Eve library

Eve — Elikem Granular Review: product scope and acceptance

PRD

Revision 13

On this page

Scope prepared on 2026-10-03 at Elikem's request. Status: proposed implementation scope, to be reviewed through Eve's existing process. The conversation authorizes scoping and backlog preparation; it does not approve this document, activate workers, enable automation, or publish content.

Revision 4 (2026-10-03) applies Elikem's instruction to adopt the assistant's review findings and a Codex read-only second opinion: owner triage and campaign tiers sized against the measured corpus, per-version revision selection, a “review completed, changes required” synthesis outcome, a transactional content watermark for current sign-off, a bridge to the existing operational approval, implementation status and claim labelling in the review context, a cut-from-scope outcome, bounded decision visibility to agents, standing-decision capture, calibrated estimates, stricter execution provenance, and two owner pilots before the Evolution and Integration phases.

1. Purpose and agreed boundary#

Give Elikem a sustainable way to personally review Oshun's PRDs, specifications, architecture, ADRs and technical specifications in coherent sessions estimated at 5–15 minutes. Start again with existing v0 material, proceed through v1.x and later product versions, and continue to admit future material. Review of the whole corpus is a campaign Elikem shapes by triage and tiers, not an obligation to read every imported draft.

The product name is Elikem Granular Review. This is an additional owner-review record in the Eve Workspace. Existing document states, reviews, task readiness and sequencing, claims, worker/reviewer switches, release gates and publications retain their current meanings. Personal reviews are substantive decisions about exact source material, not reading checkmarks. Personal document and version sign-offs are derived from these decisions plus an explicit synthesis review; they never substitute for operational approval or publication. Where a reviewed revision also awaits his operational approval, the feature leads him to that existing action (GR-08); it never performs it implicitly.

A document can simultaneously say “Operational approval: approved” and “Elikem review: 6 of 10 chunks approved.” An old document being unreviewed, deferred, questioned or personally rejected does not automatically stop current work. Starting this journey never resets earlier approvals, reopens completed tasks, or edits canonical source documents.

2. Users, attribution and visibility#

Elikem is the owner and decision maker. Agents may inventory sources, propose chunk plans, summarize context, explain terms, suggest findings and prepare follow-up drafts. They cannot originate a personal approval, sign-off, exclusion, dismissal or finding resolution in his name. An assistant may carry out an explicit decision in Elikem's authenticated Workspace session under the existing assistant policy. An owner-authenticated session proves account authority, not that Elikem instructed that particular decision: an assistant-executed decision records execution provenance and a reference to the explicit instruction bound to the exact action and subject (document, revision, chunk or manifest, and outcome), and is refused without it. A worker credential or delegate outcome never counts as Elikem's personal decision.

The initial product supports this one named owner. Authorization binds a provisioned stable owner user ID, not an editable display name, the string “Elikem,” or every admin account. It does not introduce a self-service privilege grant. Existing document access restrictions still apply.

Visibility states. Visibility is defined per record state, and the same word never means two things:

  • Drafts (unsubmitted decision text, notes, bookmarks, reading position) are private to Elikem and never readable by any agent.
  • Concerns and questions that have not been dispatched are private to Elikem. A change-request concern is such a record (GR-07); it is not the decision.
  • Submitted decisions (approve, request changes, cut from scope, reviewed as historical context, defer) and their rationale become readable in bounded form by authenticated agents through task context for the spans a task is bound to, from the moment of submission, so a worker acting on a requirement sees what Elikem decided about it. Rationale is written knowing agents will read it; private remarks belong in notes. No agent read ever includes drafts, notes, bookmarks or undispatched concerns, and a decision read confers no authority.
  • Dispatched follow-ups (a task, a proposed revision, a successor, a standing decision) are visible as the ordinary board or content records they are, with a backlink to the finding. The concern body itself stays private unless Elikem includes it in the dispatch payload shown in the preview.

Aggregate or assistant reads of private material must be separately authorized. Agents otherwise receive only authorized source context and explicitly dispatched or deliberately shared material.

3. GR-01: historical inventory, product-version order and owner triage#

Create an auditable candidate inventory through Eve's authenticated content authority, using document IDs, immutable revisions, source provenance and relations. Include existing and future PRDs, specs, technical specifications, architecture and relevant ADRs. Kind alone is insufficient: existing technical-specification documents are sometimes classified as guides. Scan metadata, authoritative titles/headings, release associations and preserved source aliases; expose uncertain classification for correction.

Product versions and document revision numbers are distinct. Support v0, v1.0, v1.1 and later v1.x, v2 onward, and future version labels without a fixed v9 ceiling or lexicographic ordering. Version order is explicit and editable. A release-less foundational document may support several releases; store those associations without copying the document or silently treating several memberships as several independent approvals.

Every candidate has a disposition: included in a version, shared foundation, historical/superseded context, excluded with reason, or unresolved/missing source. Dispositions are set by two means: a deterministic census proposes them, and an owner triage pass confirms or changes them. Show all inventory gaps. Do not fabricate a v0 corpus if provenance is absent. Preserved trackers and source checkboxes remain historical data and are never imported as new work items. No retired canonical Markdown or presentation authoring path is restored.

Revision selection is per product-version membership, not a single approval-recency rule. For each membership, select the revision that version-specific provenance supports: a linked publication snapshot, a release association recorded on the revision, or preserved source lineage. Where no such evidence exists, offer the latest approved revision at the campaign cutoff, otherwise the latest revision clearly marked draft/in-review, and label the membership “revision selected without version-specific provenance.” A shared-foundation document reviewed under v0 must not silently carry requirements added for v1.x; where the only available revision does, the chunk orientation says so. For archived sources, preserve the chosen historical revision and its archival context. Allow deliberate selection of another historical revision before activating a plan. A product-version grouping is not proof of the exact documents shipped at that time; publication snapshots are linked when available, and unknown historical release provenance remains explicit.

Owner triage and campaign tiers. Before a campaign activates, Elikem works a document-level triage queue. For each candidate he sees title, kind, version memberships, size, provenance, operational state and bound work, and sets one of: include for chunk review, spot-check (a sampled subset of chunks), reviewed as historical context, or excluded with reason, in about two minutes per document. Triage decisions are owner records under the same authority rules as chunk decisions. Campaigns carry tiers (must review, spot check, context only) derived from triage; V2 through V10 product roots default to the historical tier because those roots hold UE, docs and ops material only. The activation screen shows the estimated total owner hours per tier as an honest denominator, computed from chunk estimates, before anything is frozen. An estimate is a planning aid and never a quota.

Corpus observed during scoping (authenticated reads on 2026-10-03; a scoping read, not the census): prd 8; adr more than 100; architecture more than 100; spec between 300 and 450; guide more than 1,000, some of which are technical specifications; 11,450 documents in total. Most eligible documents are revision-1 drafts imported on 2026-09-20. At roughly five chunks of ten minutes per document, reviewing every eligible document in full is on the order of 500 to 700 owner hours. The triage and tier model exists because of this.

Initial personal progress starts at zero regardless of operational approvals. All relevant historical documents remain discoverable; superseded material is labelled, not silently omitted. Elikem can mark historical material “Reviewed as historical context,” which is tracked separately from approving its requirements for continued use.

4. GR-02: coherent chunks with complete source coverage#

A review chunk contains one coherent decision or a closely related set of decisions, such as account recovery policy, data retention requirements or an architecture boundary. It has a useful title, intent, exact source spans, context references, explicit review questions, rationale for boundaries, estimated effort and a stable identity inside a versioned plan. Every chunk carries a deterministic default review question by document kind (PRD: is this still what you want built; ADR: do you accept this decision and its consequences; architecture or specification: is this boundary or contract correct). Model-suggested specific questions are layered on top and labelled as assistance.

Suggested plans start from deterministic document structure and may receive grounded agent suggestions. Elikem can rename, split, merge, reorder and adjust boundaries before or during review. Those operations alter the review plan, not the source document. Once decisions exist, plan changes create a successor plan with lineage; they do not rewrite earlier approval subjects.

Every in-scope source span is assigned to a primary review chunk or explicitly excluded with a reason. Supplemental context can repeat but does not inflate coverage. Lists, table headers and rows, footnotes, cross-references, diagrams, code fences and requirements must retain enough context to remain interpretable. One source block may contain several review chunks; positional section IDs are not durable cross-revision identities.

Claim labelling. Within a chunk, spans are labelled as proposed behavior, asserted existing behavior, or unknown implementation status. AI-authored documents routinely assert that something exists; an exact citation proves what was written, not that the claim is true. Deterministic heuristics (modal verbs, tense, references to code paths and bound tasks) propose the label, assistance may refine it, and Elikem can correct it. The label is shown beside the span and recorded with the decision.

Estimates include reading and likely decision complexity. The estimator is a versioned formula over words, tables, code fences, diagrams and cross-references with per-kind multipliers, calibrated from Elikem's measured chunk durations once enough sessions exist; recalibration changes future estimates only and is recorded with its version. The choices 5, 10 and 15 minutes are planning aids, not enforced timers or evidence of comprehension. Dense material may exceed 15 minutes; show that honestly and offer staged reading or a user-adjusted split. Do not cut an indivisible technical decision merely to satisfy an estimate. Reading stages alone confer no approval.

5. GR-03: guided sequence and the personal queue#

Default order is v0 forward, then product intent/scope, requirements, architecture and implementation detail within a version, adjusted for explicit prerequisite decisions. Shared foundations appear when first needed. The queue groups by version and tier; must-review material leads, spot-check samples follow, and context-only material is reachable but never recommended as next. The triage queue (GR-01) is a distinct mode reachable from the same entry. The next recommendation explains why it is next and names missing context. Cycles and unknown dependencies are visible, not silently removed.

“Continue my review” resumes the exact campaign, plan, chunk, source revision and reading position. A 5/10/15-minute selector recommends a suitable next chunk without permanently starving longer ones. Elikem may jump to another version, search, filter, bookmark, reorder or defer, with skipped material still counted as outstanding. Review ordering has no relationship to worker task priority unless Elikem separately requests that change.

Filters: product version, tier, triage disposition, document, domain, kind, personal outcome, claim label, revision freshness, deferred/questioned status, historical status and estimate. Persist the current view in a shareable authenticated URL and owner preferences. Empty, loading, failed, partially indexed, access-denied and stale states must be distinct.

6. GR-04: the focused review session#

The working surface shows a concise orientation, the exact underlying content, the decision/questions, source revision and current operational status. On desktop, use a calm reading column with navigation and a collapsible context inspector; on narrow screens, stack the context behind disclosures and keep the primary action reachable. Integrate with Eve's typography, themes, density and reader components.

The context inspector provides preceding/following source text, related approved decisions, dependencies, glossary explanations and links to the complete pinned document. It also shows implementation status: the tasks bound or linked to the span with their current states (later, ready, in progress, shipped, verified) and the span's claim labels, so approving a requirement that has already shipped is visibly a different act from approving one not started. It shows the independent reviewer's findings and the operational review history for the pinned revision, and the revision's provenance (which task or agent produced it and the stated reason). Related requirements and suspected contradictions in other pinned documents appear side by side before the decision, not only at synthesis; they are suggestions with evidence. Indicate whether a dependency is merely referenced, operationally approved or personally reviewed. Summaries and explanations are labelled as assistance with pinned citations; the approval applies to the actual source spans. Missing or unreadable required source prevents approval and explains what is unavailable.

Actions: approve the chunk, request changes with a reason, cut from scope with a reason, ask a question, defer with a reason, and save a note/bookmark. Cut from scope records that the requirement should be dropped from the product; it is a personal decision, and its dispatch (GR-07) proposes a successor revision marking the span superseded and lists the bound tasks it would strand. From any rationale, Elikem can record a standing decision through the board's existing standing-decision mechanism, so later agent questions that the rationale answers are covered. Historical material also supports “Reviewed as historical context.” Questions and notes are separate from the decision, and comments can anchor to a passage. Approval can carry a rationale. There is no bulk approve, auto-approve, approval by dwell time, or automatic document approval after the last chunk.

A submitted decision shows a brief, visible undo window of a few seconds before it commits; a cancelled decision never enters history. After commit, explicit decision correction appends a superseding event. It preserves the original decision and marks affected sign-offs as needing attention. Navigation shortcuts are available with visible help and do not trigger while typing; destructive or outward actions require a clear preview. Ordinary approval should be a deliberate, labelled action without repetitive confirmation dialogs.

7. GR-05: durable drafts and interruption recovery#

Autosave notes, bookmarks, reading position and draft decisions against the owner and exact subject. Display saving/saved/unsaved/error status. Draft decision text is not a submitted decision. Resume across authenticated devices. A failed save, expired session, tab close or navigation must preserve recoverable work and show the recovery path. Elapsed time per chunk is recorded with the session, labelled as a measurement, and used only for estimate calibration (GR-02) and Elikem's own progress view.

Concurrent tabs use optimistic versions. A late response for chunk A cannot replace chunk B; an uncertain write is retried with its original request identity. Two differing decisions cannot win silently. A stale source or plan produces a comparison/reload choice without discarding the draft. Do not queue decisions for unattended submission after reconnect. Local recovery, if used, is owner-scoped, bounded and cleared at sign-out; server storage is the durable authority.

8. GR-06: immutable historical progress and current coverage#

Show two measures: baseline review progress against a frozen campaign manifest, and current revision coverage against a stated latest inventory/checkpoint. The former does not shrink when workers create a new revision. The latter says when approved material has changed, a new section/document appeared, a dependency changed or correspondence cannot be established. Current coverage is computed against two authoritative generations advanced transactionally by the content service, never against background job checkpoints alone: a content watermark per member document and dependency (new revision, archival or restore, eligibility metadata, relations and referenced assets), and an inventory generation per campaign scope (document creation, archival or restore, and kind, domain, release or eligibility metadata changes of any document), so that material which becomes eligible outside the checked membership is detected. Stale computation is shown as stale, never current.

Record the exact document ID, revision, source-span hash and plan version for every decision. Textually identical spans with unchanged required context may be shown as “unchanged since your review,” retaining the earlier decision date and revision. This is an equivalence claim scoped to the compared material: hashes prove identity of the span and its pinned context, not completeness of the dependency graph. A missing dependency mapping, unpinned mutable context or unverified media produces an explicit “unchanged span, dependencies uncertain” label, never universal current approval. Any changed normative text, code, table value, link target, diagram/media content or relevant dependency requires another look. Unknown correspondence is conservative. Whitespace inside code and meaningful punctuation are not dismissed as cosmetic; a model cannot certify equivalence.

The catch-up queue shows old/new content, why the item returned, the revision's provenance (which task or agent produced it and the stated reason) and affected dependent decisions. Changed dependencies trigger a review-needed flag without erasing historical approvals. Deleted, split, merged, reordered and successor documents retain traceability. New material is offered as an explicit campaign extension or next baseline with a visible denominator change, rather than silently moving a completion target.

A new current revision does not prevent Elikem from approving the historical revision deliberately pinned in his session. Show “A newer revision exists” and record approval of the displayed revision only. Staleness checks concern the actual review subject/plan and the claimed coverage; they must not force an endless restart whenever workers edit a document.

9. GR-07: concerns and deliberate handoff to work#

A personal concern is initially a record in this layer. It may be resolved by clarification, dismissed by Elikem with a reason, or explicitly dispatched. A question in this layer is not automatically sent through the board's clarification API, whose existing behavior can park work.

Dispatch forms: a later task in the correct family; a proposed source-document revision through Eve's proposal/successor workflow; for a cut-from-scope decision, a proposed successor revision marking the span superseded together with the list of bound tasks it would affect, shown before confirmation; and recording a standing decision through the board's existing mechanism. Before dispatch, show the source evidence, proposed change, affected current documents/tasks, target family, duplicate candidates, dependencies, priority/status and any expected operational effect. Default new tasks to later unless Elikem chooses otherwise. Product work goes to its product family; infrastructure/deployment work goes to platform-ops. Existing or leased task bodies/statuses are not changed implicitly.

Dispatch is idempotent and creates backlinks. A task being shipped or a revision being approved is evidence of progress, not automatic personal acceptance. Elikem reviews the result/diff and resolves the concern. Clarification, dismissal and acceptance of a revised requirement are different events: resolving a concern never converts a rejection of revision A into approval of revision B, and acceptance of the revised span is a new personal decision on the new revision with its own subject. Explicit blocking of operational work, if necessary, is a separate existing governed action identifying the exact affected work. Do not add a new broad “halt workers” control to this feature.

10. GR-08: document and version synthesis#

When primary chunks are decided, offer a short synthesis review: scope, combined decisions, unresolved concerns/questions, dependency impacts, differences from current material and suspected cross-document contradictions. Suspected contradictions are suggestions with evidence, never certified truth. Deterministic completeness checks run even when model assistance is unavailable.

Document/version personal sign-off binds the manifest, included document revisions, decisions, dependencies, exclusions and synthesis acknowledgement. Synthesis has three outcomes. “Review completed, approved” requires no unresolved required chunks, questions, change requests or ambiguous source coverage. “Review completed, changes required” records that every required chunk has been examined while some are rejected, cut or awaiting changes; it preserves those decisions and their follow-ups, closes the examination, and is never displayed as approval. “Completed with exclusions” discloses excluded/historical-only material in the signed scope and is not labelled universal approval. Deferral never counts as completion. A later change flags current coverage without altering the historical signed record. A deliberate fresh review campaign can start again while prior campaigns remain accessible.

The sign-off states whether it covers the frozen baseline or the current checked manifest. A baseline may be completed while newer material waits in catch-up; it must never be labelled current approval. A current sign-off verifies two things in the same transaction: the content watermark of every member document and dependency (GR-06), refused if any moved; and the inventory generation of the campaign scope, so that eligible material created or reclassified since the manifest was checked is detected. With no eligible candidate the recheck is recorded and the sign-off proceeds; with eligible candidates the sign-off is refused and lists them for triage, after which an extension or recorded dispositions let a current sign-off proceed. “Current” means the fixed membership of the checked manifest verified against the latest eligible inventory, and the receipt says so and discloses any pending candidates. Two matching background checkpoints are not sufficient.

Bridge to operational approval. When the reviewed revision is also awaiting Elikem's operational approval in the existing review queue, a completed current-manifest sign-off offers the existing operational approve command as a separate, explicitly labelled action with the sign-off linked as evidence. The two receipts are never conflated and the operational action keeps its own permission check. Review-queue entries for PRD, ADR and architecture revisions offer “review in chunks” and show personal coverage with exact-source links, so the same owner does not reread the same material in two disconnected workflows. The operational review queue remains the only approval surface; the personal layer adds none.

11. GR-09: progress and useful ergonomics#

Show reviewed, personally approved, historical-context reviewed, changes requested, cut from scope, questions, deferred, unreviewed and excluded counts with exact denominators and last-calculated time. Aggregate by document, domain, product version, tier and campaign; context-only and spot-check tiers never count toward must-review completion. Counts are based on primary source coverage; elapsed time and estimated remaining time are labelled measurements and estimates, and measured elapsed time feeds estimator calibration (GR-02). Splitting a chunk must not make approval coverage appear to increase.

Include useful additions in the delivery scope: a short “where you left off” recap, bookmarks, passage notes, keyboard navigation, in-document search, a catch-up-only view, decision rationale history, and citation-backed “explain this” using existing admitted assistant infrastructure. No external reminders, notification scheduling, new model/provider purchase or automation switch is introduced. In-workspace new-change indicators are sufficient initially.

12. Explicit exclusions#

This scope does not rebuild the operational review queue, retask workers, introduce multi-owner approval policy, revise historical source content during backfill, replay every document edit as a mandatory review, automatically adopt suggested changes, auto-approve on behalf of Elikem, perform an operational approval as a side effect of a personal sign-off, add gamification or review quotas, add a native mobile app, or change publication/UAT permissions. Presentations, research and implementation evidence can be referenced as context; comprehensive granular review of every other artifact type is an extension, not a silent expansion. No time estimate guarantees a safe decision within 15 minutes.

13. Product acceptance scenarios#

  • P01: Backfill v0 and v1.x fixtures containing approved, draft, archived, misclassified and missing sources. Personal decisions start empty; every source is accounted for and operational records remain unchanged.
  • P02: Review v0 while a worker edits v1.x and holds a task lease. Complete/defer chunks and raise a concern; the worker's claim, readiness, review and publication checks behave exactly as before.
  • P03: A 5-minute session resumes on another device with the pinned source, note draft and position. A 15-minute estimate can be exceeded without coercion or implicit approval.
  • P04: Split/merge a plan after one approval. History remains exact; the successor's coverage is recomputed and approval is not copied to altered spans.
  • P05: Approve one chunk, edit another, then change a dependency. Only proven unaffected coverage remains current; changed/ambiguous dependencies surface for another look.
  • P06: Inspect two revisions during an out-of-order network response and session expiry. No decision attaches to the wrong chunk and no draft disappears.
  • P07: Record a change-request concern privately as a draft; a bound-task agent reads nothing. Submit the request-changes decision: the decision and its rationale become readable through task context for the bound spans, while the concern body, notes and drafts stay private. Dispatch once despite a timeout/retry: the created task or proposed revision is visible as an ordinary record with a backlink. Follow the linked work, inspect its resulting revision and explicitly resolve the concern; no automatic park/reopen occurs, and the original decision on the old revision is unchanged.
  • P08: Complete chunks with a deferred item or unresolved question; sign-off is refused with the exact gap. Resolve it and sign the frozen document/version manifest; publication does not occur.
  • P09: Keyboard-only, screen-reader and narrow-screen users can read source, open context, annotate, decide and recover from errors. Code, tables, diagrams and long documents remain usable.
  • P10: Worker, reviewer, delegate and unrelated admin credentials cannot create Elikem decisions. Required missing source and a stale manifest prevent misleading approval/current-status claims.
  • P11: Triage thirty candidates including V2–V10 roots and a shared-foundation document with two memberships. Tiers and the hours denominator appear before activation; context-only material never enters must-review counts; each membership shows its selected revision and provenance label; triage dispositions are owner records.
  • P12: Examine every chunk of a document, reject two and cut one. Synthesis records “review completed, changes required”; the rejected spans and their follow-ups persist; the cut requirement's dispatch preview lists its bound tasks; a later revision does not convert the rejection into approval without a new decision.
  • P13: A chunk whose span is bound to a shipped task and asserts existing behavior shows that status and the claim label before the decision; a reviewer finding and the revision's provenance are visible in the inspector; a related requirement in another pinned document appears side by side.
  • P14: An in-review PRD revision reaches current-manifest sign-off. The bridge offers the operational approve as a separate click. A concurrent content change (a new relation or an archived dependency) moves a member watermark and the sign-off is refused; a concurrently created eligible document moves the inventory generation and the sign-off is refused with the candidate listed for triage; an ineligible creation records a recheck and the sign-off proceeds; without the click nothing is operationally approved.
  • P15: An assistant-executed decision without an instruction reference bound to the exact action and subject is refused; with it, the event records the owner, the instruction reference and execution provenance.

14. Acceptance and release boundary#

All required capabilities above are part of the full feature, including historical import, triage, revision catch-up, final synthesis, failure recovery, accessibility and isolation from existing work. Delivery may be staged behind a feature flag, but a partial slice is not “feature complete.” Full implementation requires the linked engineering and delivery contracts, independent review, automated verification, an Elikem pilot, recoverable migration and platform-owned rollout. Content approval and publication follow their existing separate permissions.

Two owner pilots gate later phases, and the board's dependency edges enforce the gates. The read-only chunk-plan preview (EGR.34) precedes the Workspace phase and the Evolution phase, because both build on the chunk and anchor semantics it validates: EGR.18 and EGR.12 depend on it, and EGR.13–15 and EGR.36 inherit the gate. The vertical-slice pilot on a small frozen campaign (EGR.35), with acceptance criteria for orientation, chunk coherence, decision effort and interruption recovery, precedes assistance (EGR.17), plan-editing surfaces (EGR.20), client/CLI/MCP exposure (EGR.25) and, through them, the Integration phase. Pilot outcomes are recorded in Eve and may change the generator and the plan. They are feedback gates, not releases; a partial slice remains not feature complete, and current coverage is labelled unavailable until the Evolution phase ships.

Attribution and evidence. The instruction to apply the assistant's review was given by Elikem in his assistant's conversation on 2026-10-03 and is not separately recorded in Eve. Until Elikem's own operational approval of this revision, which is the attributable record, that attribution is the author's claim. The corpus counts in GR-01 come from authenticated document searches by kind on 2026-10-03 under the author identity, are reproducible with the same searches, and are superseded by the EGR.01 census.

Scoping evidence: repository b5c67c1d4934f76ededdaf7f0e67729d61654592; authenticated Eve reads on 2026-10-03; existing UX audit d81ff4ab-c540-4c0f-8855-20db7559413e revision 3. Revision 4 applies the assistant's review of 2026-10-03 and a Codex read-only second opinion (gpt-6.1-sol, read-only sandbox) on Elikem's instruction; revision 9 answers the independent review of revision 5 (reviews 24125a50, d734ae8e, 654d5178). This scope is not an exhaustive live corpus census; completing that census is an explicit implementation task.