# Architecture Review — 2026-08-26 An independent review of this repository, `CrystalCore.OS-the-Crystal- Architecture-Archive`, run at the maintainer's request to keep the project's architecture sound. This is a fresh pass, not a follow-up to any prior AI-generated review — no such review of this specific repository existed on disk before this file. Scope is this repository only: its own labelling discipline, its Belt-Three and locked-name compliance, and whether its own numbered, "canonical" documents could mislead a reader about what is current. **Snapshot:** local clone of this repository at `9a1383b` (branch `main`), read in full — every top-level file and all thirteen `knowledge-base/` documents. No other repository in the portfolio was cloned for this pass; where a claim below needs one (Finding 1), that limit is stated rather than guessed past, per this archive's own `12-CONTRIBUTING.md` (*"verify the new reality directly against the repository — not against a chat session's memory"*). **Evidence tiers**, matching the scheme this repository already uses in `FULL-REVIEW-2026-07-28.md`: - **Tier A — executed.** Not used in this review; nothing here runs. - **Tier B — content-addressed / mechanically resolved.** Grep results, exact quotes, line numbers. - **Tier C — read.** A file was read and is cited or quoted. All findings below are Tier B/C. This is a documentation review of a documentation repository — there is no code to execute. --- ## What this repository actually is This is **not** an archive in the sense of frozen, superseded material — it is the portfolio's actively maintained **system ledger**: a running `STATUS.md` (last touched 2026-08-20), a thirteen-document `knowledge-base/` reconstruction, and a self-correction log (`11-CORRECTIONS.md`) that the repository's own README and `12-CONTRIBUTING.md` describe as an ongoing discipline, not a closed project. `README.md` states this plainly: *"This repository holds no application code. It documents the other [repositories]"* — and unlike most of the portfolio's other repositories, `STATUS.md` (line 10) counts this one among the **6 public living** repositories, not the 6 archived ones. The name "Archive" describes its *function* (a ledger of record) more than its *state* (it is not read-only, not frozen, and takes pull requests — PR #38 merged the same day as its last `STATUS.md` update). That framing matters for how to read the findings below. The risk this review was asked to check for — old material being mistaken for current canon — does not show up here as a frozen legacy document sitting unlabelled next to a numbered one (the pattern the project's own `proposal` repo convention warns about). It shows up as a subtler version of the same failure mode: **numbered, "canonical" documents (`06-COMPONENTS.md`, `09-GLOSSARY.md`, `02-REPOSITORY-MAP.md`) that were written once, on 2026-07-24, and describe a fact that has since moved — with the correction, when it exists at all, buried in `11-CORRECTIONS.md` rather than reflected where a reader would actually look.** This repository's own documentation template (Statement / Evidence / Historical Notes / Cross References) is designed to prevent exactly this, by carrying a Historical Notes field on every entry — the findings below are cases where that field was not used when it needed to be. --- ## Summary This repository is unusually disciplined for what it is: every factual claim follows a fixed Statement/Evidence/Historical Notes/Cross-References template, evidence tiers are used consistently, and the project holds itself to its own Belt-Three law harder than most documentation repositories hold themselves to anything (`00-INDEX.md` §2, `12-CONTRIBUTING.md`). The Indigenous Data Sovereignty boundary around "Songline" is handled correctly everywhere it was checked. The one substantive problem found is internal staleness in exactly the class of file this project's own conventions say must be trustworthy: the numbered `knowledge-base/` documents assert, as current canon, a companion name ("Lumina") that this archive's *own* later-dated evidence (Part 20 of `11-CORRECTIONS.md`, filed 2026-08-05; `00-INDEX.md`'s 2026-08-08 settled-state note) contradicts. A second, smaller instance of the same pattern affects the bus validator's name. Both are documented below with a dated flag added to the glossary in this same change, following this archive's own prescribed correction process rather than a silent rewrite. ## Strengths - **The evidence discipline is real, not decorative.** Every `knowledge-base/` Statement cites a file, line, commit, or test count a reader can independently check (`00-INDEX.md` §2, the "four-question test"). Spot-checking several citations (e.g. `06-COMPONENTS.md`'s Lumina entry citing `vision/apps/lumina/crystalcore/__init__.py: __version__ = "0.7.0"`) found them precise enough to verify, which is the actual bar this repository sets for itself. - **The Indigenous Data Sovereignty boundary is enforced, not just stated.** Every occurrence of "Songline" found in this repository is either (a) a historical fact about an early prototype's actual old name, correctly labelled as retired (`knowledge-base/09-GLOSSARY.md:74`: *"Songline — Reserved exclusively for Aboriginal culture"*; `knowledge-base/07-HISTORY.md:13`: the 2026-07-21 rename sweep that reserved the word), or (b) a quoted violation found in *inbound* material and explicitly flagged as such rather than adopted — `STATUS.md:101` and `AI-BOOT-PANEL-2026-08-08.md:80-82,218,232,500` all identify "Songline" used as a component name by outside AI responses and label it a naming-law breach on sight. No file in this repository proposes or normalises "Songline" as a component name. This is the check this review was specifically asked to run, and it passes cleanly. - **Locked names are not redefined.** `TerAustralis Incognita`, `CrystalVision`, and `CrystalCore.Lattice` appear only in ways consistent with Constitution §1 — e.g. `AI-BOOT-PANEL-2026-08-08.md:657`: *"CrystalCore.Lattice is a Constitution §1 locked name and is designed, not built"* — and the same discipline is applied to received/inbound documents that get it wrong (`GROK-REX-MYTHOS-EXPORT-2026-08-07.md`, Finding 1: a Grok-authored line claiming "Lattice integrity: 100%" is flagged, not repeated as fact). - **Received AI-generated material is handled correctly.** The Grok "Rex" mythos export, the Weaver Nexus handoff, and the AI Boot Panel are each filed with an explicit "authority: none" framing, a reception record ahead of the text, and — critically — a policy of checking claims against actual commits rather than trusting them (`GROK-REX-MYTHOS-EXPORT-2026-08-07.md`: *"a search of all 502 commits ... found none of them"*). This is the Incognita Rule applied correctly to the exact kind of material — confident, dreamed, fluent — it exists to catch. - **The repository is honest about its own staleness in general terms.** `00-INDEX.md`'s scope note and `SURVEYED.md` both state plainly that the knowledge base was built once, from six repositories, and that the portfolio has since grown; `check-freshness.py` exists precisely to make drift visible rather than silent. The findings below are about that general honesty not yet reaching two specific entries, not about the absence of the mechanism. ## Findings ### Finding 1 — HIGH: the companion's canonical name is asserted stale, and the archive's own later evidence contradicts it **Statement.** `knowledge-base/09-GLOSSARY.md`, `06-COMPONENTS.md`, and `02-REPOSITORY-MAP.md` all currently assert that the companion's name is **Lumina**, and that **Clementine** "renamed away 2026-07-21" and is retained only in a frozen legacy repository. This is presented as current canon in three of this archive's numbered, "start here" files, not flagged as historical-only. **Evidence.** - `knowledge-base/09-GLOSSARY.md:15-23` (before this review's edit): *"Clementine — Two unrelated senses, both renamed away 2026-07-21. (1) The original name for the AI companion persona ... renamed Lumina in the current canon. ... Only the frozen `crystalcore` repository still uses 'Clementine' natively."* - `knowledge-base/09-GLOSSARY.md:53-54`: *"Lumina — The AI companion. Formerly 'Clementine' ... renamed 2026-07-21."* - `knowledge-base/06-COMPONENTS.md:7-14`: a full "Lumina" component entry, **Status: Implemented**, with no mention of Clementine as a live alternative. - `knowledge-base/02-REPOSITORY-MAP.md:250`: *"persona sweep — Clementine (companion) → Lumina."* **Against this archive's own later evidence:** - `knowledge-base/11-CORRECTIONS.md` Part 17 (filed 2026-07-30, line 1726): describes the Clementine/companion name collision as *"the second naming decision waiting on the maintainer, alongside Lumina"* — i.e., as of 2026-07-30, this archive itself still treated the choice as **open**, not settled in Lumina's favour. - `knowledge-base/11-CORRECTIONS.md` Part 20 (filed 2026-08-05, lines 1903-1908): *"`The-Crystal-Vision` now carries `Clementine---Local-Soveriegn-Edge-AGI`. ... each clone's README still opens with its old identity ('This is The Crystal Vision — codex site + Clementine sovereign companion app')."* - `knowledge-base/00-INDEX.md:143-147` (2026-08-08, "Settled state"): lists **`Clementine---Local-Soveriegn-Edge-AGI`** by name as one of the six repositories now public and checkable. None of this is presented anywhere as a reversal of the Lumina naming — it sits in the correction log and the index's visibility table, never folded back into the Statement in `09-GLOSSARY.md` or `06-COMPONENTS.md` that a reader would actually consult to answer "what is the companion called." A reader following this archive's own suggested navigation order (`00-INDEX.md` §3: read `06-COMPONENTS.md` before `11-CORRECTIONS.md`) would come away certain of exactly the wrong answer, sourced to this archive's own numbered canon. **Severity and why.** High, not Medium: this is the project's most frequently-cited component name, it appears in a public-facing repo slug the project itself now uses, and the archive's own historical notes show the "correct" answer flipped at least twice (Clementine → Lumina on 2026-07-21, per `07-HISTORY.md:13`; then some signal of Clementine live again by 2026-08-05/08). A reader has no way to tell from the numbered docs alone that the ground moved. **Recommendation.** A future session with access to `TerAustralis-Incognita-Code` should re-verify the companion's current name directly against that repository (not against this file, per `12-CONTRIBUTING.md`'s own rule), then update the Statement in `09-GLOSSARY.md`, `06-COMPONENTS.md`, `02-REPOSITORY-MAP.md`, and the mentions in `01-SYSTEM-OVERVIEW.md` and `03-ARCHITECTURE.md` consistently, with a Historical Notes line recording both renames. This review does not attempt that rewrite — it cannot verify the live repository from here. As an interim, small, safe measure, this same change adds a dated flag to the two `09-GLOSSARY.md` entries pointing here, following this archive's own prescribed process (`12-CONTRIBUTING.md`: *"add a dated note ... rather than deleting the old claim outright"*). ### Finding 2 — MEDIUM: "Dreamline Narrator" does not match either current locked name for the bus validator, and has no glossary entry of its own **Statement.** Two files describe the 2026-07-21 rename as producing "Starline Weaver + Dreamline Narrator" as the two successor names for the old Songline Bus. "Starline Weaver" checks out consistently elsewhere in this archive. "Dreamline Narrator" does not appear as its own `09-GLOSSARY.md` entry anywhere, and does not match either of the locked pair the project currently uses for these two distinct roles — a message-truth validator and a memory-transport traveller, kept as two separate names in current usage rather than one compound. **Evidence.** - `knowledge-base/07-HISTORY.md:13`: *"'Songline Bus → Starline Weaver + Dreamline Narrator'."* - `knowledge-base/02-REPOSITORY-MAP.md:251`: repeats the same phrase. - `knowledge-base/09-GLOSSARY.md`: has entries for **Starline Weaver** (line 88) and **Starlines / Dreamlines** (line 92, the mythic framing) but none for "Dreamline Narrator" as a component — a reader cannot look the term up in the one document built to define it. **Severity and why.** Medium, not High: this name appears in only two places, both historical-chronology entries rather than a component's own Statement, so the practical exposure is lower than Finding 1. It is the same class of defect — a name frozen at the moment of a 2026-07-24 reconstruction, never re-squared against later naming decisions. **Recommendation.** Whoever resolves Finding 1 should also confirm the correct current pair of names for the validator and transport roles that the 2026-07-21 rename produced, add a `09-GLOSSARY.md` entry for whichever term is current, and correct or annotate the two citations above. ### Finding 3 — LOW: the portfolio's repository count drifts across this repository's own three "start here" documents **Statement.** `README.md`, `00-INDEX.md`, and `STATUS.md` — the three files this repository's own README table points a new reader to first — give three different counts of the portfolio's size, with no cross-reference reconciling them. **Evidence.** - `README.md:3-4`: *"This repository holds no application code. It documents the other ten [repositories]."* (implies eleven total, including this one) - `knowledge-base/00-INDEX.md:4`: *"The portfolio has seventeen (2026-08-08, late ...)"* - `STATUS.md:10-11`: *"Constellation: 19 repositories under CrystalArchitect (6 public living, 7 private living, 6 archived)."* (dated 2026-08-20, the most recent of the three) **Severity and why.** Low: `STATUS.md` is explicit that it is the most current count and explains its own provenance, and a careful reader following the README's link to `STATUS.md` will get the right number. The risk is narrow — a reader who stops at the README's own opening paragraph, without following its own link, walks away with a number that has been wrong since well before this repository's most recent commit. **Recommendation.** Point `README.md`'s repository count at `STATUS.md` by reference (*"documents the other repositories — see `STATUS.md` for the current count"*) rather than stating a number in the README itself, so this can't drift again the same way. ## Open questions for a human / architect 1. **What is the companion actually called right now?** This review could not check `TerAustralis-Incognita-Code` directly. Finding 1's evidence trail is entirely internal to this repository and is suggestive, not conclusive — it shows this archive's own confidence in "Lumina" eroding over time, not a confirmed final answer. Someone with access to that repository at its current HEAD should settle this and update the four files named in Finding 1's recommendation. 2. **Is "Dreamline Narrator" a typo for one of the two current locked names, or a third, now-abandoned name from between the 2026-07-21 sweep and whatever produced the current pair?** This archive alone cannot distinguish those two possibilities. 3. **Should the numbered `knowledge-base/01-10` documents be re-reconstructed wholesale**, now that the portfolio has grown from the six repositories they were built from to nineteen, or should the current model — a frozen reconstruction plus an append-only `11-CORRECTIONS.md` — continue indefinitely? The append-only model is transparent about what it is, but Finding 1 shows it can let a contradiction sit unresolved for weeks in the exact files a reader is directed to first. That is a scaling question for the maintainer, not something this review can resolve by itself. 4. **Does the project want a lightweight rule that a `09-GLOSSARY.md` entry may not be treated as reaffirmed just because it was written once** — e.g., a "last confirmed" date on load-bearing entries like component names, separate from the entry's original write date? That would have caught Finding 1 mechanically rather than requiring a manual review to notice it. --- *This review changed one file besides itself: `knowledge-base/09-GLOSSARY.md`, where dated flags were added to the Clementine and Lumina entries per this archive's own `12-CONTRIBUTING.md` process. No other content in this repository was rewritten — per this review's own brief, correctness fixes requiring re-verification against other repositories are left to a future pass with access to them, listed above as open questions.*