This hub gathers the mechanics of how feedback enters and reshapes the product and its knowledge. Two structural loops anchor the space: the [save-memory/PR round semantics] (Position_ReviewMemorySemantics) — save-memory ends a round, merge closes the PR, a corrected review always takes at least two rounds — and the [five-step annotate→document→save→re-annotate cycle] (Observation_FeedbackLoop) where annotation surfaces unsupported claims and re-annotation confirms the shift. [Convergence is framed as asymptotic] (Obs_ConvergenceAsymptotic) accumulation of review evidence, not a binary threshold. On process, the [real-time per-section loop] (Narrative_RealTimeFeedbackLoop) (Mar 19) is held superior to batch review — errors caught before they compound, insights surfaced in the moment. A dense run of Scarlet corrections (Mar 19) sharpens product framing: [expertise on-demand not ambient] (Obs_AmbientBehaviorCorrection), [perspective vs local RAG] (Obs_CustomGPTCorrection), [expanding experts not replacing] (Obs_ExpertReachCorrection), and [Tinkerer's existing solution as barrier] (Obs_TinkererBarrierCorrection). Later signals: [text-for-topic precedent] (Obs_ScarletTextForTopicQuote) (Apr 8) as rationale for the research branch methodology, and [feedback/help tool as churn prevention] (Obs_FeedbackHelpQuote) (Mar 26). [Open rhetorical questions] (StyleObs_OpenQuestions) surface unresolved design decisions as bounded scopes.
Witness phrases
verbatim The product behavior is expertise at the moment of need, not ambient
verbatim Custom GPTs don't provide perspective — they're just local RAG
verbatim We're not replacing experts — we're expanding their reach
verbatim The Tinkerer doesn't activate fastest — their existing solution is a barrier
verbatim This feels like a similar moment to the text for topic moment
verbatim Correction 7: Save memory = corrections exist, ends round, PR updates. Merge = no corrections needed, closes PR. A review with corrections always takes at least two rounds.
verbatim Real-time per-section feedback is held superior to batch review, but the asymptotic convergence framing implies no session fully closes — these two sit in tension about what 'done' means.
verbatim Tinkerer archetype corrected from 'fastest activator' to 'barrier case' — a prior hypothesis about adoption sequence was reversed by Scarlet's feedback, not refined.
verbatim Feedback/help tool framed as churn prevention (operational/retention lens) sits apart from the epistemic feedback loops that dominate the hub — a different register of 'feedback.'
verbatim Open Questions as bounded rhetorical scopes is a structural/stylistic observation about how unresolved decisions surface — distinct from the process loops and product corrections elsewhere in the hub.
In this area
Narrative_RealTimeFeedbackLoop Real-Time Feedback Loop: Read and React as Each Section Is Written
ObservationObs_AmbientBehaviorCorrection Scarlet's correction establishing the on-demand vs. ambient distinction for product behavior.
ObservationObs_ConvergenceAsymptotic Convergence is asymptotic accumulation of review evidence, not a binary threshold. More honest to how organizational knowledge actually works.
ObservationObs_CustomGPTCorrection Scarlet's correction distinguishing Custom GPTs from aswritten's perspective-based approach.
ObservationObs_ExpertReachCorrection Scarlet's correction reframing the expert relationship — expansion not replacement.
ObservationObs_FeedbackHelpQuote Feedback/help tool as primary churn-prevention mechanism
ObservationObs_ScarletTextForTopicQuote Scarlet's verbatim articulation of the text-for-topic precedent and the canon-too-soon drift mechanism — the founding rationale for the research branch methodology.
ObservationObs_TinkererBarrierCorrection Scarlet's correction to the AI's hypothesis about the Tinkerer archetype — their existing solution is a barrier, not an accelerant to adoption.
StyleObservationStyleObs_OpenQuestions Open Questions section uses direct rhetorical questions to surface unresolved design decisions — each question is a bounded scope
1 further node is not published in the public projection. Each node links from the entity index.