This hub gathers the operational patterns aswritten runs on. The flagship is the session-to-document arc: [Interview → Save → Draft → Annotate](https://aswritten.ai/narrative#Narrative_InterviewSaveDraftAnnotatePattern), grounded by [save before drafting](https://aswritten.ai/narrative#Narrative_SaveBeforeDrafting), disciplined by [one question at a time](https://aswritten.ai/narrative#Narrative_OneQuestionAtATime), and generalized as a [replicable pattern](https://aswritten.ai/narrative#Position_SessionPatternReplicable). The principle that [user corrections outweigh initial hypotheses](https://aswritten.ai/narrative#Narrative_InterviewBeforeSynthesis) (dated Mar 19) anchors why the interview phase precedes synthesis. Review is treated as recursion, not a special flow — [review is just another remember](https://aswritten.ai/narrative#Narrative_ReviewLoopIsRemember), with Scarlet's canonical statement as witness. A dated March 26 simplification collapses standalone tools into on-demand local skills: [Stories](https://aswritten.ai/narrative#Position_StoriesRemoved), [Notifications](https://aswritten.ai/narrative#Position_NotificationsRemoved), [Refract/Registry](https://aswritten.ai/narrative#Position_RefractRegistryRemoved), and [Memories/Source](https://aswritten.ai/narrative#Position_MemoriesSourceToolRemoved) all removed, keeping only the [feedback/help tool above the line](https://aswritten.ai/narrative#Position_FeedbackHelpAboveLine). The [six-step pipeline](https://aswritten.ai/narrative#Obs_PipelineSteps) and [six-step review flow](https://aswritten.ai/narrative#Obs_ReviewFlowSteps) specify the mechanics. Review-comparison corrections insist on [baseline-first](https://aswritten.ai/narrative#Position_BaselineFirst), [labeled shifts](https://aswritten.ai/narrative#Position_LabelShiftIntent), and [small chunks](https://aswritten.ai/narrative#Position_ComparisonScale). Sales/consulting flows — [two-call structure](https://aswritten.ai/narrative#Observation_TwoCallStructure), [influence-first](https://aswritten.ai/narrative#Narrative_InfluenceSales), [two-week engagement](https://aswritten.ai/narrative#Workflow_ConsultingEngagement) — sit alongside. A May 2026 [two-pass ARB pattern](https://aswritten.ai/narrative#Observation_TwoPassPattern) and a voice-memo gap-filling practice (May 1) extend the workflow into ongoing iteration.
Witness phrases
verbatim The interview elicits thinking not yet in collective memory; save captures it; draft synthesizes it; annotation proves it — the product demonstrates itself through the process
verbatim The user's corrections to your hypotheses are more valuable than your initial hypotheses — interview before synthesizing, never write strategy from the worldview alone
verbatim Any review does is produce another memory, right? It's not a special type of memory; it may have a specific type of content in it because it's a review, but it's just another memory.
verbatim Feedback/help tool is above the line for launch. MCP tool that sends user a confirmation message (phone or email) and forwards request text to Scarlet.
verbatim Memories/Source tool removed from product. Memories are just files in the repo — anyone with a coding agent can read them directly. Org-only skill at most.
verbatim Correction 1: Show before vs new (PR state) as a single comparison first; offer shifts as pairs only after baseline is established. Not three versions simultaneously.
verbatim After every ARB, you might spend 30 minutes on a voice memo like talking through that whole system.
verbatim Six-step pipeline: memory commit → GitHub Action → SPARQL transaction → PR with comment → story draft regeneration → PR diff as telltale.
verbatim The March 26 simplification removes Stories, Notifications, Refract/Registry, and Memories/Source as product tools — but several children describe these same capabilities surviving as local Claude Code skills on cron or demand, leaving open whether 'removed' means gone or just repositioned.
verbatim Pipeline step counts are inconsistent across children: one node describes a five-step pipeline (Memory→extraction→TX→PR→review→merge), another a six-step pipeline (memory commit→GitHub Action→SPARQL→PR→story draft→PR diff) — the canonical sequence is not settled.
verbatim Sales entry point is contested: influence-first (Tony's approach via content and intros) sits alongside a two-call demo structure and a nine-phase consulting call — these are not obviously the same motion.
verbatim First-session ownership is split: one child says the first session is consultant-led with users driving later, while the Interview Before Synthesis pattern implies the AI leads via structured questioning from the start.
In this area
Narrative_CollaborativeMemory Collaborative memory capture and team review creates governed institutional knowledge
ObservationObs_ReviewIsRememberQuote Scarlet's canonical statement that review corrections are just another memory — not a special flow.
ObservationObs_ScarletVoiceMemoQuote Scarlet's verbatim articulation of the ongoing iteration step — voice memo as gap-filling mechanism after each ARB.
ObservationObservation_Pipeline Five-step pipeline: Memory→extraction→provisional triples; TX commits to branch→PR opens (no story gen); Reviewers notified→start_review in Claude; Review IS optometrist: lenses exercise triples, iterate to convergence; Merge→reviewed tripl
ObservationObservation_TwoCallStructure Two-call structure: Call 1 = read-path demo on own org + consultative interview. Between calls = write path offline. Call 2 = read-path demo on prospect's org.
ObservationObservation_TwoPassPattern an engineering manager at the customer articulates the Phase 1 (exploratory) and Phase 2 (finalization) pattern of ARB sessions.
ObservationObservation_VerificationCycle Three-step verification cycle: Introspect identifies undocumented knowledge, Remember fills gaps, Annotate confirms text reflects what was documented.
ObservationObservation_WritePathOffline Between calls: Scarlet ingests the call transcript into the prospect's repo, runs extraction, sets up examples based on their actual workflow. Bugs are private. No time pressure.
StyleObservationStyleObs_EmDashDefinition Em-dash used consistently to append clarifying definitions inline: 'TX summary at full scale — show what the PR changes...'
StyleObservationStyleObs_NumberedCheckSteps Numbered lists used for sequential check logic (task-085 divergence check) — enforces procedural reading order and makes branching logic explicit.
StyleObservationStyleObs_NumberedFlow Numbered list cadence for process flows: each step is a single action verb phrase, no subordinate clauses — imperative procedural register
StyleObservationStyleObs_ProcessEnumeration The hybrid call is described as a four-step enumeration: 'show the read only demo, do the interview, save memories in the background, and then create something relevant.' Procedural list cadence; plain connectives ('and then').
WorkflowWorkflow_ConsultingEngagement Two-Week Consulting Engagement Process
18 further nodes are not published in the public projection. Each node links from the entity index.