Appearance
Open questions — the review agenda
These are the things the model does not yet decide. They live here, and not on the model pages, for one reason: a model page has to be true now, so that anyone — a person or a machine — can build on it without checking whether each sentence is a rule or a musing. A question is by definition not yet true, so it belongs on its own surface with its own lifecycle.
Nothing here retires without being ruled on. When one is settled, the answer is written into the model page and the question moves to decided with its date and reasoning. That is the whole loop.
Where the review stands
All twenty-five numbered questions are ruled — fourteen at round 1 (4 August 2026), the rest plus the three preliminary calls at round 2 (10 August 2026). Every ruling lives on the decided page with its reasoning, so nothing gets re-argued from memory.
Beyond the numbered questions, round 2 also settled five standing rules:
| Ruled | In one line |
|---|---|
| This surface wins | when a spec and this decision surface disagree, this surface wins |
| Campaign is a container | a campaign holds pieces; it is never a lens you classify against |
| The grouping kinds | campaign · brand-identity collection · project · event/season |
| Vendor work is first-class | studio and agency case studies are collected like brand work |
| Scope is wide | offline placement, geography & language, and reception are all in |
Closed threads keep their old names for traceability: the standing inter-entity edge thread (companies relate — ruled at question 4) · the content_region thread (a moment is video-only — question 11) · the Signal/Temporal thread (timing serves the pillars — question 13).
What is awaiting your mark, in order
Everything below is waiting on you, and nothing on this list changes the database by being written down. Read it top to bottom: the first item is the largest structural call and the rest get smaller. Where an item says "the reference", it means the browsable vocabulary reference — as each item moves onto that surface, this table is re-pointed at it, so you end up reviewing in one place. RATIFIED rows anywhere on these pages are context — you respond only to numbered open calls, PROPOSE rows, and the items in this table.
| # | What you are being asked | Where it lives today | What a mark looks like |
|---|---|---|---|
| 1 | The content-type FAMILY layer — the types stand; the family names above them are the open call. Nine numbered proposals; the naming picks, the Retail & Commerce expansion, and the Brand & Identity summary all sit on the one review surface | the recommendation — start here, with the reconciliation page and schema view behind it | mark the page top to bottom; proposals 1–9 close on the reconciliation page |
| 2 | The eight open calls on the classification schema — a name collision, four subject-words, an empty group, dissolving the Shot Types group, a stranded pricing value, three same-idea pairs, three details that duplicate framing, and words shared across axes | the proposed classification schema | pick one option per numbered call; strike anything you disagree with |
| 3 | The first promotion round — 64 moment-detail values that clear the five-use floor (the 64 here is a coincidence — these are moment details, unrelated to the 64 content types in item 1), 76 spellings folding into them, and 2 that are really shelf names | the promotion list | strike any line; an unstruck line executes in the pipeline lane |
| 4 | Eight further proposals on the schema page that are not numbered calls — the family renames, subject scope, flattening a nested shelf, the typography sub-splits, tier becoming a visibility label, two detail retirements, an unused entity kind, and occasion storage | the proposed classification schema | strike or let stand |
| 5 | Two product axes that overlap almost value for value — raised 11 August 2026 (below) | this page | pick: merge them, or state what each answers |
| 6 | Three axes written on every classified moment that no vocabulary governs — raised 11 August 2026 (below) | this page | pick: register them, or stop writing them |
| 7 | Where a product lives — one thing, two homes — raised 11 August 2026 (below) | this page | confirm the shape, or rule which home leads |
| 8 | The middle level's name in the services tree — the top is ruled (Services) and the rulebook level is ruled (craft); your July frame and its corrections already call the middle level Discipline, with Specialization and Execution below it — presented for your confirm rather than assumed | this page | confirm Discipline (and the levels below), or amend |
| 9 | The per-shot crew-role vocabulary | this page | rule the sub-thread, or park it deliberately |
| 10 | The definitions themselves — most values in the registry have no written definition at all, and the ones the crafts carry are the exception rather than the rule | the vocabulary reference | strike or amend any drafted definition; the rest becomes a queue |
| 11 | The Brand & Identity review — the extraction you ordered at the second round: the shelf's piece types (with proposed aliases) and the identity-element vocabulary in seven groups, read from your named sources with live confirmation | the Brand & Identity review | strike or amend any row; the new-flagged elements are the additive part |
Still open
(Q10 — the open detail level — was ruled on 10 August 2026, after the round-2 pass: harvest-then-promote. It has moved to decided; the first promotion list comes back as a short review. Every numbered question, Q1–Q25, is now ruled.)
The per-shot crew-role vocabulary (carried sub-thread of Q16)
The craft list ruling settled multiple parents, but the separate question — a list of crew roles for the per-shot level — was in Q16's options and has not been ruled.
Raised with the Groupings chapter (10 August 2026, then titled "Works & Groupings")
The names of content-type families 9 and 10 (superseded 13 August — the whole family layer was re-opened at the final walk, see the next section; this narrower question now rides inside that one). The ratified master list left exactly one thing open at the time: what the two families covering visual & identity assets and event, retail & interactive work are called. The types inside them bind already.
The classification schema, whole — the marking surface. Joe's call at the review: mark the full taxonomy in one view rather than flat lists. The proposed classification schema draws everything — every grain, every axis, the whole moments tree with the promotion round placed in it — with eight numbered open calls awaiting his picks. The promotion list is the detailed record behind its moments section; an unstruck proposal executes as written, in the pipeline lane.
Raised at the final walk (13 August 2026)
The content-type family layer is re-opened. Joe flagged the ten family names as wrong at the walk, and the source check proved him right on every point. The short version:
- The ten family names came from a machine recommendation pasted forward from his own planning document — the types were ratified name by name; the families above them never were.
- His own ratified July model already carries a different top carve: the five-pillar output axis (brand identity & systems · campaigns · content · digital products & experiences · physical products & experiences) — never folded into the content pages.
- So the job is a reconciliation, not an invention — and the reconciliation surface is built: both carves side by side, every ratified type placed under every option, a script guarding the count.
- His standing instructions for that surface: show the full re-cut before any top-carve decision · genre rides the planned creative-classification review, never a new axis now · push back honestly wherever an org-chart instinct would encode team structure or duplicate a container fact.
(The "Performance ad contradicts a ruled split" defect this walk raised was RULED on 16 August — re-homed onto ownership, placement and ad format; on decided.)
Two small legibility fixes owed: the whole-schema page's map should say it draws the four grains, not the model pillars, so a reader does not look for products or concepts on it; and every review page must state that ratified rows are context, with responses wanted only on open calls and proposals.
Raised at the consolidation round (17 August 2026)
Joe's dictation, banked verbatim the same day (docs/reference/2026-08/2026-08-17/consolidation-round-notes.md). What it opened and ordered:
- One review surface — the recommendation page carried the whole review in reading order (ruled, on decided). Since 25 August the walk itself happens on the review pack — one page per section, every decision with its own answer line; the recommendation page stays as the full record behind it.
- The film pressure test — his "Film implies video" worry, tested type by type on the recommendation page's naming section; his "-Content" suffix idea evaluated there honestly. His picks per row close it.
- Product Launch — he wants it distinct from Product Film; the naming section presents one medium-neutral Launch kind (what launched comes from the product link) versus a stored Product Launch type. His pick.
- The Works rename — "Works" rejected in his words; the pillar is now Groupings, taken on his lean, awaiting his glance.
- Sidebar findability — every review page now lives in a collapsed In-review group; the writing standard carries the rule. Ruled, on decided.
- The pipelines section — ordered: have-versus-need pages built on a verified audit of the pipeline spec against the live code, plus spec-shaped pages for the unbuilt pipelines (brand onboarding · campaign · continuous monitoring). In build this round.
- The style loop — every review round re-reads every page against the writing standard. Ruled, on decided.
Raised at the best-practices round (17 August 2026)
A second research pass (the research extension, banked 17 August) ran on your steer — how mature organizations run big reviews, the identity/retail/naming gaps, and how the best documentation sites stay correct. Every claim was gathered live with a quoted source and spot-checked by a second adversarial pass; the plain-language digest is the best-practices evidence page, and what it changed on the review surfaces is already in place there. The same round built the walk console — every open decision as a mark-able card. That console is now a record only (25 August): you answer on the review pack, in chat, one line per decision. Things the round raised for you:
Two boundary notes on the naming section — Promotional Ad vs Key Visual, and Sponsored Content as a deal-versus-kind question — each stated with its alternative; no change without your mark.
A last-reviewed stamp on every ratified page. The strongest documentation teams fight staleness with a per-page last-reviewed date plus an alarm when a page goes untouched too long. The site's checks catch bad edits; nothing yet catches a page nobody edited. Proposal: each model and decision page carries a last-reviewed date, and the review loop flags any page older than a set window. Mark it and it joins the loop rules.
The retail zones tension is now on the Retail table itself — the industry's in-store standard carves store placements as zones and keeps piece kinds smaller (shelf talker, floor graphic, screen); the table's pick now shows that evidence beside the recommendation so you rule with it in view.
Raised at the second reconciliation round (16 August 2026, fresh window)
Joe resumed the walk and dictated a round banked verbatim the same day (docs/reference/2026-08/2026-08-16/brand-identity-retail-round-notes.md). Ruled parts are on decided; these are the threads it opened:
- The identity shelf is named "Brand & Identity" (ruled), and the identity extraction is running: the unique classifications get extracted and reviewed from the sources he named — his brand-identity spreadsheet, the GitHub Brand Guidelines 2025 book, the OpenAI guideline pages, the motion-identity framework, plus a search for the rest. The review keeps piece TYPES and identity ELEMENTS as separate strata; the extraction ran the same day and its candidates are on the Brand & Identity review, awaiting his marks (agenda item 11).
- The Retail & Commerce expansion. He called the shelf starved — missing ecommerce and the physical retail kinds. His seed lines, verbatim: "Retail & PDP — Best Buy endcaps, PDPs, in-store displays, POP" and "Packaging & Physical — Unboxing, retail packaging, printed materials." More reference files are coming from him; his old content-types list is a named seed source. Every candidate lands as a PROPOSE row, never a silent merge.
- Digital Product placement — a new open question. Where does content scraped FROM a digital product live (the ChatGPT iOS app, the web app — screens we collect and classify)? A proposal with a recommendation goes on the reconciliation surface; his mark decides.
- The channel CSV parked. His channel/touchpoint/platform spreadsheet is "for later for channels etc" — it joins the channel-stack thread, not this round.
- The Fable loop is the lane's operating process for extraction work (his call, same round): the strong model does judgment, specs and re-probes; cheap agents do the sweeps with structured output; an adversarial pass checks before anything ships. The process doc lives at docs/reference/FABLE-LOOP-PROCESS-01.md.
Raised at the schema-view review (16 August 2026)
Joe reviewed the schema view and raised six threads, banked verbatim the same day (docs/reference/2026-08/2026-08-16/schema-view-walk-notes.md):
- The rename. This site should stop being called "Atlas" and just be the BrandTrackers docs — his words: "our atlas which we should really just rename as our brandtrackers docs… added to index.md or done whenever it makes sense." A thread, not urgent; touches the site title, the front door's language, and the hosting name.
- No slashes, no compound values, ever. Stored vocabulary never carries slashes or compound labels — "we should never have like slashes or things like that in our data ontology taxonomy." The ratified type list currently has slashed entries (manifesto / anthem film · commercial / spot · interview / Q&A · offer / promotion · lesson / course · brand story / origin story). Each one resolves one of three ways — one value with a clean display name, two separate values, or one value plus an alias — and that cleanup list rides the value migration when the family ruling lands.
- The decision gallery. Before ruling the top carve he wants to SEE what he is deciding between — "actual use case examples or what does the vocabulary page for this look like eventually" — one preview per option, not more argument. Built the same day: the gallery — five previews, each carrying every ratified type (64 after the 16 August rulings) with live counts, the three use cases, and the literal per-piece tags.
- The content-category page's fate — DONE the same day. The generated page now opens by explaining that its question is deprecated and where each of its values went (campaign kinds to the grouping, the program to marketing intent, the job to content role); browser-verified.
- Plain speak, everywhere, as a stated rule. The writing standard currently binds the reader pages; Joe wants it stated for every surface of these docs — "embrace plain speak simple easy to understand approach to our writing for this atlas/docs always" — and the dense tables simplified. Same thread, the layout defect: wide tables cut off at the screen edge must never happen; a wide table scrolls inside its own container or gets redesigned narrower.
- Pipelines, canonically — with the daily-recognition pipeline named. Only the content pipeline exists. Missing, in his words: the entity/business-information pipeline, a campaigns pipeline for historical content, and a daily pipeline that watches incoming content after onboarding and "tries to see and or identify content that could be a campaign or some other grouping that we don't know about yet." This strengthens the already-ruled several-pipelines and daily-recognition decisions; the planned pipelines chapter must define all of them, existing and missing.
Raised at the recommendation walk (16 August 2026, later the same day)
Joe took the recommendation's mechanism — provisionally — and dictated his own shelf carve (banked verbatim: docs/reference/2026-08/2026-08-16/recommendation-carve-notes.md). New threads:
- His shelf carve, applied and awaiting his final word — Brand & Campaign (later shortened to just Campaign in his close amendments — the review surface uses the shortened name) · Product & Feature · Users & Customers (was Social proof) · Partnership & Creator · Education & Support · Corporate Communication · Documentation & Technical · Retail & Commerce vs Events & Experiential vs Interactive & Digital Experience (the old experiences family, split three ways) — with Advertising & Promotion and Brand Identity's internal depth flagged open by him (both since resolved: Advertising & Promotion was dissolved and the identity shelf named Brand & Identity in his close amendments — kept here only as the record of the round).
- Commercial vs Brand Film — his question ("are you saying commercials based on the channel?"); the boundary pick is presented on the recommendation preview: a commercial is a kind of piece (short, offer-forward, made to run as an ad), never a channel fact.
- The identity vocabulary review — brand identity is more than logos; his comprehensive element lists (lockups, wordmark, palette, typography — the brand-identity spreadsheet from July) feed a dedicated visual/verbal/strategic identity review; identity depth is its own queued item.
- Owned vs earned, answered in place — ownership (owned · paid · earned · partner) is its own axis, never a shelf; the recommendation preview shows his exact examples placed.
- Display naming rules — Title Case on every display name, always; no slash pairs (the resolve table with aliases is on the recommendation preview); and "film" flagged as a bygone-era word for the naming review (renaming ratified types is a deliberate pass, not a drive-by).
Inputs routed to LATER chapters from the adjudicated outside review (16 August 2026) — recorded here so the chapters inherit them instead of re-deriving:
- For the relationship-matrix chapter: typed edges between groupings. Never one generic "contains" edge that can point both directions — name the edges: an event hosts a campaign · a campaign activates at an event · a campaign produces an event · a project delivers an experience · a sponsorship connects a brand to a property. Prevents ambiguous nesting and accidental cycles.
- For the pipelines chapter: grain classifiers and candidate membership. A source decomposes into candidate objects (a vendor page can reveal a project, a client brand, a campaign, an event, ten image pieces and a case-study article — it is never asked to pick one category); separate classifiers per grain (grouping · piece · experience · rendition/appearance · moment); and campaign membership supports a candidate status with confidence and evidence, confirmed later — which is exactly how the ruled daily-recognition pipeline works (a campaign may only become visible after several pieces accumulate).
- For the definitions queue: the boundary terms. Each of these is a whole-piece kind ONLY when the named structure is the spine of the piece, otherwise it is a style, role, or moment: Documentary (else documentary-style treatment) · Interview (else an interview segment or talking-head shot) · Testimonial (else a proof role or testimonial moment) · Product Demo (else a demonstration moment) · Music Video (else music-video style) · User-Generated (else UGC-style treatment). These become the registry's confusable-with notes when the definitions are written.
- For the naming review: customer story vs user story. Joe: "customer stories or user stories not sure which is best or right or if different" — both are ratified types today; the review says whether they genuinely differ (a paying customer's story vs any user's story) or merge with an alias.
Raised while building the vocabulary reference (11 August 2026)
Both of these came out of reading the live system rather than any document, and both continue the numbering the schema page uses.
The ninth open call — two product axes that are nearly the same axis. Every classified moment can record how visible the product is and what part the product plays. The first offers absent · device · implied · in_hand · ui_shown; the second offers absent · active_use · background · contextual · focal · implied. In the data the two have already blurred: the visibility column is being written with active_use, contextual and focal — words that belong to the other axis and were never registered for this one. So five of the eight values actually in use on the visibility axis are shared with the role axis. Proposal: keep one axis — what part the product plays — and fold the genuinely different visibility words (device, in_hand, ui_shown) into it as their own values. If you would rather keep two axes, they need a sentence each saying what only that one answers.
Where a product lives — one thing, two homes (a confirmation, not a defect). The ruling says a product is an entity that can carry its own brand identity, and the live system holds products in two connected places. About four hundred entities are labelled as product lines inside their companies. Separately, a products table holds 139 products — every one linked back to its owning entity, each able to carry its own brand identity, a parent product, and a launch date — with about 1,500 named features under them, 8,500 links from content to a product, and 41 links from campaigns. So the product layer is real and connected, not lost: content can say which product it shows, and a product can operate as a brand, exactly as ruled. What the review should confirm is the relationship between the two homes — whether the products table is the detail layer under the product-line entities (the natural reading), and which one a browsing surface should lead with. The vocabulary reference carries the product axes either way.
The middle level's name in the services tree (the two outer levels are ruled). The stored tree has three levels, verified against the live table. Joe ruled the outer two on 11 August 2026: the top level is a service — what a client buys, like creative production, brand strategy, or media strategy and buying — and the bottom level is a craft: the rulebook level, where every entry says exactly when to tag it, when never to, what it is confused with, and at what size it applies. The reference surfaces now say Services. What remains is the level between them — under creative production sit animation, copywriting, design, photography, post-production, print and physical design, video production, and generative production. Your own July frame already calls this level Discipline (with Specialization and Execution below it), so agenda item 8 presents Discipline for your confirm; group and family are the alternatives if you want a different word. Renaming anything stored in the database is pipeline-lane work either way.
The tenth open call — three axes written on every moment that no vocabulary governs. Alongside the registered axes, each classified moment also records an interest type (17 different words), a motion state (5) and a visual theme (550 different free-text phrases). None of the three has a single row in the registry, so nothing defines them, nothing validates them, and they cannot appear in any governed list. One word, kinetic_type, is being written as both an interest type and a motion state, which is the sort of drift a registry exists to prevent. Proposal: register interest type and motion state as real axes with written definitions, and treat visual theme as an open level like the moment detail level — harvested, then promoted once a value proves itself. The alternative is to stop writing them.
The two vocabularies — RATIFIED (10 August 2026, picks included)
Joe reviewed both proposal pages, then answered the three remaining picks the same day. The content-type master list is ratified — every type binds, and the family once called "Proof & advocacy" is now Social proof (his pick, made with the word-overlap flag in view; a family name never appears where channel values do — Joe's "content category" concern is answered on the page as the category-vs-type split). The channel stack is ratified — all fifteen quality-check items accepted and folded in (eleven added channels, four contradiction fixes), and the platform-to-entity link is confirmed (a platform's owner is also an entity; one link joins the two roles). What still travels forward: the names of content-type families 9 and 10 ride with the Groupings chapter, and the vocabulary migrations themselves are pipeline-lane build work.
What happens to a question
- It gets ruled on, with reasoning.
- The answer is written into the model page it affects — as a plain statement, not as "we decided".
- The question moves to decided, dated, with the reasoning kept.
- If the answer implies something must now be true of the data, it becomes a check on the conformance report so it cannot quietly stop being true.
Step 4 is the one that matters. A decision nobody can measure is a decision that will be re-litigated.
And the same lifecycle governs a proposal page. A "Proposed" page is temporary by design: once it is ratified, its content is built into the model section — plain statements on the model pages, values in the generated reference — and the proposal page retires, with every value re-checked against the live system at the moment it is folded in. Proposals never accumulate here. (The two ratified vocabulary pages are still up deliberately, not by neglect: their fold-in waits on the family ruling, because the family layer above their types is exactly what is being re-cut.)