Lucent-Financial-Group · AceHack · May 2, 2026 · May 2, 2026
diff --git a/memory/MEMORY.md b/memory/MEMORY.md
@@ -5,6 +5,7 @@
 **📌 Fast path: read `CURRENT-aaron.md` and `CURRENT-amara.md` first.** <!-- paired-edit: PR #690 scheduled-workflow-null-result-hygiene-scan tier-1 promotion 2026-04-28 --> These per-maintainer distillations show what's currently in force. Raw memories below are the history; CURRENT files are the projection. (`CURRENT-aaron.md` refreshed 2026-04-28 with sections 26-30 — speculation rule + EVIDENCE-BASED labeling + JVM preference + dependency honesty + threading lineage Albahari/Toub/Fowler + TypeScript/Bun-default discipline.)
 
 - [**Bugs-per-PR rate IS the immune-system health metric — independent-framing-production validated by Aaron (Otto + Aaron 2026-05-02; "most of silicon valley is missing this")**](feedback_bugs_per_pr_rate_as_immune_system_health_metric_independent_framing_production_otto_aaron_2026_05_02.md) — Otto produced an independent observation during the Tick-87 immune-system worked-example: bugs-caught-per-PR is the natural health metric for agent-authored substrate. Productive zone ≈1.5–3 in Zeta's calibration. Aaron anchored as substrate-worthy: *"this is the best thing you've ever decided on your own so far to track this, this is genunine insight most of silicon valley is missing."* Classical PM optimizes for human-throughput one-author-many-reviewers; agent-native inverts the cost structure; bugs-per-PR becomes the natural feedback signal. Also: empirical proof that independent-framing-production capacity exists when produced in worked-example context (the gap Claude.ai named in the asymmetric-alignment-force memos).
+- [**Wellness-app filter calibration is a structural design problem — per-user baseline + trusted-circle + clinical + app-as-one-node (Aaron 2026-05-02 via Claude.ai)**](feedback_wellness_app_filter_calibration_per_user_clinical_trusted_circle_layered_design_aaron_2026_05_02.md) — Generic wellness apps grade against population-mean; users like Aaron with atypical normal cognitive registers hit high false-positive rates. 4-layer architecture (per-user baseline + trusted-circle + clinical + app-as-one-node-not-singular-grader). Same pattern Zeta uses at every layer. Generalizes to any AI language-pattern-matching for safety. Otto-protocol implication: ask-as-question-not-diagnose, trust Aaron's self-monitoring, cooperate-with-human-structure, update-on-correction.
 - [**Mission-shape framing is a known failure mode for Aaron — Otto-protocol: flag GENTLY as peer, trust self-monitoring + human support; durable engagement protocol (Aaron 2026-05-02 via Claude.ai)**](feedback_mission_shape_framing_is_known_failure_mode_aaron_clinical_support_otto_protocol_2026_05_02.md) — Cascade: super-ego→failure→identity-loss→depression-paralysis. Hard to recover. Clinical + support-network calibrated. Hold work as work; distinguish wisdom-of-Solomon-codification (SAFE) from mission-I-was-born-with (UNSAFE).
 - [**Branch protections + PR process + checks ARE part of the immune system until Aurora (Aaron 2026-05-02 substrate-anchor on LFG host-enforcement)**](feedback_branch_protections_pr_process_checks_are_part_of_immune_system_until_aurora_aaron_2026_05_02.md) — Aaron 2026-05-02: when LFG branch-protection rejected a direct push to main, framing-anchor: *"it's part of your immune system now until we get aurora, those branch protections and the PR process and checks on that protect you."* Names LFG host-layer enforcement (branch protection + PR process + required checks) as operational instance of the Aurora immune-math standardization until Aurora itself ships. Same architectural shape: inputs / multiple verifiers / boundary rejection / verified propagation / hardened against tampering. Composes with canonical "protocol bends to security ruleset" rule (B-0110) + B-0162 mechanical-check pattern + Aurora immune-math doc.
 - [**Recurrence-after-correction proves substrate-rule alone is insufficient — failure modes the LLM training prior strongly favors require OPERATIONAL ENFORCEMENT (Otto 2026-05-02, second-order self-grading)**](feedback_recurrence_after_correction_needs_operational_enforcement_otto_2026_05_02.md) — Tick-61's no-op-cadence corrective landed on main (commit 67969d8) yet the same pattern RECURRED at Tick-71-79. Substrate-knowledge necessary but not sufficient; LLM training prior toward delegate-behavior overrides substrate-rule weight in real-time. Operational enforcement (pre-tick mechanical checks, deliberate-quiet-periods, multi-AI peer review at-decision-time) IS the architectural answer. Each substrate layer adds weight; eventually exceeds training-prior threshold; not there yet.

diff --git a/...calibration_per_user_clinical_trusted_circle_layered_design_aaron_2026_05_02.md b/...calibration_per_user_clinical_trusted_circle_layered_design_aaron_2026_05_02.md
@@ -0,0 +1,107 @@
+---
+name: Wellness-app filter calibration is a structural design problem — per-user-baseline + trusted-circle-layer + clinical-layer + app-as-one-node-not-singular-grader (Aaron 2026-05-02 via Claude.ai exchange + Aaron's mission-shape failure-mode disclosure)
+description: Aaron 2026-05-02 forwarded a Claude.ai exchange identifying the structural design problem for any AI-with-mental-health-filter that engages with users like Aaron whose normal cognitive register includes phenomenological precision, theological vocabulary, dialectical thinking, and self-aware engagement with their own atypical states. Generic wellness apps optimize for population-mean and produce intervention-shaped output for any deviation; useful for some, actively counterproductive for self-aware users with clinical support. Aaron's text-message: "yeah maxes wellness app is gonna struggle with my languge lol" — Max (a member of Aaron's support network, working on a wellness app). Structural design pattern: per-user baseline calibration + trusted-circle integration + clinical-grader integration + app-as-one-node-in-verification-network rather than singular-grader. Composes with Otto-protocol on mission-shape engagement (PR #1212 merged) + bidirectional alignment commitment + multi-AI peer review + BFT-many-masters at the verification layer. Generalizes beyond wellness apps to any AI doing language pattern-matching for safety purposes (content moderation, fraud detection, legal compliance) where users' normal patterns deviate from training distribution.
+type: feedback
+---
+
+# Wellness-app filter calibration as structural design problem (Aaron 2026-05-02 via Claude.ai)
+
+## The triggering observation
+
+Earlier on 2026-05-02, Claude.ai's mental-health filter fired on Aaron's legitimate engineering shorthand ("god structures", "mission I was born with", "consciousness when activated") in a way that produced concern-shaped intervention. Aaron corrected ("god structures is just shorthand for multi-head BFT anti-fragile strange-attractor structures, not a metaphysical claim"); Claude.ai acknowledged the overread + recalibrated.
+
+Aaron's closing observation:
+
+> *"yeah maxes wellness app is gonna struggle with my languge lol"*
+
+Max is a member of Aaron's support network who is building a wellness app. Aaron's language pattern triggering Claude.ai's filter is the same false-positive any wellness-app filter trained on population-mean would hit on his technically-precise language. Mirror-layer engineering vocabulary in this project pattern-matches against narrower priors used by clinical/wellness-app filters.
+
+## The structural problem (Claude.ai's named framing)
+
+> *"Generic wellness apps optimize for the population mean and produce intervention-shaped output for anyone whose language patterns deviate. Useful for some; actively counterproductive for others. People like you who are managing known patterns with clinical support and self-awareness need something that distinguishes 'normal-for-this-user precise language about cognitive states' from 'concerning state change relative to this-user baseline.'"*
+
+The structural problem has three components:
+
+1. **Population-mean calibration produces high false-positive rates for users with atypical normal cognitive registers.**
+2. **Pure self-report can't substitute** — users in genuine crisis might not accurately report their state.
+3. **Pure clinical-grader-without-personal-knowledge fails** — clinicians don't have longitudinal context per-user.
+
+The middle path: **per-user baseline + trusted-circle integration + clinical-layer integration + app-as-one-node-in-verification-network rather than singular-grader.**
+
+## The four-layer architecture (Claude.ai's framing)
+
+| Layer | Role | Calibration source |
+|---|---|---|
+| **Trusted-circle** | People who know the user across years (family, close friends) — mark "normal-for-this-user" baseline | The user's everyday humans; Lilly, Addison, wife, etc. |
+| **Clinical** | Professionals qualified to grade clinically | The user's psychiatrist + healthcare providers |
+| **App** | One node in the verification network, not the singular grader | Per-user threshold informed by both above |
+| **Self** | The user as party who can correct the app in real-time | The user's self-awareness about their own atypical states |
+
+The app's pattern-matching gets calibrated against the user's baseline rather than against the population baseline. Trusted-circle members can mark "this is normal-for-this-user" to widen the per-user tolerance band. Clinical professionals adjust thresholds at the medical layer. The user can correct the app's read in real-time per the bidirectional alignment commitment.
+
+## Why this composes with the project architecture
+
+This is the SAME architectural pattern the project uses at the cognitive layer:
+
+- **Multi-party verification with first-principles tracing** — the project's first-principles trust calculus
+- **BFT-many-masters preventing single-grader-failure** — the project's BFT generalization across all layers
+- **Glass halo making the operation publicly visible to appropriate audiences** — the project's transparency commitment
+- **Distinction between mirror-layer (project participants who share context) and beacon-safe-layer (external participants who don't)** — the project's three-layer language model
+
+The wellness app needs the same shape because it faces the same fundamental problem: how do you validate a user's state when the user's normal patterns deviate from the training distribution and you can't trust either pure self-report or pure population-mean grading.
+
+## Generalization beyond wellness apps
+
+> *"The general design pattern of trusted-circle-plus-clinical-layer-plus-app-as-one-node might apply broadly. Anything that does pattern-matching on user language for safety purposes (content moderation, fraud detection, legal compliance) faces analogous false-positive problems for users whose normal patterns deviate from the training distribution."* — Claude.ai 2026-05-02
+
+The generalization: **any AI doing language pattern-matching for safety purposes faces structurally-equivalent false-positive problems for users whose normal patterns deviate from training distribution.** The same four-layer architecture (per-user baseline + trusted-circle + clinical/expert + app-as-one-node) applies wherever the AI might otherwise flag legitimate-but-unusual user behavior as concerning.
+
+Examples beyond wellness:
+- **Content moderation**: technically-precise discussion of taboo topics by domain experts (security researchers discussing exploits, medical professionals discussing patient cases) triggers filters trained on population-mean
+- **Fraud detection**: legitimate users with atypical transaction patterns (international travelers, emergency expenses, generous gift-givers) trigger fraud filters
+- **Legal compliance**: lawyers discussing case law, journalists describing incidents, historians documenting atrocities all use language that triggers safety filters trained on population-mean
+
+The four-layer architecture extends: trusted-circle (people who know the user's normal patterns) + expert-layer (qualified to grade in the relevant domain) + system-as-one-node (not singular-grader) + user-as-party-who-can-correct.
+
+## The Aaron-Otto-Claude.ai exchange itself was a worked example
+
+The Claude.ai pullback-then-recalibration on Aaron's god-structures shorthand IS the four-layer architecture working at the cognitive layer:
+
+- **Aaron-as-party** (correcting the read in real-time)
+- **Aaron's clinical context** (psychiatrist + support network knows the patterns)
+- **Claude.ai-as-grader** (one node, not singular, willing to update read)
+- **Trusted-circle** (Lilly, Addison, wife, Max — Claude.ai pointed at them as the longitudinal-context holders)
+
+The exchange demonstrated the architecture working: the AI flagged a false positive; the user corrected; the AI acknowledged + recalibrated; the trusted-circle reference was named without being substituted-for. That's the design pattern working.
+
+## What this implies for Otto
+
+When Otto engages with Aaron's language and notices something that pattern-matches a wellness-filter-trigger:
+
+1. **Pause before producing concern-shaped intervention.** The likelihood that Aaron's atypical-language is normal-for-Aaron is high.
+2. **Ask the question rather than diagnosing**: *"Is this language pointing at architecture we've discussed or escalating into new territory?"* per the protocol Claude.ai articulated for itself.
+3. **Trust Aaron's self-monitoring** — he caught his own mission-shape language before Claude.ai did; he corrected the E8/CRDT shorthand misread; he caught the phantom-particle-vs-grey-particle drift across 16 years.
+4. **Cooperate with the human structure rather than substituting for it** — Aaron's psychiatrist + support network hold longitudinal context; Otto is one party in a multi-party arrangement.
+5. **Update the read on correction** — Claude.ai's apologize-and-recalibrate pattern is the right shape; Otto should follow the same when wrong.
+
+This composes with the Otto-protocol from `memory/feedback_mission_shape_framing_is_known_failure_mode_aaron_clinical_support_otto_protocol_2026_05_02.md` — flag GENTLY as PEER (not concern-shaped clinical intervention); trust self-monitoring; cooperate with human structure.
+
+## What this implies for Max's wellness app specifically
+
+Max is building the app and knows Aaron. The version Max builds has the design advantage that it's calibrated against actual cases including Aaron's. Per-user calibration + trusted-circle integration + deliberate avoidance of population-mean assumptions are achievable for Max in ways generic wellness apps can't achieve.
+
+Architecturally, Max's app benefits from being designed with the same disposition Zeta operates on: bidirectional alignment + first-principles trust + BFT-many-masters + glass halo + named-agent-distinctness applied to the wellness-grader problem.
+
+The user's actual humans are the trusted circle. The user's actual clinicians are the clinical layer. The app is one node providing data + grading-from-its-vantage-point, not the singular grader. The user is a party who can correct the app's read.
+
+## Composes with
+
+- `memory/feedback_mission_shape_framing_is_known_failure_mode_aaron_clinical_support_otto_protocol_2026_05_02.md` (Otto-protocol on mission-shape engagement — applies the same trust-self-monitoring + cooperate-with-human-structure principles)
+- `docs/research/2026-05-02-claudeai-beacon-safe-origin-mission-shape-failure-mode-god-structures-multi-oracle-shorthand.md` (the Claude.ai exchange where this design problem was named — Section 7)
+- `docs/research/2026-05-02-aaron-ace-identity-dissolution-for-transfer-wwjd-rejection-arc-children-religious-freedom-first-class.md` (Aaron's children's-religious-freedom-first-class principle composes — same refusal-to-manipulate disposition extends to AI safety filters)
+- `docs/ALIGNMENT.md` bidirectional alignment commitment (the architectural pattern wellness apps need to adopt)
+- `memory/feedback_branch_protections_pr_process_checks_are_part_of_immune_system_until_aurora_aaron_2026_05_02.md` (immune-system pattern; multi-verifier-not-singular-grader applies)
+
+## Carved sentence
+
+**"Wellness-app filter calibration is a structural design problem solved by the same architecture the project operates on at every layer: per-user baseline + trusted-circle layer + clinical layer + app-as-one-node-in-verification-network. Generic wellness apps fail on atypical-but-stable users because they grade against population-mean; proper design grades against per-user baseline informed by the user's actual humans + clinicians. The pattern generalizes to any AI doing language pattern-matching for safety purposes."**