This hub spans the review, evaluation, and verification disciplines that gate knowledge quality across the aswritten system. The review protocol cluster — established by Mar 2026 — is built around five corrections: [baseline-first comparison before shifts](https://aswritten.ai/narrative#Position_BaselineFirst), [labeled shift intent (improvement vs. counter)](https://aswritten.ai/narrative#Position_LabelShiftIntent), [sentence-scale comparison units](https://aswritten.ai/narrative#Position_ComparisonScale), [large-TX zone-scoping via TX summary → zone identification → targeted shifts](https://aswritten.ai/narrative#Position_LargeTXScale), and [domain-first eye-chart scoping](https://aswritten.ai/narrative#Position_EyeChartScoping). Two entry points — [warm-start (same session) and cold-start (PR-based, different session)](https://aswritten.ai/narrative#Narrative_WarmStartVsColdStart) — feed a shared core, with Scarlet's Jul 2025 kickoff preference on record. Review convergence is framed as [sharpness on branch intent, not exhaustive completeness](https://aswritten.ai/narrative#Narrative_ConvergenceNotCompleteness), with [lenses pushing both toward and away from the branch thesis](https://aswritten.ai/narrative#Position_LensesConvergeOnBranchIntent) to confirm directional correctness. [Annotation is positioned as quality control](https://aswritten.ai/narrative#Narrative_AnnotationAsQualityControl) — per-section during drafting (Mar 2026), not end-of-document, because [text length limits cause silent failures on full sections](https://aswritten.ai/narrative#Position_AnnotatePerSectionNotFullDoc). The [pre-registered eval battery discipline](https://aswritten.ai/narrative#Narrative_PreRegisteredEvalDiscipline) (pre-dating Jul 2025) holds that the bar cannot drift to fit the outcome. The [review flow is framed as peer review for worldview shifts](https://aswritten.ai/narrative#Narrative_ReviewAsPeerReview), not fact-by-fact checking. Verification doctrine extended Jul 5 2026 to [verify at the seam the user touches, not one layer above it](https://aswritten.ai/narrative#Narrative_VerifyAtUserSeam), codified after four 'verified' claims on the Jul 4 release day proved one hop short. [Every saved memory ends that review round and shifts the branch closer to the target](https://aswritten.ai/narrative#Position_EveryMemorySaveEndsReviewRound), with next-round lenses reflecting the updated graph.
Sub-areas
Narrative PreRegisteredEvalDiscipline4 nodes Framed in Jul 2025, aswritten-as-first-user discipline established the foundational bar: a pre-registered battery run in full with numbers reported counts as validated; informal sm…
Witness phrases
verbatim The goal of review is convergence on the branch's intent, not exhaustive completeness
verbatim Pre-register the comparison before building, committed to git, so the bar cannot drift to fit the outcome.
verbatim Annotate per-section as you draft — the text length limit causes silent failures on full sections.
verbatim Verify at the seam the user touches, not one layer above it
verbatim If I save a memory after remember returns, I want to have a review immediately started, or it offered to immediately start review.
verbatim The review flow is exactly peer review, the difference is we're reviewing fundamental worldview shifts, not necessarily all of the specific facts.
verbatim Every saved memory ends that review round and shifts the branch closer to the target.
verbatim Maximum comparison unit is a few sentences. Longer content must be broken into comparison-sized chunks.
verbatim Two convergence framings coexist without explicit reconciliation: one frames the goal as 'sharpness on the branch's intent' while the other frames it as 'convergence on the branch's intent' — subtly different orientations on what review is optimizing for.
verbatim The peer-review framing (Mar 2026) positions review as evaluating worldview shifts, not specific facts — which sits in tension with the fact-review step in the shared core protocol that explicitly checks conviction levels and relationships.
verbatim Annotation doctrine says 'write full narrative footnotes from the start; the narrative IS the value' — a stronger claim than quality control, positioning annotation as the primary artifact, not a check on it.
verbatim The eval discipline pre-dates Jul 2025 and holds 'report actual numbers before the words at par or ready' — but the hub carries no child recording what happened when a metric fell short and the iterate instruction was applied, leaving the failure-mode path undocumented here.
In this area
Narrative_AnnotationAsQualityControl Annotation as Quality Control, Not Decoration
NarrativeNarrative_ConvergenceAsSharpness The goal of review is sharpness on the branch's intent, not exhaustive coverage
NarrativeNarrative_ConvergenceNotCompleteness The goal of review is convergence on the branch's intent, not exhaustive completeness
Narrative_ReviewAsPeerReview Review Flow as Peer Review: Worldview Shifts, Not Every Fact
NarrativeNarrative_VerifyAtUserSeam Verify at the seam the user touches, not one layer above it
NarrativeNarrative_WarmStartVsColdStart Warm-Start vs Cold-Start Review: Two Entry Points into Shared Core
ObservationObs_ScarletReviewKickoffQuote Scarlet's verbatim articulation of the desired review kickoff behavior after remember returns.
8 further nodes are not published in the public projection. Each node links from the entity index.