Finish the Docs review — revised plan

Current work — whole-program audit, approved 5 October 2026. Start with Program audit: current state and path to delivery. It connects the original scope, current decisions, documentation gaps, system evidence and the retained product phases. All four current graphics recommendations are now approved. The active work is continuity recovery and the whole-program audit; no new small packet is needed.

The 4 October decision-pass plan below is retained as the review method and history. The 5 October amendment changes the immediate order: recover and publish the whole-program map before more small naming packets, audit the connected model and system, then discuss findings before substantive restructuring. Existing product requirements and approved meanings remain binding. Audit approval does not authorize production changes.

Morning reset — 6 October

The overnight audit produced a much clearer problem map. Production has not been repaired, and Docs is not yet fully reconciled. The next outcome should be corrected, connected model guidance and a small set of executable repair contracts. More investigation is justified only when it resolves a named gap in those deliverables.

What changed overnight

Area	Completed work	What that does not establish
Earlier decisions	Connected the original 28 review questions to their historical answers, compared all 24 substrate implementation proposals with later decisions, and mapped 45 additional approved choices to their recorded replies.	Old implementation recipes are not newly approved. Partly answered bundles remain partial.
Connected model	Read the overview, five pillars and relationship reference together; identified fourteen documentation findings and drafted twelve relationship contract groups. Prepared the missing Concepts explanation from recorded decisions.	The Concepts draft is not yet the canonical pillar. Every relationship and every model sentence has not passed complete system verification.
Database and pipelines	Extended catalog review and bounded checks of campaign recovery, later vendor evidence, captures, classification ownership, prompts, onboarding, scheduling, access and saved research. Several failure cases were reproduced in isolated tests.	Local reproduction is not deployed-code parity or proof of a historical incident. Catalog coverage is not semantic correctness of every stored row.
Vocabulary	Reproduced the 95 generated craft pages from their saved source and grouped definition problems by shared rules. Follow-up checked strategy evidence, production-process claims and contributor scope.	Renderer parity does not approve all definitions. Source examples and remaining consumers still need bounded checks.
Publication	Published the audit findings and supporting evidence through the existing Docs project; preserved original plans, approvals and historical links.	Publishing a finding is not repairing it, and a successful Docs build is not product acceptance.

The program report carries the findings and evidence. The existing inventory currently tracks 636 obligations, including 289 Markdown pages. These are not unanswered questions or a completion percentage. Its conservative assessments still contain substantial unexamined documentation, unverified implementation and unexercised acceptance. Those gaps must remain visible.

What matters most now

1. Access and truthful saved references. Selected access paths and missing-media behavior need repair contracts before wider sharing. Do not infer access from source ownership or a visible link.
2. Human corrections and recoverable writes. Reanalysis and partial failures must preserve reviewed relationships and allow safe recovery. These come before expanding automatic processing.
3. Identity and accumulating evidence. Separate collection source, depicted product, contributor, work, version and appearance. Connect later vendor material to existing campaigns without inventing membership.
4. One classification contract. Definitions, effective prompts, validation, writes and queries must agree about scope and evidence. Group repeated vocabulary defects by their common cause.
5. Canonical explanation. Publish approved meanings consistently across the model, generated reference and machine-readable output. Make dated observations and unresolved capability explicit.

These are dependencies for delivery, not permission to change production. Detailed evidence and limits remain in the report.

What to review today

Correction order approved on 6 October. You reviewed the connected relationships and model-page corrections. No repeat answer is needed. The approved order is Concepts and shared rules; connected pillar explanations; grouping relationships; then shared vocabulary. Residual choices return only when a checked example exposes a meaningful difference.

You also authorized the move to docs.brandtrackers.xyz, preserving old links, and requested current guidance above decisions, history and internal notes. First finish the address change, then discuss the bounded execution plan before enabling a goal. This approval does not authorize production data or pipeline repairs.

Today's bounded execution sequence

Step	Deliverable	Checkpoint
Finish this handoff	One morning entry, saved evidence, preserved approvals and explicit gaps.	This page and the program report agree about what did and did not happen.
Correction order approved	The approved sequence is recorded above and in the existing plan.	No repeat approval or replacement PRD.
Reconcile the first model unit	Concepts plus the affected cross-pillar explanations, consistent diagrams and linked evidence. Follow the approved order; publish bounded units after their checks.	Show the corrected unit and remaining contradictions. Do not restart the complete audit.
Prepare first repair contracts	Access/citation boundaries and correction-preserving, recoverable writes, with consumers, compatibility, failure tests and rollback.	Review each concrete repair before production implementation. Unavailable runtime checks stay explicit.
Finish remaining coverage	Work through the existing inventory by relationship or workflow family; reuse unchanged evidence.	Separate unresolved meaning, documentation correction, implementation defect and missing proof.
Return to delivery	Execute separately approved repairs, then resume the retained research-experience phases.	Demonstrate actual journeys; frontend expansion remains held until its contracts and review gates are ready.

Database-connector authentication and SSH authentication blocked the latest fresh checks. Earlier dated receipts remain useful; they cannot certify the current running system. Restoring access enables deployed verification. Drafting, documentation reconciliation and local contract work can continue independently.

At this morning check the goal control reports usage-limited, despite the credit reset. That indicator does not establish ongoing execution. Manual work in this chat continues under your request; automatic continuation needs the goal's control to resume. Preserve the chat, repository and product scope. The older overnight goal should not be resumed unchanged; discuss a finite replacement objective before editing its control.

The next goal should finish a defined Docs/model milestone with explicit evidence and a pipeline handoff. Its exact wording will follow the requested discussion. The existing Docs PRD, product plan and build contract remain the authorities; boards, trackers, reels, grounded Ask, discovery, documents, collaboration and phone acceptance remain the destination.

The audit finishes only with explicit coverage and limitations for every obligation and system path. Repairs finish with verified behavior. The product finishes with accepted user journeys. None of those milestones is replaced by the volume of documentation produced overnight.

What the reset found

Fourteen numbered sittings are planned. Only five numbered sitting pages exist: 1, 2, 10, 11 and 14. The others have source questions in the older review pack, but are not completed sittings ready to send. New follow-up pages have answered parts of those planned sittings, so page order no longer describes progress reliably.

The historical inventory has 62 question identities. At the reset checkpoint, 23 had banked responses, including two partly answered bundles; one additional item is explicitly recorded as already settled. The remaining 38 lack a response or settled disposition in that intake. That does not mean 38 new questions for you. Some overlap, some have earlier approvals to reconcile, and some contain several independent choices.

The inventory also misses important work outside those identities: eight numbered classification calls, the rest of that page's proposals, moment promotions, and the definition review that was planned without numbered questions. Finishing the 62-item list alone would not finish Docs.

The reset inventory recorded 316 source and review files, including 245 Markdown pages. On 5 October, the coverage check found 38 newer Markdown pages missing their later audit obligations and added them to the same queue. Those 283 Markdown pages received those obligations; two new feature-review pages were subsequently added to the same queue, and the check guards against future omissions. This establishes coverage to track, not that every sentence, vocabulary value or database relationship has been verified. The complete live audit remains a separate deliverable.

Why progress has felt slow

The old sequence groups questions by their shape: naming, placement, splitting or confirming a list. That helps a page read consistently, but it does not establish what matters most or what another decision depends on. Later approvals have also changed the premises of some older cards.

For example, the older classification proposal pointed toward merging product appearance and role, although you approved keeping them separate. It also described catalogue shots as never useful creative references, which conflicted with the recorded rule that visibility and usefulness are different. Those two statements and its implied-consent instruction have now been corrected. Other proposals remain for reconciliation, not blanket approval.

The answer is neither a fresh application nor a full database investigation before each wording choice. It is one visible queue that distinguishes a human decision, an already-approved rule, an implementation defect and an evidence gap.

Approved order

Review package	What needs attention	What it unlocks
1. Remaining relationship boundaries	A product a capture came from versus one it depicts; a page versus its useful parts; physical work versus photographs and original assets; brand strategy versus what a piece expresses. Carry existing approvals forward and ask only residual choices.	Consistent identities and links across entities, products, groupings, content and moments.
2. Classification and discovery	Moment shelves, production-shot and narrative discovery, visibility, craft levels and attribution; distinguish genuinely different questions from duplicate labels. Reuse the pending video, advertising, sponsorship and Offer choices.	Definitions that can govern prompts and useful searches without mislabeling a whole piece.
3. Identity, retail and shared headings	Visual, verbal, strategic, audio, motion, environment and digital identity; retail objects and placements; remaining type-list headings and membership.	One coherent browse structure, with overlapping research paths and no invented hierarchy.
4. Remaining names and definitions	Unresolved aliases and type boundaries, podcast wording, smaller additions, and named definition/promotion batches.	A vocabulary ready for implementation review without silent synonym merges.

This is the approved execution order. It preserves every original question through a mapping to its package; combining questions does not merge their answers. A relationship already settled needs an engineering check, not another question. If package 1 reveals no residual human choice, move directly to package 2.

The video and promotion choices are now answered. The Event Recap choice is now answered. Answers can arrive in any order. Do not repeat existing notifications or let waiting for one block unrelated work.

Morning review — 5 October

The prepared morning packets are answered. Your latest approval settles both feature-comparison choices. Event Recap remains approved. The preceding five approvals settle podcast episodes and brand stories, plus books, one-pagers and speaker cards. All earlier answers remain recorded. Broader bundles remain partial where individual entries have not been reviewed.

The table contains 19 answered packets. The product-use packet is answered: deliberate handling and adjustment are included. Both product-feature choices are approved. The graphics packet now includes the requested page-type and section check, including image, video and text content. Digital Wallpaper is approved as searchable intended use. The four video/promotion choices were additional to that table and are also answered. The answered Event Recap choice, Corporate Film and remaining definition and promotion work are additional; these counts do not measure completion of the whole audit.

The next work, in priority order:

1. Product features and comparisons — both choices are now approved. Record their consequences and verify the implementation later; no repeat answer is needed.
2. Graphics, sections and placement — the requested bounded page-type and section check is documented, with implementation gaps kept separate. All four current recommendations are approved. The older first-three filter proposals remain superseded history. No repeat answer is needed.
3. Using and handling products — approved: include deliberate handling and adjustment as well as operation. No repeat answer is needed; product-role separation stays approved.
4. Remaining relationship and classification boundaries — reconcile product/feature links, source versus depicted work, campaign and later vendor evidence, and whole-piece versus part descriptions against prior decisions. Bring back only an unresolved meaning; keep implementation defects in the audit queue.
5. Definitions and membership — finish remaining service/identity entries and individual moment descriptions using the approved structure. Then resolve smaller names, including the prepared launch and manifesto work, without silently merging terms.

Event Recap is now answered. No new answer is needed for it or the two packets approved before it. Answered pages now live under Recorded reviews; older walks remain accessible separately. Neither section is a second queue.

The bounded feature check recovered the July category approval and inspected current shopping/comparison records and their source pages. Its two choices are now approved. The complete feature audit still needs identity resolution, provider context, product and specific demonstration links. The complete services vocabulary audit must check each name, definition and relationship; the level-name approval does not approve the existing tree. Channel/touchpoint mappings, exact credits and later-arriving vendor campaign work remain required checks. These checks inform the next recommendation without turning every implementation defect into another approval request.

No answer is currently required for these prepared packets. Remaining definitions, original console choices and implementation evidence are being reconciled before another question is sent. You do not need to reread or reapprove answered sittings.

The table below keeps every prepared packet accessible. Its order reflects preparation, not a required answer sequence. The current recommended order is above.

Packet, in preparation order	What remains to review	Status
Moment discovery	Keep the shot group, offer a visible entrance and separate framing from other descriptions.	Answered 5 October; all three accepted. Exact vocabulary and individual merge proposals remain open.
Classification boundaries	Chat versus AI participation; Data Analysis versus AI involvement. Named-feature handling applies the existing product model.	Answered 5 October; both accepted. Format's rename-only question is withdrawn. Other activity descriptions still need checked examples.
Brand identity: names and motion	Guideline names, Key Art and a separate Motion heading.	Answered 5 October; all three accepted. The broader identity elements and exact relationships remain open.
Retail and complete pages	Display form versus location, Product Detail Page type and Shoppable interactions.	Answered 5 October; all three accepted. Installation, booth and signage meanings remain in their separate packet below.
Names and definitions	Podcast Episode across media, preserving excerpts; Brand Story versus origin-focused narrative.	Answered 5 October; both accepted. Exact definitions, other naming pairs and implementation remain open.
Shared headings and credits	Discipline naming and Interactive Kiosk placement.	Answered 5 October; both accepted. Moment-credit capability stays approved; exact role vocabulary and implementation remain open.
Books, overviews and speaker cards	Book/eBook, One-Pager and Speaker Card meanings.	Answered 5 October; all three accepted. The four graphics choices are also answered; Template was already approved.
Images, sections and where they appear	Three revised choices connect assets to email, article and social context. Digital Wallpaper stays separate.	Answered 5 October: evidenced email/article/social context; Digital Wallpaper as intended use. Complete section vocabulary and implementation remain open.
Installations and signs	Installation versus Exhibition Booth, Signage across settings and digital-screen presentations.	Answered 5 October; all three accepted. Physical storage and the remaining retail list stay open.
Visual identity elements	Themed palettes, gradient systems, shape systems and brand characters.	Answered 5 October; all four accepted. Other identity elements and physical implementation remain open.
Verbal identity	Voice versus tone, messages versus taglines, naming rules versus terminology.	Answered 5 October; all three accepted. Calls to action and the displayed strategy boundaries are also approved. Other vocabulary and implementation remain open.
Strategy statements	Source names for purpose statements, stated values, promise versus value proposition.	Answered 5 October; all three accepted. Other strategy definitions and full implementation remain open.
Digital design systems	Shared system meaning across platforms and a searchable responsive-layout description.	Answered 5 October; both accepted. The environment and icon/component meanings are also answered below; exact storage and implementation remain open.
Design in physical spaces	Environmental Graphics, Retail Design, Exhibition Design and Workplace Branding.	Answered 5 October; all four accepted. Registry placement and implementation remain separate.
App icons and interface components	App identity versus interface symbols, and searchable component kinds.	Answered 5 October; both accepted. Exact component vocabulary and implementation remain separate.
Interactive experiences	A shared browsing entrance for interactive work.	Answered 5 October; heading accepted. Individual type names were already approved; exact definitions, registry reconciliation and capture support remain separate.
Music labels	Music Label as a company type rather than a separate basic kind.	Answered 5 October; company-type placement accepted. General company/type approval stays settled; creation-path reconciliation and relationship details remain separate.
Using and handling products	Deliberate handling and adjustment versus physical operation only.	Answered 5 October; include handling and adjustment. Existing assignments and shelf scope remain separate.
Product features	Supported category overlap and selective, defined capability comparisons.	Answered 5 October; both accepted. Exact category lists and storage remain separate.

The video, advertisement, sponsorship and Offer choices are answered. The Event Recap choice is answered; no repeat answer is needed. The source-product review is answered.

One more repeated question has been removed. The shared-name rule was already approved: different lists can share a word when they describe different things. No new blanket naming answer is needed. Reveal, Demo and Transformation still require individual definition checks; the pending packet questions above are unchanged.

The old umbrella-heading question does not need another answer. Your approved rule already places headings on the type list. That closes the repeated mechanism question; it does not approve every heading name or placement. The older social proposal still needs separate treatment of creator relationships, a possible Meme type and existing social labels. Reaction Video is already approved and no longer waits on that proposal.

Next work: continue continuity recovery and the whole-program audit. This supersedes the earlier wait-for-graphics sequence. Do not add small packets merely to keep preparing. A dependent definition waits for its answer; independent auditing continues. All remaining work stays in the existing inventory.

Four more directions were recovered from your earlier answers: flatten nested shelves, separate visibility from browsing categories, distinguish typography uses, and describe occasion independently of audience and market. Those directions need no new approval. Exact definitions, data mappings and implementation still need verification. Moment discovery and Offer are now answered separately. Exact descriptions, event relationships and implementation still need their own evidence.

Each packet starts with what I need from you and a short list of recommendations. Every independent choice explains what is already decided, the remaining question, a checked example, a counterexample, my recommendation, the alternative and what changes downstream. Detailed source and technical evidence is linked separately. Most packets contain three to five choices; use fewer when more would repeat an answer.

A page becomes Ready only after prior answers are reconciled, relevant sources and examples are inspected, system checks cover facts that affect the recommendation, an independent challenge is resolved, and the hosted page is verified. Otherwise it remains Preparing or Waiting, with a reason. Answered pages remain accessible as records. A blank answer is never approval.

You can dictate or answer a packet in one message: “accept the recommendations except …”. That applies only to the explicit choices and recommendations on that version of the page. We will save your exact reply immediately, map each choice separately and ask about ambiguous parts rather than guessing. It does not approve unrelated vocabulary lists or physical implementation.

During the morning session, independent ready packets can follow one another without waiting for a deployment after every answer. Related answers may share a publication. A dependent packet waits until its underlying answer is recorded, reflected in the affected explanations and published. New ready choices use the mobile question tool; existing pending questions are not sent again.

At each working checkpoint, update this table with what is ready, what is waiting and the next action. The morning handoff reports independently answerable choices, their suggested order, exact blockers and the last verified hosted revision. It does not count files as completed decisions. If tools or goal execution stop, report that explicitly.

This preparation work preserves the complete decision pass and the later comprehensive audit. It does not expand the frontend or change production schema, prompts or automation.

Review the whole model — now in progress

Approved for execution on 5 October. The feature and hands-on approvals are published. The later continuity instruction confirms the audit sequence; graphics meanings are now approved. The first checkpoint report and its technical evidence show the current coverage and limits. This does not declare the decision pass or audit complete.

Read every current model page and reconcile its definitions, diagrams and relationships with the decision records. Check the human explanation against machine-readable answers and references. Include generated vocabulary pages and the sources that generate them: a displayed database value is not proof of an approved meaning. Preserve historical questions and replies, but make their status unmistakable.

Audit the sidebar as part of that review. Current guidance, unanswered reviews and historical records need clear homes. Remove stale approval requests from active navigation without deleting their evidence or breaking links. Check that resolved proposals have reached the affected model pages, including definitions and diagrams, rather than living only in a decision log.

The deliverable is one prioritized gap report in the existing inventory. Each item identifies the affected page, prior decision, current evidence, consequence and next action. Distinguish settled meaning, stale explanation, contradiction, unresolved product choice, implementation defect and missing evidence. Rank by what other decisions depend on and by effects on collection, classification, relationships, queries and research journeys. Do not rank by which naming question is easiest to prepare.

Candidate areas to reconcile include grouping relationships and later vendor evidence; product and feature identity; page, part and asset appearances; classification scope and corrections; service and credit vocabulary; and channel/touchpoint mappings. These are areas to check, not a claim that their underlying meanings are all undecided. Recover earlier answers before asking anything again.

Discuss the findings with Joe before revising the plan, goal wording, product brief or task sequence. Preserve the existing phases and their history. The checkpoint should show what remains in the decision pass, what belongs to the comprehensive technical audit, and what implementation and product acceptance must later prove. Continue using the current Docs address and keep frontend expansion held.

How each review will work

1. Recover the subject before preparing three to five related choices. Search current Docs, canonical specifications, relevant Markdown research, implementation notes and Git history using current and former names. Follow explicit historical references; inspect referenced conversations when accessible and label unavailable history. Read relevant current documents fully before editing them. Reuse the verified subject-to-source map and inspect changes instead of repeatedly rereading unrelated pages. Use fewer choices when more would repeat a settled answer. Start from real examples; inspect the actual example, including the relevant interval for motion. For each choice, say what is already settled, what remains, the recommendation and the consequence of choosing otherwise.
2. Check only evidence that can change that choice. Read the related prior decisions, inspect the relevant live schema and records, and trace the affected path far enough to expose a false premise. Flag incomplete access. Do not turn a missing implementation into a question about whether an approved concept should exist.
3. Publish one review and ask through the question tool. Keep the evidence available without requiring you to read every receipt. Each independent choice gets its own answer. No response means no approval.
4. Bank answers immediately. Update the decision record, affected model guidance and the same queue. Preserve original replies and historical pages. Several related answers may share one publication; a meaning must be published before a later question relies on it.
5. Report progress by resolved decisions and blockers. A sitting count, a document count or a passing build is not product completion.

Use roughly 30 minutes as an initial preparation target for an ordinary batch, not a cutoff on accuracy. If evidence requires longer, report the specific conflict and continue independent preparation. Keep the existing 60–90-minute outcome checkpoints. Do not spend that whole interval polishing a page whose question is already answerable.

Then perform the complete audit

After the decision pass, reconcile every inventoried page and every vocabulary row against the combined decisions. Each gets a disposition: checked, needs correction, needs a human choice, inaccessible or deferred with a stated reason. Plain-language definitions need inclusion and exclusion boundaries, examples and explicit uncertainty. Historical pages remain history; current guidance must stop contradicting the record.

The technical audit covers identities, data types, required values, primary and foreign keys, uniqueness, relationship direction and cardinality, junctions, deletion and merge behavior, indexes and actual queries. It also covers time, provenance, evidence, permissions, human corrections, versions, publications, captures and independently useful parts. Check the real system before specifying changes. This inventory is not a claim that those checks have passed.

Trace discovery → capture → routing → composed prompt → validation → writes → publication → queries → interface. Include retries, corrections, missing media and partial failure. Review onboarding, campaign recognition, continuous monitoring, schedules, cost controls and delivery. Approved semantics, a working writer and a working user journey are three separate acceptance checks.

Technical findings receive a repair plan and rollback evidence. Only genuinely unresolved product meanings come back to you. No schema change, bulk reanalysis or automation activation follows from approving this review method.

The larger product plan remains intact

The existing phases remain: record and advance Docs; compare guided discovery and collection layouts; deliver one complete research journey; repair the production data path; prove grounded Ask and controlled processing; extend the broader experience; complete hosted phone and release acceptance.

For now, further frontend expansion remains held. Existing layouts may illustrate a decision. Boards, trackers, reels, scoped conversation, reusable discovery/subject/collection pages, daily Briefing and Feed, documents, collaboration, permissions and responsive behavior all remain in scope. The documentation and schema work must lead back to those journeys.

What closes this stage

The decision pass closes when every original question and every additional review surface has a recorded disposition, no bundled choice is falsely marked complete, and the remaining engineering questions have owners and evidence requirements. The audit closes only with item-level evidence. Repairs close with verified behavior. Product acceptance still requires the actual journey to work and be useful.

Execution has started. The first review, source products and complete web pages, is answered: source and subject remain separately browsable with evidence. Whole-page library membership, document parts, photographs and source strategy claims already have answers. The first classification and discovery meanings are now answered. Continue the program audit; exact definitions and complete verification remain open.

The existing review inventory now has an execution section. Original questions and replies remain intact; partial naming bundles, classification calls and individual promotion/merge candidates have their own entries. Unreconciled history is an evidence gap, not an assumption that you owe another answer. The complete verification pass must expand every remaining bundle and generated definition into item-level evidence before it can close.

The maintained source inventory is docs/reference/2026-10/2026-10-04/review-reset-inventory.json in the repository. It records original sources, dispositions, dependencies, affected documents, downstream consumers, next actions and evidence requirements. The earlier inventory remains a dated snapshot; the execution section and its ordered execution updates own the current queue. Later updates amend the named obligation without rewriting its earlier disposition.

Last reviewed 6 October 2026.