes. The gaps are unpublished material\n  (drafts, private items, media), which a hold must turn from deletes into soft deletes.\n- **Retention schedules and compelled deletion** run the other way. A records policy, a\n  court order or an erasure request can require removing a published item that §4\n  promised would answer 200 forever and a pin promised to keep (§8). The removal is an\n  act outside the protocol: the document answers 404, with no status code, marker or\n  reason (F5, ruled in section 10; the §14 sentence is quoted in section 3.4).\n- **Archives of external content.** Keeping other origins' withdrawn content because the\n  law requires it is outside the protocol too. A client that is both reader and archive\n  keeps the archive off every surface the spec governs: no presentation, no re-serving,\n  no quote from it (§13.4, F5).\n\n### 4.5 Hosting and data residency\n\nThe protocol is host-neutral: any server that can serve the files conforms (§4), so the\npublished surface can sit on an internal file server behind any gate. It loses only the\nWebmention endpoint, which is optional (§15.7). Where the authoring side runs, and where\nits data lives, is the client's design. Residency is a property of that design and of\nits hosts.\n\n### 4.6 AI provider controls\n\nAI is never in the protocol (#5). An enterprise client needs:\n\n- an operator-controlled list of allowed models and providers, with no free-entry way\n  around it;\n- the ability to point at an internal gateway;\n- AI calls inside the egress policy;\n- the boundary rule (section 3.2) applied to generation sources, since a TK source from\n  a narrower audience is a cross-boundary reference.\n\n### 4.7 The publish-time boundary (DLP)\n\nThis is F3, and the most important enterprise control. Section 10 rules that it is\nclient policy with no spec text: declining a SHOULD for a stated reason is what SHOULD\nmeans, and §14 already warns that anything published can be quoted out of its\npublisher's reach. Revisit only if two boundary-building clients disagree.\n\n- **Audiences are configuration:** named sets of origin prefixes.\n- **Publish block.** A version may carry no reference (transclusions, `stub_of`,\n  `forked_from`, `links[]`, `generated[].sources[]`, `cited`) to an origin narrower than\n  its own. Say which references block, and offer to remove them. This is the same block\n  `plans/private-items-proposal.md` §3.4 describes for private items, keyed on audience.\n- **Mentions.** Send none to a receiver outside the origin's own audience, and show the\n  author why the §15.2 SHOULD was declined.\n- **Content scan.** Optionally, check `content_html` for absolute URLs under narrower\n  origins (#53 makes them easy to find): a plain link leaks a hostname.\n- **Egress as backstop.** A wider-audience deployment that cannot fetch a narrower origin\n  cannot quote it either.\n\n### 4.8 Multi-tenant hosting\n\nN origins on one host are safe on the wire: #61 stops one tenant's documents verifying in\nanother's name, #84 makes target membership exact, and #87 fixes the manifest name at\nevery identity. Each tenant is the publisher of its origins (#11), not the vendor.\nIsolation of data, egress policy, audit and IdP per tenant is the vendor's design.\n\n## 5. Wire impact\n\n**None.** F1 and F5 are each one sentence in the living 0.4 text that readers already\nhandle; F2 and F3 add nothing; F4's shape E stays L3. Shapes A–D use only existing\nconstructs, and the boundary rule is a client publish error and a declined SHOULD.\n\n## 6. The five items and their tests\n\nThe four questions are those in `CLAUDE.md` → *Model routing*: (1) readers must change;\n(2) reader-visible addition; (3) narrows, widens or reinterprets a locked decision;\n(4) a locked decision already settles it. The items are kept as the record of the test\nas the draft applied it; each *Proposal* now opens with the ruling.\n\n**F1. Audience is a property of the origin.**\n\n- *Proposal:* Ruled, section 10: confirmed, with one sentence in §4. The draft: the\n  spec governs the bytes an origin serves to whoever can reach it.\n  Reachability, including an identity-aware gate across the *whole* surface, is\n  deployment. A gated origin conforms at L1 and L2, provided that every reader who\n  passes the gate gets the same bytes. Per-reader difference within an origin stays L3.\n  §3.2's \"required to be public\" reads as \"served to its audience\". A blyg origin MUST\n  NOT be split-horizon.\n- *Test:* Q1 no. Q2 no. **Q3 yes**: it reads #1 and #7 (\"privileged-group access is\n  deferred to L3\") as being about per-item audiences, not whole-origin ones, and that is\n  an interpretation Opus should not make alone. Q4: #3 and #14 point the same way, but\n  do not settle it.\n- *Touches:* #1, #3, #7, #14, #17; §3, §3.2, §4 (200 forever, CORS), §12.2, §16.7.\n\n**F2. Verification across a network boundary.**\n\n- *Proposal:* Ruled, section 10: confirmed, no text. The draft: a mention whose source\n  the receiver cannot fetch is *failed*, with no new\n  state. §14's \"refusal to fetch private-network addresses where the deployment could be\n  used to probe them\" is met by a deployment-scoped allowlist, because internal fetches\n  that the deployment intends are not probes.\n- *Test:* Q1 no. Q2 no. Q3 *maybe*: it reads §14's private-network SHOULD as\n  configurable. Q4: #28 and #61 settle the mechanics, but not §14's wording. Small.\n- *Touches:* #13, #28, #61, #84; §14, §15.2, §15.4.\n\n**F3. References that cross a boundary.**\n\n- *Proposal:* Ruled, section 10: confirmed as client policy, no text. The draft: client\n  policy in this RFC (section 3.2's rule), with **no** spec text yet.\n  Revisit only if a second client builds a boundary and the two disagree.\n- *Test:* Q1 no and Q2 no while it stays client policy. Q3 *maybe*: declining §15.2's\n  SHOULD wholesale for a class of receivers, and a publish error the spec does not list\n  in §10.2 step 3. Q4: #3 lets the client refuse anything at publish. If Fable wants a\n  sentence in §14 (\"a quote is an export\"), that is spec text.\n- *Touches:* #9, #12, #26, #53; §5.6, §5.7, §5.9 `cited`, §10.2, §10.3a, §14, §15.2;\n  `private-items-proposal.md` §3.4.\n\n**F4. Group visibility is L3, and is enterprise its trigger?**\n\n- *Proposal:* Ruled, section 10: confirmed, with C1–C3 classified. The draft: enterprise\n  does not need L3 to start (section 3.3). Shape E is the trigger,\n  measured when a deployment hits the publish-anew cost in practice. Until then, group\n  tokens wait, as `private-items-proposal.md` §8.3 says.\n- *Test:* any form of shape E is **Q1, Q2 and Q3 yes** (C2), or Q3 yes (C1).\n- *Touches:* #1, #7, #11, #35, #43; §3, §4, §16.7; `private-items-proposal.md` §5, §8.3;\n  backlog L3 row.\n\n**F5. Compelled removal and compliance retention.**\n\n- *Proposal:* Ruled, section 10: a sentence in §14, no marker. The draft: say in §14\n  that a legally compelled removal is an out-of-protocol act. The\n  origin breaks a promise and readers see a 404, which §12.3 already tolerates. Add no\n  status code or marker. For the reader side, ask whether a non-presenting compliance\n  archive is \"hoarding\" under §13.4.\n- *Test:* a §14 sentence is Q1 no, Q2 no, **Q3 yes** (it qualifies #8's \"200 forever\"\n  and the irrevocable pin). A marker such as 410 with a reason would be Q1 and Q2 yes.\n- *Touches:* #8; §4, §8, §9, §13.4, §14.\n\n## 7. Open questions for comment\n\nFor implementers building enterprise products on Blygger. F1–F5 are ruled (section 10);\nthese questions are about what implementers need, and where an answer bears on a ruling\nit says which.\n\n1. **Is one origin per audience enough?** Name the case where publishing anew across\n   audiences fails you. That case is what opens L3 (F4).\n2. **Reader credentials.** How should a subscribing client authenticate to a gated\n   origin: service tokens, mTLS, workload identity, something else? Is a shared\n   convention worth an RFC, or is it each product's business?\n3. **Outbound mentions from inside.** Should an internal blyg ever notify a public one\n   that it quoted it? If so, under what explicit act?\n4. **Promotion.** When an internal item is republished publicly, should the public item\n   name its internal ancestor (leaking the origin), carry a public lineage note, or\n   carry nothing?\n5. **Retention.** Which do you need: compelled deletion of published items, legal hold\n   over drafts, a compliance archive of external content, or all three? Does the §14\n   sentence (F5) cover what you need to build, and if not, what else do you need — legal\n   hold on drafts, archives, something the sentence does not reach?\n6. **Hosting.** Where must the authoring side run, and what residency do your buyers\n   need?\n7. **Audit.** Which events, which format, which sink?\n8. **Teams.** For your users, is a team blyg shape A (members' own origins) or shape B\n   (one masthead)? Who withdraws?\n9. **Should the boundary rule (section 3.2) ever become spec text?** It is client policy\n   today (F3); name the disagreement between two clients that would change that.\n\n## 8. Rejected designs\n\nRecorded because each will be proposed.\n\n1. **Split-horizon blyg origins.** One name, two servers, two contents. This breaks\n   §12.2's determinism and is per-reader content under one origin (section 3.1).\n2. **Gating the HTML pages only.** The JSON carries the full text (private-items\n   proposal §5, \"not an option\").\n3. **A 401 or 403 on chosen item documents within an otherwise readable origin.** This\n   is C2. §4 has no third answer, and §15.4 cannot verify against it. It is L3.\n4. **Turning private-network refusal off wholesale for an internal deployment.** It\n   works, and it makes the Webmention endpoint an internal fetch proxy for anyone who can\n   reach it. Use an allowlist (section 4.3).\n5. **An enterprise \"audience\" member in the manifest or item documents.** Readers cannot\n   act on it without identity, which is #11's line. Audience is the client's\n   configuration plus the gate.\n6. **A removal marker (a 410 with a reason, a tombstone).** A removal marker is the\n   tombstone #8 refused, and a delete signal invites delete requests. A compelled removal\n   is a 404 with no marker, which readers already tolerate (§12.3; F5, section 10).\n\n## 9. Relation to other documents\n\n- `plans/private-items-proposal.md`: the owner-only level and its publish block. This\n  RFC generalises that block to audiences (section 4.7) and supplies F4's trigger\n  framing for its §8.3.\n- `docs/notes/tn-3-groups-and-aggregation.md`: shapes A and B are the enterprise team\n  shapes.\n- `docs/notes/tn-6-mentions-and-identifier-schemes.md` (formerly RFC-1): bylines, `ids`\n  and agent members under SSO (section 3.5).\n- `docs/notes/tn-7-comments-sections.md` (formerly RFC-2): hosted commenters, which\n  inside an organisation are SSO users commenting through the house's comments origin.\n\n## 10. Rulings (Fable 5.1, session 60, 2026-11-05)\n\nThe draft closed with five items for Fable. The rulings follow, each with the principle\nit rests on. All five confirm the draft's direction; F4 adds a classification the draft\ndid not make, and F5 settles the reader side the draft left as a question.\n\n**F1. Audience is a property of the origin — confirmed, with spec text.** Decision #1\nmakes the protocol a static file contract, and #14 makes an origin any absolute URL;\nnothing in the text requires that an origin be publicly reachable. \"Public\" in #7 and in\n§3.2 means the same bytes to every reader who can reach the origin. Reachability is\ndeployment, like TLS. Per-reader difference within one origin is L3 (#7), and a\nsplit-horizon origin is that difference by the back door, because it breaks §12.2's\ndeterminism. The tenth revision of 0.4 adds the sentence to §4: \"What an origin serves\ndoes not depend on who asks: every reader who can reach it gets the same bytes, and\nreachability is the deployment's.\"\n\n**F2. Verification across a network boundary — confirmed, no text.** §14's clause is\nalready conditional: it asks for refusal \"where the deployment could be used to probe\nthem\". A fetch the deployment intends, under an allowlist, is not a probe. A source the\nreceiver cannot reach is *failed* (§15.4), which is honest: the receiver could not verify\nit.\n\n**F3. References that cross a boundary — confirmed as client policy, no text.** §14\nalready says the sentence the draft wanted: \"anything published can be quoted into a\nthread whose snapshot its publisher cannot recall\". The boundary rule (section 3.2: no\nreference of any kind, `cited` included, to a narrower origin; no mention to a receiver\noutside the origin's audience) lives in this note as client guidance. Declining a SHOULD\nfor a stated reason is what SHOULD means, so a client that withholds §15.2's mentions\nacross the boundary conforms. Revisit only if two boundary-building clients disagree.\n\n**F4. Group visibility is L3 — confirmed, and the three shapes classified.** Enterprise\ndoes not need L3 to start. Shape E is the trigger, measured when a deployment has hit the\npublish-anew cost with a named case. Of the private-items proposal's three gate shapes:\n\n- **C2** (a 401 or 403 on chosen documents, under any gate descriptor) is refused. It is\n  not static, and it is L3 by the back door.\n- **C3** (encrypted static content) is the shape the L3 round starts from when the\n  trigger fires.\n- **C1** (a redacted wire, with a fuller rendition served off the mount to signed-in\n  readers) is **client territory under #3, not L3**, under one invariant: nothing on the\n  protocol surface may name, link to or depend on the fuller rendition, and the public\n  document is complete in itself. The redaction is the author's own visible edit (#78).\n  That is a paywalled newsletter beside a blyg, which the protocol has never forbidden.\n\n**F5. Compelled removal and compliance retention — a sentence in §14, no marker.** \"200\nforever\" and the irrevocable pin are the publisher's promises. Law may force the\npublisher to break them, and the protocol does not legislate the law. Readers already\ntolerate a 404 (§12.3). There is no status code and no reason: a removal marker is the\ntombstone #8 refused, and a delete signal invites delete requests (section 8, item 6).\nOn the reader side, §13.4 governs what a conformant reader presents and re-serves. A\nrecords system that retains under law is outside the protocol in the same way, and a\nclient that is both reader and archive keeps the archive off every surface the spec\ngoverns. The tenth revision of 0.4 adds to §14: \"Law may compel a publisher to remove\nwhat it promised to serve, or a reader to retain what an origin withdrew. Both are acts\noutside the protocol: a compelled removal is a 404 with no marker, and a compelled\nretention never reaches any surface this document governs.\"\n\n**What happens next.** Published for comment under #66 on 2026-11-05; the period closes\non 2026-12-03 (four weeks). Decision #93 records the five rulings, with this section as\ntheir reasoning (`v0.4-plan.md` §18). Comments are stubs on the item, and\ncounter-proposals are forks of the pin.\n\n\n\n*Source: [`docs/rfcs/rfc-5-enterprise-deployment.md`](https://github.com/blygger/blygger-spec/blob/main/docs/rfcs/rfc-5-enterprise-deployment.md) in blygger-spec, which is where revisions are made; each one is published here as a new version. To comment, respond to this item from your own blyg: a stub is a comment, a partial quote names the passage, and a fork of the pinned version is a counter-proposal.*\n","content_html":"<div class=\"blyg-tk-gen\"><h1>RFC-5 — Blygger inside an organisation: one audience per origin, a boundary the publisher enforces</h1>\n<p><strong>Request for comments · non-normative · OPEN FOR COMMENT until 2026-12-03 on\n<a href=\"https://blyg.blygger.org/t/3t4kfb63bsyqajphwz14ngz52t/\">blyg.blygger.org</a> (#66); without a blyg, comment on <a href=\"https://github.com/blygger/blygger-spec/issues/27\">blygger-spec#27</a> ·\nsession 59–60, 2026-11-04/05.</strong> Drafted by Opus 5.5 (session 59) at Venkat Rao's request\nas research only: no code, no client change. Edited and ruled by Fable 5.1 (session 60),\nwhose rulings are section 10. Like a technical note, an RFC constrains nothing. When the\ncomment period closes it becomes a technical note, for people building enterprise\nproducts on Blygger (decision #90: the close is the promotion, and the first build is a\nrevision trigger). It is about the protocol and about what any client serving enterprise\nuse has to handle. An enterprise product is expected to be a client of its own, so it\ndescribes no particular client. Written against protocol 0.4 at its tenth revision, which\ncarries the sentences F1 and F5 add; no pre-1.0 version promises anything (#21).</p>\n<p>Five questions in this RFC needed Fable under the four-question test (F1–F5). Section 6\nstates each item, with the test as it was applied, and section 10 rules it.</p>\n<h2>Summary</h2>\n<p><strong>The protocol has no fundamental blocker for enterprise use.</strong> Nothing on the wire\nassumes public DNS or the public internet. An origin is a URL string. Resolution and\nverification are fetches. Every fetch failure already degrades safely: a poll failure\ndiscards nothing (§12.3), and an unreachable source is a failed mention (§15.4). The\n<code>blyg:</code> namespace URI is an opaque token that is never fetched (§7).</p>\n<p>Three things need care, and only one of them is a missing capability:</p>\n<ol>\n<li><strong>One audience per origin.</strong> A whole origin behind SSO or a private network works\ntoday. Different readers seeing different items <em>within one origin</em> is L3, which #7\ndefers. Enterprises get audiences by running one origin per audience: the N-origins\ngroup shape, <code>tn-3</code> shape A. This is the missing capability (F1, F4).</li>\n<li><strong>References leak across a boundary.</strong> A quote bakes the bytes. Every reference\ncarries an origin string, and <code>cited</code> can carry an excerpt. A sent mention POSTs the\ninternal URL to the outside receiver. The protocol permits all of this. An enterprise\nclient has to refuse it at publish time (F3).</li>\n<li><strong>Retention law runs against two promises</strong> the medium makes: published documents\nanswer 200 forever, and pins are irrevocable (F5).</li>\n</ol>\n<p>Everything else is client work, and section 4 lists it: SSO and roles, reader\ncredentials, egress policy, audit and retention, hosting and residency, AI controls, and\nthe publish-time boundary.</p>\n<h2>1. The question</h2>\n<p>Venkat asked whether anything fundamental stops Blygger being used as an enterprise\ntool, deployed inside an organisation's boundary with the usual enterprise features:\nSSO, an internal DNS scope, private networks, audit, retention.</p>\n<p>The answer matters beyond enterprises. The protocol's claim is that anything able to\nserve files can host a blyg (§1). An intranet file server can serve files. If the claim\nheld only on the public internet, it would be narrower than it says.</p>\n<p><code>plans/private-items-proposal.md</code> §8 (same session) recorded Venkat's answer that\n&quot;enterprise deployments are under discussion&quot; and are the concrete case likely to open\nL3. This RFC is the enterprise half of that thread. It does not repeat the\nprivate-items design. It asks what an enterprise needs <em>before</em> L3, and whether L3 is on\nits critical path.</p>\n<h2>2. Deployment shapes</h2>\n<table>\n<thead>\n<tr>\n<th>Shape</th>\n<th>What it is</th>\n<th>Audience per origin</th>\n<th>Protocol status</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><strong>A. Fully internal</strong></td>\n<td>Every blyg sits on an internal DNS zone (for example under <code>.internal</code>, which ICANN reserved for private use in 2024) or behind a gateway. Nothing is reachable from outside.</td>\n<td>uniform (employees)</td>\n<td>conformant today, given F1</td>\n</tr>\n<tr>\n<td><strong>B. Internal plus selectively public</strong></td>\n<td>Some origins internal, some public: a public engineering blog next to internal team blygs.</td>\n<td>uniform per origin; different origins, different audiences</td>\n<td>conformant today; the boundary between origins is F3</td>\n</tr>\n<tr>\n<td><strong>C. Mixed federation</strong></td>\n<td>Internal blygs subscribe to and quote public blygs, and the reverse is attempted.</td>\n<td>as B</td>\n<td>inbound (public → internal) is safe; outbound is F3</td>\n</tr>\n<tr>\n<td><strong>D. Vendor multi-tenant hosting</strong></td>\n<td>A vendor hosts many organisations' blygs, each tenant one or more origins.</td>\n<td>uniform per origin</td>\n<td>conformant today: #14, #36, #61, #84 make N origins on one host safe</td>\n</tr>\n<tr>\n<td><strong>E. Per-item audiences in one origin</strong></td>\n<td>One team blyg where some items are for the team, some for the company, some public.</td>\n<td><strong>mixed</strong></td>\n<td><strong>L3</strong> (#7). Not available.</td>\n</tr>\n</tbody>\n</table>\n<p>Shapes A–D need no new protocol. Shape E is the only one that does, and the\nrecommendation in section 3.3 is to build A–D first and treat E as the measured need\nthat opens L3 (F4).</p>\n<p><strong>Shape B is &quot;promotion&quot; by publishing anew.</strong> Withdrawal is the only exit from public\n(#8), and an origin's audience never changes (F1). So an internal item does not <em>become</em>\npublic. Its author publishes a new item in the public origin. Whether that new item may\nname its internal ancestor is F3: <code>forked_from</code> or a quote would carry the internal\norigin string outward.</p>\n<h2>3. Protocol analysis: is anything blocking?</h2>\n<p>Each claim below is checked against the 0.4 text, not assumed.</p>\n<h3>3.1 Nothing assumes public DNS</h3>\n<table>\n<thead>\n<tr>\n<th>Construct</th>\n<th>What it needs</th>\n<th>Internal origin?</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Origin, identity origin (§2, §12.2, #14, #17)</td>\n<td>an absolute base URL; identity is the final fetch URL minus its last segment</td>\n<td><strong>Fine.</strong> No rule requires a publicly resolvable name. §12.1 step 1 expects HTTPS; an internal CA is a transport matter.</td>\n</tr>\n<tr>\n<td>Reference shape (§5.9)</td>\n<td><code>{origin, id, version}</code> as strings</td>\n<td><strong>Fine</strong> as data. A reader outside cannot check staleness, which §12.3 already treats as an ordinary failure.</td>\n</tr>\n<tr>\n<td><code>content_html</code> absolute URLs (§5.2, #53)</td>\n<td>every URL in the content absolute</td>\n<td><strong>Fine</strong> inside. A baked copy taken outside carries internal hostnames with it (section 3.2).</td>\n</tr>\n<tr>\n<td>Namespace URI <code>https://blygger.org/ns/0.1</code> (§7)</td>\n<td>&quot;an opaque wire token&quot;</td>\n<td><strong>Fine.</strong> Never fetched, so an air-gapped reader needs no access to <a href=\"http://blygger.org\">blygger.org</a>.</td>\n</tr>\n<tr>\n<td><code>generator_url</code> (§6.1)</td>\n<td>informative, MUST NOT gate</td>\n<td><strong>Fine.</strong></td>\n</tr>\n<tr>\n<td>Resolution (§12.1)</td>\n<td>at most six fetches from the reader's network position</td>\n<td><strong>Fine</strong> for any reader that can reach the origin.</td>\n</tr>\n<tr>\n<td>Verification (§15.4)</td>\n<td>the receiver fetches the source from its asserted origin</td>\n<td><strong>Fine</strong> when the receiver can reach the source. Otherwise <em>failed</em>, which is honest (F2).</td>\n</tr>\n<tr>\n<td>Webmention endpoint (§15.1)</td>\n<td>optional at every level</td>\n<td><strong>Fine.</strong> A static internal blyg omits it (§15.7).</td>\n</tr>\n<tr>\n<td>CORS (§4)</td>\n<td><code>Access-Control-Allow-Origin: *</code> is a SHOULD</td>\n<td><strong>Fine.</strong> A gated origin answers credentialed CORS instead; the SHOULD bends for a stated reason.</td>\n</tr>\n<tr>\n<td>Census (§3.2)</td>\n<td>&quot;the manifest is a file the protocol requires to be public&quot;</td>\n<td><strong>Wording only.</strong> The census simply never sees internal blygs. &quot;Public&quot; here means the same bytes to every reader who can reach the origin (F1, ruled in section 10; §4 now says so).</td>\n</tr>\n</tbody>\n</table>\n<p><strong>One hazard to rule out: split-horizon DNS for a blyg origin.</strong> If one name resolves\nto an internal server inside the network and a different server outside, then &quot;the item\nat that origin&quot; depends on where the reader stands. That breaks §12.2's guarantee that\ntwo readers given the same URL arrive at the same identity, and it reproduces per-reader\ncontent under one origin, which is shape E by the back door. Internal origins should use\nnames that do not exist publicly. Section 8 lists this as a rejected design.</p>\n<h3>3.2 References across the boundary</h3>\n<p>Every cross-origin construct carries the target's origin string, and two of them carry\nwords:</p>\n<ul>\n<li><strong>A quote bakes bytes.</strong> §10.2 bakes the publisher's local snapshot into\n<code>content_html</code>. §10.4 says the snapshot survives anything the source does, and §14\nsays so plainly (&quot;anything published can be quoted into a thread whose snapshot its\npublisher cannot recall&quot;). A quote of an internal item into a public blyg copies the\ninternal text, and its internal absolute links (#53), out of the boundary for good.</li>\n<li><strong>A <code>cited</code> object</strong> (§5.9) may carry an <code>excerpt</code> and a <code>url</code>, so even a bare <code>stub_of</code>\nor <code>forked_from</code> can carry words.</li>\n<li><strong>Origin strings alone leak</strong> internal hostnames, mount paths and item ids. That is\ntrue of <code>stub_of</code>, <code>forked_from</code>, <code>links[]</code> (§10.3a), <code>transclusions[]</code> and\n<code>generated[].sources[]</code> (§5.7).</li>\n<li><strong>A sent mention leaks in the other direction.</strong> §15.2 says a publisher SHOULD send a\nmention for every remote reference. From an internal blyg that quotes a public one, the\nPOST hands the public receiver the internal page URL as <code>source</code>. The receiver then\ntries to fetch it: the fetch fails (good), but it has learned the hostname and the\nfact of the quote (bad).</li>\n<li><strong>Generation is a quote by model.</strong> A TK span drawing on an internal item, published\npublicly, carries the internal origin in <code>generated[].sources[]</code> and paraphrased\ncontent in the text. The 0.4 rationale for the <code>source</code> relation, that &quot;paraphrase by\nmodel must not be the one way to use a stranger's words without telling them&quot;, cuts\nthe same way here.</li>\n</ul>\n<p>None of this is a protocol defect. The protocol is a public-speech medium, and quoting is\nits primitive (#12). It does mean that <strong>the boundary is enforced by the publisher at\npublish time, or not at all.</strong> No reader-side or wire-side mechanism can recall a baked\nquote. That is the same conclusion <code>private-items-proposal.md</code> §3.4 reached for private\nitems, generalised from &quot;private versus public&quot; to &quot;narrower versus wider audience&quot;.</p>\n<p>The rule a boundary-aware client applies:</p>\n<blockquote>\n<p><strong>A version published at an origin with audience W may carry no reference (of any\nkind, <code>cited</code> included) to an origin whose audience is narrower than W, and sends no\nmention to a receiver outside the audience of its own origin.</strong></p>\n</blockquote>\n<p>The audiences are client configuration: a list of origin prefixes per audience. Both\nhalves are within what the spec already permits. Refusing to quote is a publish error\nthe client chooses. Not sending a mention is a SHOULD the client declines for a stated\nreason. The spec says nothing more (F3, ruled in section 10): §14 already carries the\nsentence that matters, &quot;anything published can be quoted into a thread whose snapshot its\npublisher cannot recall&quot;, and the rule above is client guidance in this note.</p>\n<h3>3.3 &quot;Readable only by members&quot;: static, L1/L2, or L3?</h3>\n<p>The distinction that settles this is <strong>where the audience boundary falls</strong>.</p>\n<ul>\n<li><strong>Origin-uniform audience.</strong> The whole surface (manifest, feed, index, item documents,\npins, media, pages) sits behind one gate: a private network, a VPN, an identity-aware\nproxy, Cloudflare Access. Everything inside the gate is plain static files, every\nreader who passes the gate sees the same bytes, and §4's 404/200 rules hold for every\nsuch reader. The gate is transport, like TLS. This is conformant at L1 and L2 and\nneeds nothing new (F1, confirmed in section 10). §4 now says it in one sentence: &quot;What\nan origin serves does not depend on who asks: every reader who can reach it gets the\nsame bytes, and reachability is the deployment's.&quot;</li>\n<li><strong>Per-item or per-reader audience in one origin.</strong> Two readers asking for the same\n<code>items/{id}.json</code> get different answers (200 versus 401/404, or full versus redacted\ncontent). That is the gate <code>private-items-proposal.md</code> §5 analyses as C1–C3. Section\n10 classifies them: C2 is refused, C3 is where an L3 round starts, and C1 is client\nterritory under one invariant. The proposal's &quot;not an option&quot;, a gated page over an\nopen JSON, applies here too.</li>\n</ul>\n<p>So an enterprise gets members-only reading <strong>today</strong> by running one origin per audience\nbehind a gate. A company-wide blyg, a team blyg and a public blyg are three origins, on\none host if it likes (<code>tn-3</code> shape A). The cost is the one <code>tn-3</code> names: N origins, N\narchives, N sets of pins.</p>\n<p><strong>What this does not give</strong> is an item that is team-only today and company-wide\ntomorrow: the publish-anew path (section 2). That is where L3, or a studio-side\ncross-origin &quot;promote&quot; with honest lineage, would earn its cost. Section 10 rules it the\ntrigger for L3 (F4), measured when a deployment hits that cost with a named case.</p>\n<h3>3.4 Reader side</h3>\n<ul>\n<li><strong>Readers must reach the origin.</strong> A subscribing studio behind the gate needs network\nreach and, when the gate is identity-aware, a machine credential (a service token,\nmTLS, workload identity) for each gated origin. That credential is transport. The\nprotocol has no opinion on request headers, and §12.3's polling rules are unchanged.\nIt is client work (section 4.2).</li>\n<li><strong>Failures are already safe.</strong> A reader that loses access keeps its stored state and\nbacks off (§12.3). It never treats 401 as withdrawal, because only an endcap is\nwithdrawal (§13.4).</li>\n<li><strong>Sanitize-at-render is unchanged</strong> (§14). The internal-content risk is egress, not\nrendering.</li>\n<li><strong>Retention follows the origin</strong> (§13.4). A records system that keeps withdrawn\ncontent because the law requires it acts outside the protocol, and keeps the archive\noff every surface the spec governs (F5, ruled in section 10). §14 now says: &quot;Law may\ncompel a publisher to remove what it promised to serve, or a reader to retain what an\norigin withdrew. Both are acts outside the protocol: a compelled removal is a 404 with\nno marker, and a compelled retention never reaches any surface this document governs.&quot;</li>\n</ul>\n<h3>3.5 Identity and SSO</h3>\n<p>Invariant 4 and #11 say that identity is never in the protocol, and that the only\nauthenticated entity is the origin. This <strong>helps</strong> enterprise use rather than blocking\nit. The protocol takes no position on how an owner, an editor or a reader proves who\nthey are, so SSO is entirely the client's:</p>\n<ul>\n<li><strong>Write side.</strong> §14's last bullet and §16.6 put authoring auth out of scope, and #31\nrules out a normative write surface. A client's authoring auth is its own; an IdP\nsession can be its root credential. The pattern #31 recommends (scoped,\nowner-revocable tokens for tools, a root credential no tool holds) carries over\nunchanged.</li>\n<li><strong>Bylines.</strong> An SSO-backed studio can assert verified employee bylines (<code>author.name</code>,\n<code>author.url</code> pointing to an internal directory profile). #35's <code>rel=&quot;me&quot;</code> proof works\ninside the network exactly as outside. RFC-1's <code>ids</code> and address book apply unchanged.\nBylines are still the origin's claim (§5.5), which inside one organisation is the\nright trust model.</li>\n<li><strong>Multi-author.</strong> The invariant is single <em>publisher</em> per origin, not single author\n(#11). A team blyg with many SSO users is <code>tn-3</code> shape B, the masthead. #36's test,\n&quot;who can withdraw&quot;, decides between shape A and shape B per team.</li>\n<li><strong>Reader identity.</strong> This is where SSO touches the protocol, and only in shape E,\nwhere a group needs named readers. That is F4, and #11 calls it out (&quot;keys imply\nidentity&quot;).</li>\n</ul>\n<h3>3.6 Verdict</h3>\n<table>\n<thead>\n<tr>\n<th>Item</th>\n<th>Kind</th>\n<th>Where</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Public DNS or internet assumed anywhere</td>\n<td><strong>none found</strong></td>\n<td>3.1</td>\n</tr>\n<tr>\n<td>Whole-origin gating (SSO proxy, private network)</td>\n<td>interpretation; one §4 sentence</td>\n<td><strong>F1</strong>, ruled, §10</td>\n</tr>\n<tr>\n<td>Verification when the source is unreachable</td>\n<td>interpretation, small</td>\n<td><strong>F2</strong>, ruled, §10</td>\n</tr>\n<tr>\n<td>Cross-boundary quoting and reference leakage</td>\n<td>client rule, no spec text</td>\n<td><strong>F3</strong>, ruled, §10</td>\n</tr>\n<tr>\n<td>Per-item audiences in one origin</td>\n<td><strong>missing capability (L3)</strong></td>\n<td><strong>F4</strong>, ruled, §10</td>\n</tr>\n<tr>\n<td>Compelled deletion; compliance archives of others' content</td>\n<td>out of protocol; one §14 sentence</td>\n<td><strong>F5</strong>, ruled, §10</td>\n</tr>\n<tr>\n<td>SSO, roles, audit, residency, egress, AI controls, DLP</td>\n<td>client and product work</td>\n<td>4</td>\n</tr>\n</tbody>\n</table>\n<h2>4. Concerns for clients serving enterprise use</h2>\n<p>The protocol governs only the published surface (#3). Everything below is the client's\nbusiness. It is listed so that implementers see the whole job, and so that two enterprise\nclients can make compatible choices where they touch.</p>\n<h3>4.1 Authoring identity: SSO and roles</h3>\n<ul>\n<li>The organisation's IdP (OIDC, or SAML through a broker) is the root credential. No\ntool holds it; tools get scoped, revocable tokens (the pattern of #31).</li>\n<li>Many people may author in one origin. The protocol needs one <em>publisher</em> per origin,\nnot one author (#11), so a team blyg is <code>tn-3</code> shape B with bylines, or shape A with\none origin per member. #36's test, &quot;who can withdraw&quot;, decides between them.</li>\n<li>IdP groups map onto the client's roles. Deprovisioning (SCIM or equivalent) revokes a\ndeparting user's sessions and tokens.</li>\n<li>Verified bylines: <code>author.url</code> can point at the internal directory profile, and #35's\n<code>rel=&quot;me&quot;</code> proof works inside the network as outside.</li>\n</ul>\n<h3>4.2 Reading gated origins</h3>\n<ul>\n<li><strong>Gate the whole surface</strong> (manifest, feed, index, item documents, pins, media, pages)\nwith one gate, so that every reader past it gets the same bytes (F1). Gating only the\nHTML leaves the full text in the JSON.</li>\n<li><strong>A subscribing client needs a machine credential</strong> per gated origin (service token,\nmTLS, workload identity). Attach it only on an exact origin-prefix match, and never\nfollow a redirect with it to another origin: §12.2 allows cross-host redirects, and a\ncredential that follows one leaks.</li>\n<li><strong>Caches</strong> must be keyed so that nothing cached for one audience is served to another.</li>\n</ul>\n<h3>4.3 Network reach and egress</h3>\n<ul>\n<li><strong>Internal names.</strong> A client fetches manifests, feeds and documents (§12), and a\nreceiver fetches mention sources (§15.4). Inside the network those names often exist\nonly in internal DNS, and a resolver that consults public DNS will not find them.</li>\n<li><strong>The probe risk.</strong> §14 asks receivers to refuse private-network addresses where they\ncould be used to probe them. The Webmention endpoint is the dangerous one: anyone who\ncan POST to it chooses a URL for the receiver to fetch. An internal deployment needs an\nallowlist (host suffixes or origins, plus address ranges), not a switch that turns the\ncheck off, and the endpoint should fetch allowlisted sources only for claims that\narrived from inside (F2).</li>\n<li><strong>Air gaps.</strong> In an air-gapped deployment, no feature may reach a public host: CDN\nscripts, update checks, embeds, hosted AI providers.</li>\n<li><strong>Rebinding.</strong> A resolve-then-fetch check does not pin the connection. Enforce egress\nat the network boundary as well.</li>\n</ul>\n<h3>4.4 Audit, retention and legal hold</h3>\n<ul>\n<li><strong>Audit.</strong> An append-only, exportable log of sign-ins, token grants and revocations,\npublish, withdraw and pin, visibility and settings changes, subscriptions, outbound\nmentions, and AI calls (model and provider; not prompt text by default).</li>\n<li><strong>Legal hold</strong> mostly comes free on the published side: withdraw-not-delete (#8) keeps\nevery published version, and pins keep their bytes. The gaps are unpublished material\n(drafts, private items, media), which a hold must turn from deletes into soft deletes.</li>\n<li><strong>Retention schedules and compelled deletion</strong> run the other way. A records policy, a\ncourt order or an erasure request can require removing a published item that §4\npromised would answer 200 forever and a pin promised to keep (§8). The removal is an\nact outside the protocol: the document answers 404, with no status code, marker or\nreason (F5, ruled in section 10; the §14 sentence is quoted in section 3.4).</li>\n<li><strong>Archives of external content.</strong> Keeping other origins' withdrawn content because the\nlaw requires it is outside the protocol too. A client that is both reader and archive\nkeeps the archive off every surface the spec governs: no presentation, no re-serving,\nno quote from it (§13.4, F5).</li>\n</ul>\n<h3>4.5 Hosting and data residency</h3>\n<p>The protocol is host-neutral: any server that can serve the files conforms (§4), so the\npublished surface can sit on an internal file server behind any gate. It loses only the\nWebmention endpoint, which is optional (§15.7). Where the authoring side runs, and where\nits data lives, is the client's design. Residency is a property of that design and of\nits hosts.</p>\n<h3>4.6 AI provider controls</h3>\n<p>AI is never in the protocol (#5). An enterprise client needs:</p>\n<ul>\n<li>an operator-controlled list of allowed models and providers, with no free-entry way\naround it;</li>\n<li>the ability to point at an internal gateway;</li>\n<li>AI calls inside the egress policy;</li>\n<li>the boundary rule (section 3.2) applied to generation sources, since a TK source from\na narrower audience is a cross-boundary reference.</li>\n</ul>\n<h3>4.7 The publish-time boundary (DLP)</h3>\n<p>This is F3, and the most important enterprise control. Section 10 rules that it is\nclient policy with no spec text: declining a SHOULD for a stated reason is what SHOULD\nmeans, and §14 already warns that anything published can be quoted out of its\npublisher's reach. Revisit only if two boundary-building clients disagree.</p>\n<ul>\n<li><strong>Audiences are configuration:</strong> named sets of origin prefixes.</li>\n<li><strong>Publish block.</strong> A version may carry no reference (transclusions, <code>stub_of</code>,\n<code>forked_from</code>, <code>links[]</code>, <code>generated[].sources[]</code>, <code>cited</code>) to an origin narrower than\nits own. Say which references block, and offer to remove them. This is the same block\n<code>plans/private-items-proposal.md</code> §3.4 describes for private items, keyed on audience.</li>\n<li><strong>Mentions.</strong> Send none to a receiver outside the origin's own audience, and show the\nauthor why the §15.2 SHOULD was declined.</li>\n<li><strong>Content scan.</strong> Optionally, check <code>content_html</code> for absolute URLs under narrower\norigins (#53 makes them easy to find): a plain link leaks a hostname.</li>\n<li><strong>Egress as backstop.</strong> A wider-audience deployment that cannot fetch a narrower origin\ncannot quote it either.</li>\n</ul>\n<h3>4.8 Multi-tenant hosting</h3>\n<p>N origins on one host are safe on the wire: #61 stops one tenant's documents verifying in\nanother's name, #84 makes target membership exact, and #87 fixes the manifest name at\nevery identity. Each tenant is the publisher of its origins (#11), not the vendor.\nIsolation of data, egress policy, audit and IdP per tenant is the vendor's design.</p>\n<h2>5. Wire impact</h2>\n<p><strong>None.</strong> F1 and F5 are each one sentence in the living 0.4 text that readers already\nhandle; F2 and F3 add nothing; F4's shape E stays L3. Shapes A–D use only existing\nconstructs, and the boundary rule is a client publish error and a declined SHOULD.</p>\n<h2>6. The five items and their tests</h2>\n<p>The four questions are those in <code>CLAUDE.md</code> → <em>Model routing</em>: (1) readers must change;\n(2) reader-visible addition; (3) narrows, widens or reinterprets a locked decision;\n(4) a locked decision already settles it. The items are kept as the record of the test\nas the draft applied it; each <em>Proposal</em> now opens with the ruling.</p>\n<p><strong>F1. Audience is a property of the origin.</strong></p>\n<ul>\n<li><em>Proposal:</em> Ruled, section 10: confirmed, with one sentence in §4. The draft: the\nspec governs the bytes an origin serves to whoever can reach it.\nReachability, including an identity-aware gate across the <em>whole</em> surface, is\ndeployment. A gated origin conforms at L1 and L2, provided that every reader who\npasses the gate gets the same bytes. Per-reader difference within an origin stays L3.\n§3.2's &quot;required to be public&quot; reads as &quot;served to its audience&quot;. A blyg origin MUST\nNOT be split-horizon.</li>\n<li><em>Test:</em> Q1 no. Q2 no. <strong>Q3 yes</strong>: it reads #1 and #7 (&quot;privileged-group access is\ndeferred to L3&quot;) as being about per-item audiences, not whole-origin ones, and that is\nan interpretation Opus should not make alone. Q4: #3 and #14 point the same way, but\ndo not settle it.</li>\n<li><em>Touches:</em> #1, #3, #7, #14, #17; §3, §3.2, §4 (200 forever, CORS), §12.2, §16.7.</li>\n</ul>\n<p><strong>F2. Verification across a network boundary.</strong></p>\n<ul>\n<li><em>Proposal:</em> Ruled, section 10: confirmed, no text. The draft: a mention whose source\nthe receiver cannot fetch is <em>failed</em>, with no new\nstate. §14's &quot;refusal to fetch private-network addresses where the deployment could be\nused to probe them&quot; is met by a deployment-scoped allowlist, because internal fetches\nthat the deployment intends are not probes.</li>\n<li><em>Test:</em> Q1 no. Q2 no. Q3 <em>maybe</em>: it reads §14's private-network SHOULD as\nconfigurable. Q4: #28 and #61 settle the mechanics, but not §14's wording. Small.</li>\n<li><em>Touches:</em> #13, #28, #61, #84; §14, §15.2, §15.4.</li>\n</ul>\n<p><strong>F3. References that cross a boundary.</strong></p>\n<ul>\n<li><em>Proposal:</em> Ruled, section 10: confirmed as client policy, no text. The draft: client\npolicy in this RFC (section 3.2's rule), with <strong>no</strong> spec text yet.\nRevisit only if a second client builds a boundary and the two disagree.</li>\n<li><em>Test:</em> Q1 no and Q2 no while it stays client policy. Q3 <em>maybe</em>: declining §15.2's\nSHOULD wholesale for a class of receivers, and a publish error the spec does not list\nin §10.2 step 3. Q4: #3 lets the client refuse anything at publish. If Fable wants a\nsentence in §14 (&quot;a quote is an export&quot;), that is spec text.</li>\n<li><em>Touches:</em> #9, #12, #26, #53; §5.6, §5.7, §5.9 <code>cited</code>, §10.2, §10.3a, §14, §15.2;\n<code>private-items-proposal.md</code> §3.4.</li>\n</ul>\n<p><strong>F4. Group visibility is L3, and is enterprise its trigger?</strong></p>\n<ul>\n<li><em>Proposal:</em> Ruled, section 10: confirmed, with C1–C3 classified. The draft: enterprise\ndoes not need L3 to start (section 3.3). Shape E is the trigger,\nmeasured when a deployment hits the publish-anew cost in practice. Until then, group\ntokens wait, as <code>private-items-proposal.md</code> §8.3 says.</li>\n<li><em>Test:</em> any form of shape E is <strong>Q1, Q2 and Q3 yes</strong> (C2), or Q3 yes (C1).</li>\n<li><em>Touches:</em> #1, #7, #11, #35, #43; §3, §4, §16.7; <code>private-items-proposal.md</code> §5, §8.3;\nbacklog L3 row.</li>\n</ul>\n<p><strong>F5. Compelled removal and compliance retention.</strong></p>\n<ul>\n<li><em>Proposal:</em> Ruled, section 10: a sentence in §14, no marker. The draft: say in §14\nthat a legally compelled removal is an out-of-protocol act. The\norigin breaks a promise and readers see a 404, which §12.3 already tolerates. Add no\nstatus code or marker. For the reader side, ask whether a non-presenting compliance\narchive is &quot;hoarding&quot; under §13.4.</li>\n<li><em>Test:</em> a §14 sentence is Q1 no, Q2 no, <strong>Q3 yes</strong> (it qualifies #8's &quot;200 forever&quot;\nand the irrevocable pin). A marker such as 410 with a reason would be Q1 and Q2 yes.</li>\n<li><em>Touches:</em> #8; §4, §8, §9, §13.4, §14.</li>\n</ul>\n<h2>7. Open questions for comment</h2>\n<p>For implementers building enterprise products on Blygger. F1–F5 are ruled (section 10);\nthese questions are about what implementers need, and where an answer bears on a ruling\nit says which.</p>\n<ol>\n<li><strong>Is one origin per audience enough?</strong> Name the case where publishing anew across\naudiences fails you. That case is what opens L3 (F4).</li>\n<li><strong>Reader credentials.</strong> How should a subscribing client authenticate to a gated\norigin: service tokens, mTLS, workload identity, something else? Is a shared\nconvention worth an RFC, or is it each product's business?</li>\n<li><strong>Outbound mentions from inside.</strong> Should an internal blyg ever notify a public one\nthat it quoted it? If so, under what explicit act?</li>\n<li><strong>Promotion.</strong> When an internal item is republished publicly, should the public item\nname its internal ancestor (leaking the origin), carry a public lineage note, or\ncarry nothing?</li>\n<li><strong>Retention.</strong> Which do you need: compelled deletion of published items, legal hold\nover drafts, a compliance archive of external content, or all three? Does the §14\nsentence (F5) cover what you need to build, and if not, what else do you need — legal\nhold on drafts, archives, something the sentence does not reach?</li>\n<li><strong>Hosting.</strong> Where must the authoring side run, and what residency do your buyers\nneed?</li>\n<li><strong>Audit.</strong> Which events, which format, which sink?</li>\n<li><strong>Teams.</strong> For your users, is a team blyg shape A (members' own origins) or shape B\n(one masthead)? Who withdraws?</li>\n<li><strong>Should the boundary rule (section 3.2) ever become spec text?</strong> It is client policy\ntoday (F3); name the disagreement between two clients that would change that.</li>\n</ol>\n<h2>8. Rejected designs</h2>\n<p>Recorded because each will be proposed.</p>\n<ol>\n<li><strong>Split-horizon blyg origins.</strong> One name, two servers, two contents. This breaks\n§12.2's determinism and is per-reader content under one origin (section 3.1).</li>\n<li><strong>Gating the HTML pages only.</strong> The JSON carries the full text (private-items\nproposal §5, &quot;not an option&quot;).</li>\n<li><strong>A 401 or 403 on chosen item documents within an otherwise readable origin.</strong> This\nis C2. §4 has no third answer, and §15.4 cannot verify against it. It is L3.</li>\n<li><strong>Turning private-network refusal off wholesale for an internal deployment.</strong> It\nworks, and it makes the Webmention endpoint an internal fetch proxy for anyone who can\nreach it. Use an allowlist (section 4.3).</li>\n<li><strong>An enterprise &quot;audience&quot; member in the manifest or item documents.</strong> Readers cannot\nact on it without identity, which is #11's line. Audience is the client's\nconfiguration plus the gate.</li>\n<li><strong>A removal marker (a 410 with a reason, a tombstone).</strong> A removal marker is the\ntombstone #8 refused, and a delete signal invites delete requests. A compelled removal\nis a 404 with no marker, which readers already tolerate (§12.3; F5, section 10).</li>\n</ol>\n<h2>9. Relation to other documents</h2>\n<ul>\n<li><code>plans/private-items-proposal.md</code>: the owner-only level and its publish block. This\nRFC generalises that block to audiences (section 4.7) and supplies F4's trigger\nframing for its §8.3.</li>\n<li><code>docs/notes/tn-3-groups-and-aggregation.md</code>: shapes A and B are the enterprise team\nshapes.</li>\n<li><code>docs/notes/tn-6-mentions-and-identifier-schemes.md</code> (formerly RFC-1): bylines, <code>ids</code>\nand agent members under SSO (section 3.5).</li>\n<li><code>docs/notes/tn-7-comments-sections.md</code> (formerly RFC-2): hosted commenters, which\ninside an organisation are SSO users commenting through the house's comments origin.</li>\n</ul>\n<h2>10. Rulings (Fable 5.1, session 60, 2026-11-05)</h2>\n<p>The draft closed with five items for Fable. The rulings follow, each with the principle\nit rests on. All five confirm the draft's direction; F4 adds a classification the draft\ndid not make, and F5 settles the reader side the draft left as a question.</p>\n<p><strong>F1. Audience is a property of the origin — confirmed, with spec text.</strong> Decision #1\nmakes the protocol a static file contract, and #14 makes an origin any absolute URL;\nnothing in the text requires that an origin be publicly reachable. &quot;Public&quot; in #7 and in\n§3.2 means the same bytes to every reader who can reach the origin. Reachability is\ndeployment, like TLS. Per-reader difference within one origin is L3 (#7), and a\nsplit-horizon origin is that difference by the back door, because it breaks §12.2's\ndeterminism. The tenth revision of 0.4 adds the sentence to §4: &quot;What an origin serves\ndoes not depend on who asks: every reader who can reach it gets the same bytes, and\nreachability is the deployment's.&quot;</p>\n<p><strong>F2. Verification across a network boundary — confirmed, no text.</strong> §14's clause is\nalready conditional: it asks for refusal &quot;where the deployment could be used to probe\nthem&quot;. A fetch the deployment intends, under an allowlist, is not a probe. A source the\nreceiver cannot reach is <em>failed</em> (§15.4), which is honest: the receiver could not verify\nit.</p>\n<p><strong>F3. References that cross a boundary — confirmed as client policy, no text.</strong> §14\nalready says the sentence the draft wanted: &quot;anything published can be quoted into a\nthread whose snapshot its publisher cannot recall&quot;. The boundary rule (section 3.2: no\nreference of any kind, <code>cited</code> included, to a narrower origin; no mention to a receiver\noutside the origin's audience) lives in this note as client guidance. Declining a SHOULD\nfor a stated reason is what SHOULD means, so a client that withholds §15.2's mentions\nacross the boundary conforms. Revisit only if two boundary-building clients disagree.</p>\n<p><strong>F4. Group visibility is L3 — confirmed, and the three shapes classified.</strong> Enterprise\ndoes not need L3 to start. Shape E is the trigger, measured when a deployment has hit the\npublish-anew cost with a named case. Of the private-items proposal's three gate shapes:</p>\n<ul>\n<li><strong>C2</strong> (a 401 or 403 on chosen documents, under any gate descriptor) is refused. It is\nnot static, and it is L3 by the back door.</li>\n<li><strong>C3</strong> (encrypted static content) is the shape the L3 round starts from when the\ntrigger fires.</li>\n<li><strong>C1</strong> (a redacted wire, with a fuller rendition served off the mount to signed-in\nreaders) is <strong>client territory under #3, not L3</strong>, under one invariant: nothing on the\nprotocol surface may name, link to or depend on the fuller rendition, and the public\ndocument is complete in itself. The redaction is the author's own visible edit (#78).\nThat is a paywalled newsletter beside a blyg, which the protocol has never forbidden.</li>\n</ul>\n<p><strong>F5. Compelled removal and compliance retention — a sentence in §14, no marker.</strong> &quot;200\nforever&quot; and the irrevocable pin are the publisher's promises. Law may force the\npublisher to break them, and the protocol does not legislate the law. Readers already\ntolerate a 404 (§12.3). There is no status code and no reason: a removal marker is the\ntombstone #8 refused, and a delete signal invites delete requests (section 8, item 6).\nOn the reader side, §13.4 governs what a conformant reader presents and re-serves. A\nrecords system that retains under law is outside the protocol in the same way, and a\nclient that is both reader and archive keeps the archive off every surface the spec\ngoverns. The tenth revision of 0.4 adds to §14: &quot;Law may compel a publisher to remove\nwhat it promised to serve, or a reader to retain what an origin withdrew. Both are acts\noutside the protocol: a compelled removal is a 404 with no marker, and a compelled\nretention never reaches any surface this document governs.&quot;</p>\n<p><strong>What happens next.</strong> Published for comment under #66 on 2026-11-05; the period closes\non 2026-12-03 (four weeks). Decision #93 records the five rulings, with this section as\ntheir reasoning (<code>v0.4-plan.md</code> §18). Comments are stubs on the item, and\ncounter-proposals are forks of the pin.</p>\n</div>\n<div class=\"blyg-tk-gen\"><p><em>Source: <a href=\"https://github.com/blygger/blygger-spec/blob/main/docs/rfcs/rfc-5-enterprise-deployment.md\"><code>docs/rfcs/rfc-5-enterprise-deployment.md</code></a> in blygger-spec, which is where revisions are made; each one is published here as a new version. To comment, respond to this item from your own blyg: a stub is a comment, a partial quote names the passage, and a fork of the pinned version is a counter-proposal.</em></p>\n</div>\n","content_hash":"sha256:4daae57014c3f84a95ebe75c73e0aaec00ae50ee935c94e600265f8ad3406b8f","transclusions":[],"generated":[{"sources":[],"model":"claude-opus-5-5+claude-fable-5-1"},{"sources":[],"model":"claude-opus-5-5"}]}