# nfrs.sgit.ai — everything, in one file Site version: v0.1.2. Generated by admin/build/gen_llms_full.py — do not edit by hand. This is /llms.txt, then the front page, then all nine source documents, concatenated in reading order. It exists because agent fetch tools frequently refuse URLs a search has not already returned, which makes link-following unreliable and makes a single-file surface the practical one. If you can make more than one request, prefer the individual documents at https://nfrs.sgit.ai/briefs/ — they are the source of truth and this file is a concatenation of them. One document from the source pack is deliberately absent: the source manifest, which indexes do-not-publish material by repository path. The boundaries it defines are stated in full at https://nfrs.sgit.ai/shipped/index.html#boundaries. All content is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). Third-party material quoted inside these documents stays under its own terms. Contents, in order: 1. llms.txt — the annotated map, each entry carrying its page's single most important fact 2. index.md — the front page as markdown 3-11. the source documents, verbatim ============================================================================== == llms.txt ============================================================================== # nfrs.sgit.ai — the non-functional requirements, from the inside > NFRs — non-functional requirements — is this estate's own operative term, defined in > its own words as "the whole version control, reliability, resilience, security, > backups, consistency, explainability, and documentation". That sentence is this site's > table of contents. Other sites in this network argue that maintaining non-functional > requirements is a scarce and valuable thing; this one is the same argument from the > inside — how the NFRs actually get done here, including the places where they do not. Site version: v0.1.2 (25 August 2026). All content CC BY 4.0. ## What kind of site this is This is a HUB. Most NFR topics already have an owner elsewhere in the sgit.ai network, so this site does two things: it owns the disciplines with no other home, and it holds the topic map that says where everything else lives. It operates under one rule, and the rule is enforced by the pre-release gate rather than by good intentions: LINK THE MEASUREMENT, NEVER RESTATE IT. GENERATE OR DATE EVERY NUMBER. Consequences an agent can rely on: - No measurement on this site is a re-measurement. Where a figure exists, the sibling site that took it is linked, and that link is the citation. - Every page carrying a figure carries an as-of date. CI fails the build otherwise. - Every discipline page links at least one sibling site. CI fails the build otherwise. - Every discipline page ends with its counter-evidence — the measured places the practice being taught did not hold here. ## Properties agents may rely on - Every source document is fetchable at a stable constructed URL: /briefs/v0.33.62__nfrs-brief-pack__.md. This is a promise, not an accident. - The whole site's document set is available in one fetch at /llms-full.txt, for agents whose fetch tools refuse URLs a search has not already returned. - The front page has a markdown twin at /index.md. ## The pages, each with its single most important fact /index.html The eight-NFR sentence is quoted verbatim wherever this site states its scope, and the scorecard teaser appears on the front page rather than being buried. /map/index.html — THE TOPIC MAP. Read this first. Seven topics are OWNED here (testing, CI, documentation, IFD, budgets-as-discipline, project management, resilience patterns); five are LINKED and never restated (security domains, serverless, explainability's grounding ladder, architecture conventions, the villagers market argument); one — backups — is on the estate's own list and has no doctrine anywhere. /memory/index.html — THE MEMORY THESIS "These sites are a more evolved and focused version of what is usually called LLM memory." Memory you can read, cite, version, license and hand to any agent, because it is a website. The evidence it was already the design: guides with for_llms in the filename, the markdown twin at every URL, and an agent-access report diagnosing a memory-retrieval failure in those terms. Its honest limit: in an agent-memory network, a stale page is a false memory, and this estate has already produced four stale artefacts. /testing/index.html No mocks, no patches — and the reason it is affordable here is that cheap typed objects make composing the real thing as easy as configuring a stand-in. THE TESTING PHILOSOPHY AND THE TYPE SYSTEM ARE ONE DECISION; adopting the rule without the substrate will be unaffordable. Counter-evidence: CI runs about two-thirds of collectible tests, and one of the four structural guards has never worked. /ci/index.html Build once, verify before naming: nothing gets a name until it has passed as an anonymous digest, and a machine image must prove itself twice, once as built and once as booted. Counter-evidence: a workflow invoking a binary defined nowhere, sixteen times, and a second version file that nothing reads. /documentation/index.html "If the reality document doesn't list it, it does not exist. Briefs are aspirations, not facts." The rule does not make documents accurate; it makes the corpus ANSWERABLE. Counter-evidence: the reality index was itself dozens of versions stale, and a README described a repository that does not exist. /ifd/index.html — IFD, ITERATIVE FLOW DEVELOPMENT The estate's named methodology, published outside the estate for the first time here. It names developer attention as the scarce resource the process exists to protect, and derives its rules from that. Counter-evidence: the guides are summarised faithfully and have NOT been verified against current practice. /resilience/index.html Every resilience mechanism in this estate traces to a named incident. That is the doctrine, it keeps the mechanism set small and every mechanism explicable, and it was nowhere written down before this page. /budgets/index.html Profitability-first, pre-approve-the-ladder, budget-on-the-step. "The person who approved the first million is, on the evidence, the worst available decider on the second." NONE OF THIS ESTATE'S OWN FINANCIAL FIGURES APPEAR ANYWHERE ON THIS SITE, in any form — the discipline is published and the numbers are not. /pm/index.html "The project manager is where the register becomes work." Plus the original page: how a brief becomes work here — acceptance criteria, numbered asks and tasks, same-day cross-team review, day-indexes, classified debriefs, handover guides. Demonstrated across roughly 4,000 documents and described as a system nowhere until this page. /scorecard/index.html The estate against its own eight-item list, with no plain tick in it. Reading down the failure column produces one sentence: WHAT A MACHINE ENFORCES, HOLDS; WHAT ATTENTION ENFORCES, DRIFTS. The table is hand-assembled and dated, not generated, and says so. /backups/index.html The one NFR on the estate's own list with no doctrine at all. The question to answer first: is the vault model — versioning plus escrow — already the doctrine, unnamed? And the requirement that survives either answer: a restore that has never been performed is not a backup. /shipped/index.html What this site asserts versus what it can demonstrate, by status: measured elsewhere, position, described-from-the-corpus, proposed-and-not-built, absent. Also the do-not-publish boundaries and the licensing. /network/index.html The fourteen sites by owner. Organised by site; /map/ is the same set organised by topic and carries the positions. A position stated on /network/ would be a bug. /documents/index.html The eight source documents published whole, with the raw markdown as the source of truth. The pack's ninth file, a source manifest indexing do-not-publish material by path, is deliberately not published. /admin/index.html · /admin/comms.html · /admin/versions.html How the site is built and gated; the numbered asks still open with the project lead; the release history. ## What this site will not tell you - Any figure from this estate's own finances. Not rates, margins, pricing, projections or runway, in any form, on any page. - Live hostnames, account identifiers or stack addresses. - Market statistics about technical debt or NFR spend. The source brief flags its own market figures as loosely attributed vendor commentary, so this site carries none. - Anything a sibling site measured. It links the sibling instead. This is the rule, not an omission. ## The sibling sites, and what they own https://sgit.ai — the parent project: the vault layer, the CLI, the platform docs https://sg-compute.sgit.ai — the serverless platform, and most of the measurements this site links: testing, CI, resilience artefacts https://coding.sgit.ai — coding conventions and formatting https://open-source.sgit.ai — the villagers argument as economics (this site is its mirror: the same argument from the inside) https://pki.sgit.ai — public key infrastructure for agents https://nhi.sgit.ai — non-human identity, the problem statement https://sg-sentinel.sgit.ai — detection and monitoring https://standards.sgit.ai — standards and the grounding vocabulary https://risks.sgit.ai — risk registers, scoring, evidence https://llms.sgit.ai — the LLM boundary and provenance https://graphs.sgit.ai — graphs and the visual layer https://twins.sgit.ai — digital twins as a primitive https://skills.sgit.ai — agent skills and the drift between them https://wardley-maps.sgit.ai — evolution and lifecycle ## Source documents /briefs/v0.33.62__nfrs-brief-pack__README.md — the pack's reading order /briefs/v0.33.62__nfrs-brief-pack__00__BRIEF.md — the commission, the name, the hub shape /briefs/v0.33.62__nfrs-brief-pack__01__the-topic-map.md — the map, in full /briefs/v0.33.62__nfrs-brief-pack__02__the-memory-thesis.md — the thesis /briefs/v0.33.62__nfrs-brief-pack__03__the-owned-disciplines.md — the five in depth /briefs/v0.33.62__nfrs-brief-pack__04__site-architecture.md — page by page /briefs/v0.33.62__nfrs-brief-pack__05__boundaries-and-licensing.md — the contract /briefs/v0.33.62__nfrs-brief-pack__06__gaps-and-open-questions.md — published unresolved /briefs/v0.33.62__nfrs-brief-pack__LICENCE.md — CC BY 4.0 Attribution: Dinis Cruz, with AI co-authorship. Repository: https://github.com/SGit-AI/SGit-AI__Website__NFRs ============================================================================== == index.md — the front page ============================================================================== # nfrs.sgit.ai — the non-functional requirements, from the inside > **NFRs** — non-functional requirements — is this project's own operative term for > *"the whole version control, reliability, resilience, security, backups, consistency, > explainability, and documentation."* That sentence was written a month before this site > was conceived, and it is the site's table of contents. *Source: · site v0.1.2 · markdown twin of the front page.* --- Other sites in this network sell the argument that [somebody has to be the villagers](https://open-source.sgit.ai). This one is the same sentence from the inside: how the NFRs actually get done here, including where they do not. ## This is a hub, and it behaves like one | The rule | What it means here | |---|---| | **Link the measurement, never restate it** | When a sibling site has measured something, this site links that page. It does not re-measure and does not paraphrase the number. | | **Generate or date every number** | Every figure carries an as-of date or is generated at build time. The pre-release gate rejects a page that carries figures without one. | | **Every discipline page ends with its counter-evidence** | A page teaching a practice closes with the measured places that practice failed here. | The stakes are raised by [the memory thesis](memory/index.html): if these sites are memory an agent reads, a stale page is not an inconvenience, it is a false memory. ## What this site owns | Page | The position, in a line | |---|---| | [The topic map](map/index.html) | Every NFR topic: what it means here, where the canonical material lives, which site owns it. **Read first.** | | [Testing](testing/index.html) | No mocks, no patches — and the testing philosophy and the type system are one decision, not two. | | [CI pipelines](ci/index.html) | Build once, verify before naming. Nothing gets a name until it has passed as an anonymous digest. | | [Documentation](documentation/index.html) | *If the reality document doesn't list it, it does not exist. Briefs are aspirations, not facts.* | | [IFD](ifd/index.html) | The named methodology, published outside the estate for the first time. It protects attention, and derives its rules from that. | | [Resilience](resilience/index.html) | Design for the failure you had. Every mechanism here traces to a named incident. | | [Budgets](budgets/index.html) | Profitability-first, pre-approve the ladder, budget-on-the-step — with none of this estate's own figures, anywhere. | | [Project management](pm/index.html) | The project manager is where the register becomes work — plus how a brief becomes work here. | ## The estate, measured against its own eight-item list *As of 24 August 2026. The full table, with each cell linked to the sibling pack that measured it, is on [the scorecard](scorecard/index.html).* | NFR | The estate, measured | |---|---| | Version control | ✅ auto-increment tags, the `version` file convention | | Reliability / resilience | ✅ watchdog, poller, bake-and-verify — **and** a dead workflow nobody noticed | | Security | ✅ key discipline, allowlists — **and** one disclosed key-in-history incident | | Backups | ⚠️ **[no doctrine at all](backups/index.html)** | | Consistency | 🟡 total compliance where a script checks, partial where it does not | | Explainability | ✅ reality docs, provenance — **and** no evals anywhere | | Documentation | ✅ a governing rule and a very large corpus — **and** a README describing a repository that does not exist | Reading down the failure column produces one sentence: **what a machine enforces, holds; what attention enforces, drifts.** That is the transferable finding of the whole site. ## Why a website, and not a memory feature > *"The idea of these sites is to provide good briefs for agents to learn about how we work > and think — **these sites are a more evolved and focused version of what is usually > called LLM memory**."* Conventional memory is accumulated transcripts, retrieved by similarity, private to one vendor, unversioned and invisible to the person it describes. This is the same function built as publishing: **memory you can read, cite, version, license and hand to any agent, because it is a website.** [The thesis, its evidence, and its honest limits](memory/index.html). ## Honest about itself - [What this site ships](shipped/index.html) — asserted, demonstrated, or proposed-and-not-built, as a table. - [Backups](backups/index.html) — the one item on the list with nothing behind it, published as an absence. - [Admin & engineering](admin/index.html) — how the site is built, and the two rules it enforces on itself in CI. ## For agents - [/llms.txt](llms.txt) — the annotated map, each entry carrying its page's most important fact. - [/llms-full.txt](llms-full.txt) — the whole document set in one fetch. - Source documents at constructed URLs: `/briefs/v0.33.62__nfrs-brief-pack__.md`. --- All content CC BY 4.0. Attribution: Dinis Cruz, with AI co-authorship. Repository: ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__README.md ============================================================================== # nfrs.sgit.ai — brief pack **For:** the agent commissioned to build `nfrs.sgit.ai` **From:** Dinis Cruz, via the SG/Send Librarian **Version:** v0.33.62 · 24 August 2026 **Licence:** CC BY 4.0 — see `LICENSE.md` --- ## What this is The home for the non-functional-requirements topics — CI, testing, security, scalability, resilience, explainability, serverless, architecture, documentation, budgets, project management — built as **the network's hub**: it owns the disciplines that have no other home, and the topic map that shows an agent where everything else lives. **The name is right.** `nfrs.sgit.ai` is the corpus's own operative term — the villagers brief defines it verbatim: *"the whole version control, reliability, resilience, security, backups, consistency, explainability, and documentation."* That sentence is the site's table of contents. Expand the acronym in sentence one; otherwise ship it. **And the commissioning message contains the network's purpose statement:** > *"these sites are a more evolved and focused version of what is usually called LLM memory."* `02__` develops that into a page — memory you can **read, cite, version, license and hand to any agent, because it is a website** — and recommends it for the `sgit.ai` hub too. --- ## Read in this order | File | Words | What it does | |---|---:|---| | **`00__BRIEF.md`** | 1.1k | **Start here.** The naming verdict, the hub shape, the owns/links split, the scorecard | | **`01__the-topic-map.md`** | 0.8k | The hub's core artefact — every NFR topic: position in one line, canonical source, owner site | | **`02__the-memory-thesis.md`** | 0.7k | The sites-as-agent-memory page, with the evidence it was already the design (`for_llms` filenames, the markdown twin, the agent-access report) | | `03__the-owned-disciplines.md` | 0.8k | Testing, CI, documentation/reality, budgets, PM — position, verbatim material, counter-evidence | | `04__site-architecture.md` | 0.3k | Page by page; the generate-or-date rule | | `05__boundaries-and-licensing.md` | 0.3k | The hub's contract: owns vs links-and-never-restates | | `06__gaps-and-open-questions.md` | 0.7k | 6 build-fresh items, 5 open questions, 4 tensions | | `09__source-manifest.csv` | 23 rows | Every source, tiered 0–3, **every path verified on disk** | --- ## The four things worth your attention **1. The scorecard is the credibility move.** The estate measured against its own eight-NFR list: version control ✅, resilience ✅-with-a-dead-workflow, security ✅-with-one-disclosed-incident, **backups ⚠️ no doctrine at all**, consistency 🟡, explainability ✅-with-no-evals, documentation ✅-with-a-fictional-README. Every lapse is already published in a sibling pack — this site adds assembly, not exposure. A clean column would be dishonest; the mixed table is the teaching. **2. IFD is the most original owned asset.** Iterative Flow Development — the estate's named methodology, versioned guides at `v1.2.x`, written explicitly *"for LLMs assisting with IFD-based development… preserving developer flow state."* **Nobody outside the estate has ever seen it.** Publishing it is also a test of the open-source position (*"the moat is a rate, not a wall"*) — `06__` Q3 says decide that explicitly rather than assume it. **3. The most original PM page is a description of what already happens.** "How a brief becomes work here" — acceptance criteria, same-day cross-team reviews, day-indexes, numbered asks, good-failure/bad-failure debriefs, ~4,000 documents of demonstrated practice, described nowhere as a system. It doubles as onboarding for every new agent session, which is the memory thesis in action. **4. Backups is the honest gap.** The eighth item on the corpus's own list, 151 mentions, no doctrine — in the estate with irreversible publishing and frozen vaults. `06__` Q5 asks the right first question: is the vault model (versioning + escrow) already the doctrine, unnamed? An hour of thinking, then either the page or the dated stub. --- ## The hub's one rule **Link the measurement, never restate it; generate or date every number.** This site quotes more sibling measurements than any other — it has the most to lose from drift, and the memory thesis raises the stakes: a stale page in an agent-memory network is a false memory. --- This file is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__00__BRIEF.md ============================================================================== # 00 — The Brief: `nfrs.sgit.ai` **Version** v0.33.62 · 24 August 2026 **From** Dinis Cruz, via the SG/Send Librarian **To** the agent commissioned to build `nfrs.sgit.ai` **Licence** CC BY 4.0 --- ## 1. The commission > *"the home for lots of the NFR related topics we have covered in quite detail — for example CI pipelines, security, scalability, resilience, explainability, serverless, architecture, testing, documentation, budgets/finance, project management…"* And the framing that matters more than the topic list: > *"the idea of these sites is to provide good briefs for agents to learn about how we work and think — **these sites are a more evolved and focused version of what is usually called LLM memory**."* That second sentence is the clearest statement of the whole `*.sgit.ai` network's purpose made anywhere in this project, and it should not stay buried in a chat message. `02__` develops it into a page — and recommends it also land on the `sgit.ai` hub, because it explains all fourteen sites at once. --- ## 2. The name — verdict: yes, `nfrs.sgit.ai` Three reasons it is right: 1. **It is your own operative term, already defined in the corpus.** The villagers brief: *"what I call the non-functional requirements, which fundamentally is **the whole version control, reliability, resilience, security, backups, consistency, explainability, and documentation**."* That sentence is this site's table of contents, written a month before the site was conceived. 2. **It fits the network's naming pattern** — plural subject nouns: `risks.` `graphs.` `skills.` `twins.` `standards.` `nfrs.` reads as one of the family. 3. **The jargon objection does not apply here.** "NFR" is opaque to a general reader — but the stated audience is **agents and technical readers**, for whom it is precise and unambiguous. The alternatives are worse: `quality.` is vague, `engineering.` overclaims, `operations.` misses half the list. Two conditions: **expand the acronym in the first sentence of the front page**, and note that the corpus itself writes it both ways ("non-functional requirements" in prose, NFR in speech) — the site should use *NFRs* as the name and the full phrase on first use per page. --- ## 3. The shape: this is a hub site Unlike `twins.` (a primitive) or `sg-compute.` (a platform), this site is a **hub over topics that mostly already have owners** in the network. The measured densities and the split: | Topic | Files | This site's role | |---|---:|---| | **Testing** | 279 | **OWNS** — no-mocks-no-patches, deploy-via-pytest, the structural-guard pattern, 4,785-tests-in-81s | | **CI pipelines** | 273 | **OWNS** — digest-first multi-arch, AMI bake-and-verify, auto-increment versioning | | **Documentation** | 651 | **OWNS** — the reality-document system, briefs-are-aspirations, the markdown twin, day-indexes | | **IFD (the methodology)** | guides | **OWNS** — Iterative Flow Development, versioned guides written explicitly *"for LLMs"* | | **Budgets / finance** | 410 | **OWNS the discipline** — pre-approve-the-ladder, profitability-first, budget-on-the-step | | **Project management** | 31 | **OWNS** — the PM-is-where-the-register-becomes-work brief, the brief/debrief system itself | | **Resilience** | 100 | **OWNS the patterns** — the watchdog, health poller, teardown paths (artefacts live in `sg-compute.`) | | **Explainability** | 65 | **LINKS** — the grounding ladder is `risks.`/`standards.`; this site owns "computed, not claimed" as an engineering habit | | **Serverless / scalability** | 347 / 104 | **LINKS** — `sg-compute.` owns the platform; this site owns the *requirement* framing | | **Security** | — | **LINKS** — `pki.` `nhi.` `sg-sentinel.` own domains; this site owns the *NFR posture* (audit-before-the-key, key-leak CI gates) | | **Architecture** | — | **LINKS** — `coding.` owns conventions; this site owns responsibility-boundaries as a requirement class | **The rule: this site owns the disciplines that have no other home, and the topic map that shows an agent where everything else lives.** `01__` is that map. --- ## 4. The spine: the villagers sentence, from the inside The network already tells the NFR story from the *market* side — `open-source.sgit.ai` owns *"somebody has to be the villagers"* as an economic argument. **This site is the same sentence from the inside: how *we* actually do the NFRs the villagers sell.** That gives the site an unusual honesty obligation, because this session has already measured the estate against its own NFR list: | NFR (his list) | The estate, measured | |---|---| | Version control | ✅ 2,777 commits/100 days, auto-increment tags, the `version` file convention | | Reliability / resilience | ✅ watchdog, poller, bake-and-verify — **and** a dead workflow (`sg-play` ×16) nobody noticed | | Security | ✅ key discipline, allowlists — **and** one disclosed key-in-history incident, handled well | | Backups | ⚠️ **thinnest topic in the corpus** — 151 mentions, no doctrine | | Consistency | 🟡 100% banner compliance beside 39% import alignment; four stale sources of truth | | Explainability | ✅ reality docs, provenance — **and** no evals anywhere | | Documentation | ✅ 1.27M words with a governing rule — **and** a README describing a repo that does not exist | **Publish that table.** A site teaching NFRs from a corpus that visibly practises *and* visibly lapses is credible in a way no clean-room guide can be — and every lapse is already documented in a sibling pack with the evidence. --- ## 5. The numbers | | | |---|---| | **Densities** | documentation 651 · budget 410 · observability 393 · serverless 347 · unit test 279 · CI 273 · backup 151 · scalability 104 · resilience 100 · reliability 74 · explainability 65 · PM 31 · "non-functional" 26 | | **The definition** | one sentence in the villagers brief, listing **eight NFRs verbatim** | | **Owned disciplines** | testing · CI · documentation/reality · IFD · budgets · PM · resilience patterns | | **The methodology** | IFD — versioned guides (`v1.2.1`), written *"for LLMs assisting with IFD-based development"* | | **This pack** | 7 documents · manifest of 26 rows · every path verified | --- ## 6. Build order 1. **`/map/`** — the topic map (`01__`). The hub's reason to exist: one page an agent reads to know where everything lives. 2. **`/memory/`** — the sites-as-agent-memory thesis (`02__`). The page that explains the network. 3. **`/testing/` and `/ci/`** — the two strongest owned disciplines, with the measured numbers. 4. **`/documentation/`** — the reality-document system: *"if the reality document doesn't list it, it does not exist… briefs are aspirations, not facts."* 5. **`/budgets/` and `/pm/`** — pre-approve-the-ladder, profitability-first, the brief/debrief system. 6. **`/scorecard/`** — §4's table, unsoftened, linked to the sibling packs' evidence. 7. **`/shipped/`** — what this site asserts vs what the estate demonstrably does. --- This document is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__01__the-topic-map.md ============================================================================== # 01 — The topic map The hub's core artefact: one page an agent reads to know **what each NFR means here, where the canonical material lives, and what the estate actually does about it.** Each entry: the position in one line, the best source, the owner site. --- ## Testing — OWNED **Position:** *"No mocks. No patches."* Assert on contracts, not implementation; real Chromium gated on an env var, skipping cleanly; deploy **via pytest**, numbered tests run top-down. Affordable because `Type_Safe` objects are cheap to compose in memory — the type system and the testing philosophy are one decision. **Evidence:** 4,785 tests passing in 81 seconds; 799 test files; the four `tests/ci/` structural guards (one broken — the honest footnote). **Canonical:** `library/guides/v3.1.1__testing_guidance.md` + the sg-compute measurements. **Cross-link:** `coding.sgit.ai` owns the conventions; this site owns the philosophy and the numbers. ## CI pipelines — OWNED **Position:** the pipeline is a *verifier*, not a runner: native per-arch builds, **push by digest only**, integration-test the pre-tag image, only then assemble the manifest; AMIs are baked, **relaunched, and re-verified** before tagging `healthy`; versions auto-increment from one repo-root file. **Evidence:** the 36,815-byte `ci-pipeline.yml`, the two-phase `bake-ami.yml` — and the lapse: that workflow invokes a binary (`sg-play`) defined nowhere, sixteen times. **Owner of the artefacts:** `sg-compute.sgit.ai`; this site owns the *pattern* write-up. ## Documentation — OWNED **Position:** documentation is a **truth system, not prose**. The reality-document rule: *"**If the reality document doesn't list it, it does not exist.** Briefs are aspirations, not facts."* Proposed features labelled `PROPOSED`. Every page has a markdown twin. Day-indexes, debriefs with good-failure/bad-failure classification, CC BY footer on ~1,100+ files. **Evidence:** 1.27M words in one repo alone; 72,339 words of reality docs across 12 domains — and the lapse: the reality index itself 41 versions stale, and a README describing a repo that does not exist. **The page writes itself: the system, the rule, and the two places it failed.** ## IFD — Iterative Flow Development — OWNED **Position:** the estate's named methodology — *"rapid software development using AI assistance while maintaining engineering rigor… centred on **preserving developer flow state**."* Versioned guides (intro, testing, versioning) at `v1.2.x`, written explicitly *"for LLMs assisting with IFD-based development."* Minor versions are the Explorer team's output unit; majors are Villager releases. **Canonical:** `library/guides/development/ifd/`. **Nobody outside the estate has ever seen this methodology written up.** It is this site's most original owned asset. ## Budgets and finance — OWNED (the discipline, not the business) **Position:** three pillars. **Profitability-first:** *"until we know the traction… financial projections and future predictions are made-up, because we do not yet have the data."* **Pre-approve the ladder:** approve the overrun positions at approval time — 1.5×, 2×, 5× — *"with the stopping point named while it is still cheap to name"*, because the research says *"the person who approved the first million is, on the evidence, the worst available decider on the second."* Plus the novel acceptance: *"the things **not** done because this project was funded should themselves be accepted, with an owner."* **Budget-on-the-step:** budgets attached to workflow steps, not projects. **Boundary:** the *discipline* is publishable; the estate's own figures are Tier-3 everywhere. **Cross-link:** token-spend-as-engineering-problem is `wardley-maps.`'s line. ## Project management — OWNED **Position:** *"the project manager is where the register becomes work"* — a register per project, net score, erosion, the missing reference class. Plus the estate's own PM system as the exhibit: briefs with acceptance criteria, cross-team reviews, day-indexes, numbered asks (N1…) and tasks (T1…), version-prefixed filenames, session handover guides. **The brief/debrief system is itself the estate's PM methodology, demonstrated across ~4,000 documents.** ## Resilience — OWNED (patterns); artefacts live in `sg-compute.` **Position:** design for the failure you had, not the one you imagine. The watchdog (`os._exit(2)` through a deadlock, GIL reasoning in the source), the two-phase health poller, halt-means-terminate, idle reconciliation, build-time guards each citing a production incident. **Every resilience mechanism in the estate traces to a named incident** — that is the doctrine, and it is nowhere written down. ## Security — LINKS The NFR posture only: audit-before-the-key · deny-by-default allowlists · key-prefix validation · no-credentials-in-git with CI gates · the disclosed-incident norm. Domains belong to `pki.` `nhi.` `sg-sentinel.` `standards.`; the LLM boundary to `llms.`. ## Serverless and scalability — LINKS The requirement framing only — *"no servers running when there is no traffic, with a defined cost to start"* — and the honesty note that throughput is argued, never measured. Platform: `sg-compute.`. ## Explainability — LINKS The engineering habit only: **computed, not claimed** · unanswered-is-an-output · estimates rendered with `~`. The grounding ladder belongs to `risks.`/`standards.`; provenance to `llms.`. ## Architecture — LINKS Responsibility boundaries as a requirement class (*"X is the ONLY class that…"*), single-source-of-truth by structure. Conventions: `coding.`. ## Backups — ⚠️ THE GAP 151 mentions, **no doctrine.** The one estate with irreversible publishing (frozen vaults), a keys-vault design, and no written backup/recovery discipline beyond the keys-vault `RECOVERY.md` proposal. **The site should say so — it is the eighth item on the corpus's own NFR list and the only one with no material.** --- This document is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__02__the-memory-thesis.md ============================================================================== # 02 — The memory thesis: these sites are agent memory, done better The commissioning message contains the clearest statement of the network's purpose ever made: > *"the idea of these sites is to provide good briefs for agents to learn about how we work and think — **these sites are a more evolved and focused version of what is usually called LLM memory**."* This document develops that into the page it deserves — and recommends it also appear on `sgit.ai`, because it explains all fourteen sites at once. --- ## 1. The claim What products call "memory" today is an accumulation: embeddings of past conversations, retrieved by similarity, private to one vendor's silo, unversioned, uncurated, and invisible to the person it describes. The `*.sgit.ai` network is the same function built as **publishing**: | | Conventional LLM memory | These sites | |---|---|---| | **Content** | accumulated transcripts | **curated briefs** — written, reviewed, pruned | | **Retrieval** | similarity search, opaque | **addressable URLs** + `llms.txt` + `llms-full.txt` | | **Versioning** | none | git, version-prefixed filenames, dated pages | | **Truthfulness** | whatever was said | the reality-document rule: *"briefs are aspirations, not facts"* | | **Portability** | locked to one vendor | **any agent, any vendor, one HTTP GET** | | **Licence** | unclear | **CC BY 4.0, explicit, irrevocable** | | **Inspectable by the human** | rarely | it is a website — read your own memory | | **Shared across agents** | no | yes — one memory, N agents, no sync | The one-line version for the front page: **memory you can read, cite, version, license and hand to any agent — because it is a website.** ## 2. The evidence that this was already the design The estate has been building toward this explicitly, before the sentence was said: - Guides with **`for_llms` in the filename** — `v3.1.1__for_llms__type_safe__testing_guidance.md`, `v3.63.4__for_llms__python_formatting_guide.md` — and the IFD guide's purpose line: *"Complete reference **for LLMs** assisting with IFD-based development."* - **The markdown twin at every URL**, *"so a traversing agent never has to parse HTML."* - **The agent-access report's finding** — *"the audience is disproportionately agents… it can read the map and cannot walk it"* — which is a memory-retrieval failure diagnosed in exactly those terms. - The house pattern itself: `/llms.txt` as the whole agent surface, single-file concatenation, numbered asks an agent can act on. ## 3. What "more evolved" concretely means — the disciplines this site teaches The memory thesis is an NFR story, which is why this page lives here: a memory is only better than a transcript if somebody maintains its non-functional requirements. 1. **Curation is the villagers' work applied to knowledge.** Someone owns each site, prunes it, and keeps it true — the reality-doc discipline as memory hygiene. 2. **Versioning makes memory correctable** — the thing a frozen vault and a transcript both cannot do. 3. **The do-not-publish tiers are memory's security model.** Every pack in this series ships one; a memory without redaction discipline leaks. 4. **`shipped/` pages are memory's calibration.** A memory that cannot distinguish designed from built teaches agents to overclaim — the single most common failure this session found across the estate's stale artefacts. 5. **Deconfliction is memory's normalisation.** One canonical copy, siblings link — the skills pack's diverging duplicate is what happens otherwise. ## 4. The honest limits - **Staleness is the failure mode**, and the estate has already demonstrated it four times (capabilities.json, the reality index, the README, `sg_compute/version`). A memory-site network needs the generate-or-date rule everywhere. - **Curated memory is opinionated memory.** These sites teach *how we think* — an agent trained on them inherits the positions, including the wrong ones. The tensions-published-unresolved convention is the mitigation, and it is a real one. - **Fourteen sites is itself a retrieval problem.** Without the topic map (`01__`) and cross-site deconflicts, the network reproduces the discovery failure the agent-access report found in one site. --- This document is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__03__the-owned-disciplines.md ============================================================================== # 03 — The owned disciplines, in depth The five topic areas this site owns outright, each with its position, its best verbatim material, and its honest counter-evidence. These become the site's core pages; `01__` gave each a summary, this gives the substance. --- ## 1. Testing **The four non-negotiables, verbatim:** > 1. **No mocks. No patches.** Use `register_playwright_service__in_memory()` and `in_memory_stack`-style composition. > 2. **Assert on contracts** — schemas, status codes, persisted artefacts — not implementation details. > 3. **Real Chromium for integration tests.** Gate on `SG_PLAYWRIGHT__CHROMIUM_EXECUTABLE`; skip cleanly when absent. > 4. **Deploy-via-pytest.** Deploy tests are numbered (`test_1__create_lambda`, `test_2__invoke__health_info`, …) and run top-down. **Why no-mocks is affordable here and not elsewhere:** `Type_Safe` objects are cheap to construct, and every service has an in-memory composition path — so the real thing is as easy to instantiate as a mock would be. The philosophy and the type system are one decision. The proof: **4,785 tests in 81 seconds.** **The structural-guard pattern** — CI tests that enforce architecture rules rather than behaviour (`object = None` banned, UI-in-the-wheel, component snapshots) — with the rule that makes the set healthy: **every guard encodes a rule that was violated at least once.** Grown from incidents, not checklists. **Counter-evidence to publish:** CI runs 67.4% of collectible tests; six import-level breakages hide in the un-run third; one guard has never worked; three tests fail on date arithmetic. All measured, all in the sg-compute pack. ## 2. CI pipelines The pattern, distilled from the 36,815-byte pipeline: **build once, verify before naming.** Native per-arch builds → **push by digest only** → integration-test the *pre-tag* image → only then combine digests into a manifest and tag. Nothing gets a name until it has passed as an anonymous digest. Same shape one level up: the AMI bake **relaunches from the baked image and re-verifies** before tagging `healthy` — the artefact must prove itself twice, once as built and once as booted. Versioning: one repo-root `version` file, auto-incremented by branch policy (dev bumps minor, main bumps major), read at runtime, used as the image tag. **One source, many consumers, no drift** — and the counter-evidence is `sg_compute/version`, a second version file read by nothing. ## 3. Documentation and the reality system The governing rule, which is the estate's single best NFR idea: > *"**If the reality document doesn't list it, it does not exist.** … **Briefs are aspirations, not facts.**"* With its supporting disciplines: `PROPOSED` labels on unbuilt features · the markdown twin · day-indexes closing *"all documents are em-dash-free and released under CC BY 4.0"* · debriefs classifying failures as good-failure/bad-failure · the session-handover guide (*"Don't improvise"*). **The full honest story:** the system exists, 72,339 words across 12 domains — and the index was 41 versions stale while the README described a repository that does not exist. **The rule is right and enforcement is manual**, which is the same finding as the coding pack's: discipline without tooling is real but fragile. The page should propose the fix the estate would recognise: a reality-doc freshness check in CI, exactly like the licence audit. ## 4. Budgets and finance Three pillars, all publishable as discipline (never the estate's own figures): **Profitability-first** — *"until we know the traction, the product lines, the services, the cost lines, and what users actually buy, financial projections and future predictions are made-up… The path is to discover the market and ship quasi-daily; investment only accelerates the path to profitability."* The Explorer-phase corollary: do not build Town-Planner financial artefacts on hypothesis. **Pre-approve the ladder** — the overrun positions approved at approval time, the kill point *"named while it is still cheap to name"*, value milestones and sponsor probabilities recorded as a calibration record, grounded in the escalation literature (*"the person who approved the first million is, on the evidence, the worst available decider on the second"* — with the preregistered finding that a public conditional pledge makes stopping *raise* trust). Plus the original move: **opportunity cost as a first-class acceptance, with an owner.** **Budget-on-the-step** — budgets attached to workflow steps rather than projects, so containment is structural. ## 5. Project management Two layers. **The position:** *"the project manager is where the register becomes work"* — a register per project, net score, erosion, the missing reference class; risk management and project management as one discipline with two vocabularies. **The demonstrated system:** the estate's own way of working *is* its PM methodology — briefs with acceptance criteria (the naming brief's nine, scored in the sg-compute pack), cross-team reviews the same day, numbered asks and tasks, version-prefixed filenames sorting chronologically, handover guides, and the good-failure/bad-failure debrief convention. Roughly 4,000 documents of it. **No page anywhere describes it as a system.** Writing that page — "how a brief becomes work here" — is this site's most original PM contribution, and it doubles as onboarding for every new agent session. --- This document is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__04__site-architecture.md ============================================================================== # 04 — Site Architecture ## The house pattern Copy `pki.sgit.ai`; add `/llms-full.txt`. This site's character: **the hub** — it owns the topic map and the homeless disciplines, and links out relentlessly. Resist the pull to duplicate; the deconflict table in `01__` is the contract. ## Page by page **`/`** — NFRs expanded in sentence one; the villagers definition as the epigraph (*"version control, reliability, resilience, security, backups, consistency, explainability, and documentation"*); the hub claim; and the scorecard teaser. **`/map/`** — `01__` in full. **Build first.** One page an agent reads to know where everything lives. Machine-readable twin required. **`/memory/`** — `02__`. The thesis page — and propose it to `sgit.ai` for the hub as well. **`/testing/` · `/ci/` · `/documentation/` · `/budgets/` · `/pm/` · `/resilience/` · `/ifd/`** — the owned disciplines from `03__`, one page each, each closing with its counter-evidence and a link to the sibling pack holding the measurements. **`/scorecard/`** — the estate measured against its own eight-NFR list (`00__` §4). Links each cell to evidence. Regenerate per release. **`/backups/`** — the honest stub: the one NFR on the list with no doctrine. State it, date it, and let the stub shame the gap closed. **`/shipped/`** — what this site asserts vs what the estate demonstrably does; the generate-or-date rule for every number. **`/network/` · `/admin/`** — house pattern; build order published unresolved. ## Generated, not written The topic densities · the scorecard · every measurement quoted from sibling packs (link, do not restate) · the map's link targets. This site quotes more numbers than any other in the network — it has the most to lose from drift. --- This document is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__05__boundaries-and-licensing.md ============================================================================== # 05 — Boundaries and licensing ## 1. Licensing **CC BY 4.0** site-wide, stamped and gated with `licence-audit.py --check`. Quoted code is Apache-2.0 — retain notices. The IFD guides and `for_llms` documents are in-repo Apache-2.0 material; republishing them here as pages is fine with attribution, but **keep the repo canonical and generate** — the second-source-of-truth rule applies to methodology guides exactly as it did to the `sg.llm` reference. ## 2. Do not publish - **Any of the estate's own financial figures.** The budget *discipline* is Tier-0; the numbers around it are Tier-3 everywhere — `library/alchemist/materials/` (whole tree), the payments-platform briefs (pricing, margins, provider detail), the commercial-model briefs (rates, delivery location), the investor briefs. - `team/roles/grc/reviews/02/19/` — named private individual. - The appsec review classified as *"an attack roadmap for live code."* - The villagers brief's **market statistics** — the brief itself flags them as loosely attributed; re-source or soften (same ruling as three prior packs). - Live hostnames, the AWS account ID, stack FQDNs — the sg-compute redaction list applies wherever its material is quoted. ## 3. Network boundaries — the hub's contract This site **owns**: the topic map · testing · CI · documentation/reality · IFD · budgets-as-discipline · PM · resilience patterns · the memory thesis · the scorecard. This site **links and never restates**: security domains (`pki.` `nhi.` `sg-sentinel.` `llms.`) · serverless/platform (`sg-compute.`) · conventions (`coding.`) · grounding/explainability (`risks.` `standards.`) · the villagers *market* argument (`open-source.`) · the lifecycle (`wardley-maps.`) · skills (`skills.`). **One rule above the table:** when a sibling pack has measured something, this site links the measurement — it never re-measures and never paraphrases numbers. The hub that drifts from its spokes is worse than no hub. ## 4. House style Every discipline page ends with its counter-evidence · every number generated or dated · the eight-NFR list quoted verbatim wherever the scope is stated · -ise, no em-dashes. --- This document is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__06__gaps-and-open-questions.md ============================================================================== # 06 — Gaps, open questions and honest tensions --- ## 1. Build-fresh items | # | Item | Why | |---|---|---| | **G1** | **The backups doctrine** | The only NFR on the corpus's own list with no material — in the estate with irreversible publishing and a keys-vault design. Write it or publish the stub | | **G2** | **"How a brief becomes work here"** | The estate's PM system exists in ~4,000 documents and is described nowhere. The most original PM page available, and onboarding for every agent session | | **G3** | **The resilience doctrine** | Every mechanism traces to a named incident; the pattern ("design for the failure you had") is unwritten | | **G4** | **The reality-doc freshness check** | The best documentation rule in the estate, enforced manually, already failed twice. A CI check like the licence audit | | **G5** | **The scorecard generator** | `00__` §4's table, regenerated per release, cells linked to evidence | | **G6** | **The memory-thesis page for `sgit.ai`** | The network's purpose statement, currently living in a chat message | ## 2. Open questions | # | Question | Where it stands | |---|---|---| | **Q1** | **Is a hub site worth its drift risk?** Fourteen spokes, all moving | The mitigations are the link-never-restate rule and generated numbers. If those hold, yes; if not, the map alone (one page) is the fallback | | **Q2** | **Where does "consistency" live?** On the eight-NFR list, but split across `coding.` (conventions), this site (sources of truth), and `skills.` (drift) | Recommend: this site owns the *requirement*, the map points at the three homes | | **Q3** | **Should IFD be published at all?** It is the estate's competitive methodology — and the memory thesis says publish | The open-source position (*"the moat is a rate, not a wall"*) answers yes; say so explicitly rather than assuming | | **Q4** | **Do the budget disciplines survive contact with real figures?** Pre-approve-the-ladder is research-grounded and untested here | Publish as discipline with the research citations; mark the estate's own use as pending | | **Q5** | **Is "backups" actually the gap, or is the vault model the answer?** Vault versioning + escrow may *be* the backup doctrine, unnamed | An hour of thinking, then either the doctrine page or the honest stub | ## 3. Honest tensions 1. **A site about NFRs, from an estate whose own scorecard is mixed.** One green column would be dishonest; the mixed table is the credibility. Every lapse cited is already published in a sibling pack — this site adds no new exposure, only assembly. 2. **The hub teaches what the spokes measured.** Its value is curation, and curation is the first thing to rot. The generate-or-date rule is load-bearing here more than anywhere. 3. **The memory thesis makes every site's flaws teachable.** If these sites are agent memory, then a stale page is a false memory — which raises the stakes of the staleness failures the estate has already had four of. The thesis and the freshness discipline arrive together or not at all. 4. **"For agents to learn how we work" cuts both ways.** An agent that learns the no-mocks rule also learns the 67.4%-CI-coverage reality. Teaching the practice honestly means teaching the gap between rule and practice — which is, in the end, what an NFR site is *for*. ## 4. Loose ends worth an hour each - Decide Q5 (backups vs vault-model) before writing G1. - Extract the eight-NFR sentence's history — did the list ever appear earlier than the villagers brief? - Check `library/guides/development/` for guides not yet surfaced in any pack (the `code-formating` and `dependencies` folders were only skimmed). - Confirm the IFD guides' current version against actual practice — they are `v1.2.x` in a repo at `v0.33.x`, two version schemes side by side. - Count the day-indexes and debriefs for the G2 page — the PM system's scale is part of its argument. --- This document is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). ============================================================================== == briefs/v0.33.62__nfrs-brief-pack__LICENCE.md ============================================================================== # Licence ## This pack Everything in this brief pack — the seven numbered documents, `09__source-manifest.csv`, this file and `README.md` — is released under the **Creative Commons Attribution 4.0 International licence (CC BY 4.0)**. Copyright (c) 2026 Dinis Cruz Licensed under CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/ Attribution: **Dinis Cruz**, with AI co-authorship (Claude, Anthropic). ## The site this pack commissions **The entire content of `nfrs.sgit.ai`** is CC BY 4.0, consistent with the network. Stamp every markdown document; gate with `licence-audit.py --check`. Quoted code and in-repo guides (IFD, testing, formatting) are **Apache-2.0** — retain the notice, keep the repo canonical, and generate rather than copy: the second-source-of-truth rule applies to methodology guides exactly as it did to API references. ## Do not publish - **Any of the estate's own financial figures.** The budget discipline is Tier-0; the numbers are Tier-3 everywhere: `library/alchemist/materials/` (whole tree), the payments-platform briefs, the commercial-model briefs. - `team/roles/grc/reviews/02/19/` — named private individual. - The villagers brief's market statistics — flagged by the brief itself as recycled vendor commentary; re-source or soften. - Live hostnames, the AWS account ID, stack FQDNs — the sg-compute redaction list applies wherever its material is quoted. ## The hub's accuracy rule This site quotes more sibling measurements than any other in the network. **Link the measurement, never restate it; generate or date every number.** A hub that drifts from its spokes is worse than no hub. --- This file is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0).