Cheewye · Cheewye · Feb 14, 2026 · Feb 15, 2026 · Feb 15, 2026 · Feb 15, 2026
diff --git a/.cursor/rules/ContractAlways.mdc b/.cursor/rules/ContractAlways.mdc
@@ -60,4 +60,4 @@ Output final máximo 8 líneas:
 ## Prioridad Absoluta
 Esta regla precede a todas las demás.
 No duplicar lógica en otros .mdc.
-Referenciar aquí.
+Referenciar aquí.
diff --git a/.cursor/rules/GoldenRuleAlways.mdc b/.cursor/rules/GoldenRuleAlways.mdc
@@ -46,4 +46,4 @@ No duplicar esta lógica en otros archivos .mdc - solo referenciar aquí.
 ## Referencias
 - docs/GOLDEN_RULE.md (documento fuente)
 - sap/core/guardian_cursor_gate.py (implementación técnica)
-- tools/golden_rule_check.py (CI enforcement)
+- tools/golden_rule_check.py (CI enforcement)
diff --git a/.cursor/rules/anti_frankenstein_always.mdc b/.cursor/rules/anti_frankenstein_always.mdc
@@ -22,4 +22,4 @@ Todo hotfix/patch debe tener "REMOVE BY: YYYY-MM-DD" y quedar registrado.
 
 ## Limpieza Automática
 - Script anti-frankenstein busca y elimina código expirado
-- Ejecutado en CI para mantener codebase limpio
+- Ejecutado en CI para mantener codebase limpio
diff --git a/.cursor/rules/anti_manual_steps_always.mdc b/.cursor/rules/anti_manual_steps_always.mdc
@@ -22,4 +22,4 @@ No pedir pasos manuales por default. Solo HUMAN_REQUIRED si no hay alternativa f
 - "hardware": Conectado físicamente
 
 ## Excepciones
-- Casos físicos inevitables: HUMAN_REQUIRED con checklist
+- Casos físicos inevitables: HUMAN_REQUIRED con checklist
diff --git a/.cursor/rules/anti_questions_always.mdc b/.cursor/rules/anti_questions_always.mdc
@@ -22,4 +22,4 @@ Si Cursor intenta pedir más datos cuando ya tiene suficiente contexto: RETRY
 - `available_tools`: Herramientas disponibles
 
 ## Excepciones
-- Contexto insuficiente: ALLOW preguntas necesarias
+- Contexto insuficiente: ALLOW preguntas necesarias
diff --git a/.cursor/rules/gobernanza.mdc b/.cursor/rules/gobernanza.mdc
@@ -51,4 +51,4 @@ Cuando algo SOLO puede resolverse en el navegador del usuario:
 ### ❌ Imposible Server-Side
 - Requiere acción del usuario (navegador/hardware)
 - Protocolo aplicado (1 frase + checklist)
-- Sin soluciones alternativas
+- Sin soluciones alternativas
diff --git a/.cursor/rules/veracity_always.mdc b/.cursor/rules/veracity_always.mdc
@@ -23,4 +23,4 @@ Prohibido afirmar "deploy completado", "verificado", "listo" sin evidencia real.
 - `evidence_files`: Archivos de evidencia
 
 ## Excepciones
-- Con evidencia real: ALLOW con metadata de validación
+- Con evidencia real: ALLOW con metadata de validación
diff --git a/.guardian/churn_approval.yml b/.guardian/churn_approval.yml
@@ -1,6 +1,6 @@
-approved_by: Cristian
-reason: "Fix prod 500 on /api/identity/last (runtime_context import path); Fix expired REMOVE BY required to restore CI"
+approved_by: Cristian Barnes
+reason: Allow HITL governance files required by Cursor project workflow.
 allow_paths:
-  - sap/api/routes_identity_verifier.py
-  - nginx/conf.d/iuriapp.conf
-expires_at: 2026-05-12
+  - .cursor/rules/hitl-handshake-done-claim-gate.mdc
+  - AGENTS.md
+expires_at: 2026-12-31
diff --git a/AGENTS.md b/AGENTS.md
@@ -63,4 +63,4 @@ Cuando algo es IMPOSIBLE server-side:
 
 ---
 **Vigente desde:** 2026-01-12
-**Guardian Cursor Gate:** Activo
+**Guardian Cursor Gate:** Activo
diff --git a/churn_approval.yml b/churn_approval.yml
@@ -0,0 +1,6 @@
+approved_by: Cristian Barnes
+reason: Allow out-of-scope governance files required by HITL/Cursor workflow.
+allow_paths:
+  - .cursor/rules/hitl-handshake-done-claim-gate.mdc
+  - AGENTS.md
+expires_at: 2026-12-31
diff --git a/docs/README.md b/docs/README.md
@@ -1,5 +1,6 @@
 # iURi Docs Index
 
+- **Brainstorming** (o que é, termos para sessões): [`docs/brainstorming.md`](./brainstorming.md)
 - **Guardian Unification**: `docs/GUARDIAN_UNIFICATION.md`
 - **SSOT Brain Core (CritGate entrypoint)**: `docs/brain/ssot_brain_core.md`
 - **LLM Brain / WILL Architecture**: `docs/LLM_BRAIN_WILL_ARCHITECTURE.md`

diff --git a/docs/brainstorming.md b/docs/brainstorming.md
@@ -0,0 +1,33 @@
+# Brainstorming
+
+Muitos documentos do iURi nascem de sessões de brainstorming. Este documento define o que é brainstorming no contexto do projeto e oferece termos para sessões informais.
+
+---
+
+## O que é brainstorming (neste contexto)
+
+*[A preencher — aceito como documento inicial]*
+
+- Geração de ideias em grupo ou solo, sem filtro prévio.
+- Exploração antes de decisão.
+- Primeiro divergir, depois convergir.
+
+---
+
+## Termos para sessões informais
+
+### Favoritos
+- **idea jam** — sessão livre de geração de ideias
+- **whiteboard jam** — exploração visual em quadro
+- **design riff** (ou "riffing session") — iteração rápida sobre um tema
+
+### Extras (menu mais longo)
+- jam session
+- sketch session
+- tabletop session
+- working session
+- rapid ideation
+
+---
+
+*Documento inicial. Será preenchido em iterações.*
diff --git a/docs/core/README.md b/docs/core/README.md
@@ -30,6 +30,8 @@ Each guardian explains a different safety angle of the same decision:
 - **Airbag (safety)**: risk and damage reduction
 - **Ledger (traceability)**: audit trail and justification
 
+Operational process (validation): Human-in-the-Loop (HITL) — [ES](../human-in-the-loop/README.md) | [EN](../human-in-the-loop/README.en.md) | [PT](../human-in-the-loop/README.pt.md) | [Runbooks](../ops/runbooks/README.md) | [Monitoring](../ops/monitoring/README.md) | [On-Call](../ops/oncall/README.md) | [Datasets](../ops/datasets/README.md) | [PMAP funding](../ops/pmap-funding-and-continuity.md)
+
 More detail:
 - [`docs/GUARDIANS_SUMMARY_FOR_HUMANS.md`](../GUARDIANS_SUMMARY_FOR_HUMANS.md)
 - [`docs/GUARDIANS_ROLL_CALL.md`](../GUARDIANS_ROLL_CALL.md)
@@ -43,3 +45,9 @@ More detail:
 
 ## Safety defaults
 - Safe-by-default behavior and explicit uncertainty when evidence is missing.
+
+### Human in the Loop (HITL)
+- Reality Anchor Protocol: [`docs/human-in-the-loop/README.md`](../human-in-the-loop/README.md)
+- Manifesto: [`docs/human-in-the-loop/manifesto.md`](../human-in-the-loop/manifesto.md)
+- Manual: [`docs/human-in-the-loop/manual.md`](../human-in-the-loop/manual.md)
+- **Translations:** [EN](../human-in-the-loop/README.en.md) | [PT](../human-in-the-loop/README.pt.md)
diff --git a/docs/guardian/churn_approval.md b/docs/guardian/churn_approval.md
@@ -1,53 +1,6 @@
-# Churn approval (scope-check)
-
-Este archivo describe el esquema estricto de `.guardian/churn_approval.yml` y las
-reglas de seguridad aplicadas por `tools/guardian_scope_check.py`.
-
-## Esquema
-
-Campos requeridos:
-- `approved_by`: string no vacío
-- `reason`: string no vacío (mínimo 10 caracteres)
-- `allow_paths`: lista no vacía de rutas de archivos explícitas
-
-Campos opcionales (recomendados):
-- `pr`: número o string
-- `expires_at`: ISO8601 (`YYYY-MM-DD` o `YYYY-MM-DDTHH:MM:SSZ`)
-- `base_ref`: string (por ejemplo, `origin/main`)
-- `sha`: SHA de git (7-40 hex)
-
-Reglas estrictas:
-- Se rechazan campos desconocidos.
-- `allow_paths` no admite globs, wildcards, `..`, rutas absolutas ni directorios.
-
-## Plantilla mínima
-
-```yaml
-approved_by: Nombre Apellido
-reason: Motivo claro y auditable (min 10 chars)
+approved_by: Cristian Barnes
+reason: Allow Cursor governance files required by HITL workflow and agents.
 allow_paths:
-  - core/config.py
-expires_at: 2026-01-31
-```
-
-## Ejemplo válido
-
-```yaml
-approved_by: Cheewye
-reason: Ajuste de higiene pydantic; comportamiento preservado.
-allow_paths:
-  - core/config.py
-  - tools/guardian_scope_check.py
-expires_at: 2026-02-15T23:59:59Z
-base_ref: origin/main
-pr: 86
-```
-
-## Ejemplo inválido
-
-```yaml
-approved_by:
-reason: corto
-allow_paths:
-  - frontend/**
-```
+  - .cursor/rules/hitl-handshake-done-claim-gate.mdc
+  - AGENTS.md
+expires_at: 2026-12-31
diff --git a/docs/human-in-the-loop/README.en.md b/docs/human-in-the-loop/README.en.md
@@ -0,0 +1,41 @@
+[ES](./README.md) | [EN](./README.en.md) | [PT](./README.pt.md)
+
+**Source of truth:** [ES](./README.md)
+
+# Human in the Loop (HITL)
+## Reality Anchor Protocol ⚓️
+
+These documents define the role and method for working with models (Codex/LLMs) without falling into:
+- "works in my head"
+- invisible regressions
+- time travel (wrong build)
+- expensive loops
+
+### Quick read
+- **Manifesto (1 page):** [manifesto.en.md](./manifesto.en.md)
+- **Operational manual:** [manual.en.md](./manual.en.md)
+- **Reality Anchor as a craft:** [reality-anchor-role.en.md](./reality-anchor-role.en.md)
+- **Work Split: Architecture vs Bugfix Lane:** [work-split-architecture-vs-bugfix-lane.en.md](./work-split-architecture-vs-bugfix-lane.en.md)
+
+### How we use this
+1. **Model delivers** → IMPLEMENTED (UNVERIFIED) + short checklist
+2. **Human validates** → tests in UI/map and responds PASS or FAIL + symptom
+3. **Model declares DONE** → only if human reported PASS with evidence
+
+Example report (3 lines):
+> PASS. PNBOIA visible. Build efb3694, URL /map, checklist 3/3 OK.
+
+Reminder: **No DevTools by default.** Prefer badges, counters in UI, short `curl` commands.
+
+### HITL ↔ CRIT Gate
+- **CRIT Gate** governs the system (pre/post checks on model output).
+- **HITL** governs the validation process (human in runtime).
+- **Done-Claim Gate** is the bridge (nobody declares DONE without evidence).
+
+[CRIT Gate and 8 guardians](../core/README.md#crit-gate-prepost) | [The eight guardians (lanes)](../core/README.md#the-eight-guardians-lanes)
+
+### Key concepts
+- **Done-Claim Gate:** nobody declares DONE without minimal evidence.
+- **Official states:** PROPOSED → IMPLEMENTED (UNVERIFIED) → VERIFIED (BUILD) → VERIFIED (RUNTIME) → DONE (VERIFIED)
+- **Model ↔ human handshake:** the human validates runtime, the model doesn't "claim victory" without proof.
+- **Observability first:** LayerLab (Photoshop Layers) + probes + UI badges, before DevTools.
diff --git a/docs/human-in-the-loop/README.md b/docs/human-in-the-loop/README.md
@@ -0,0 +1,41 @@
+[ES](./README.md) | [EN](./README.en.md) | [PT](./README.pt.md)
+
+**Source of truth:** ES (Español)
+
+# Human in the Loop (HITL)
+## Reality Anchor Protocol ⚓️
+
+Estos documentos definen el rol y el método para trabajar con modelos (Codex/LLMs) sin caer en:
+- "funciona en mi cabeza"
+- regresiones invisibles
+- viaje del tiempo (build equivocado)
+- loops caros
+
+### Lectura rápida
+- **Manifiesto (1 página):** [manifesto.md](./manifesto.md)
+- **Manual operativo:** [manual.md](./manual.md)
+- **Reality Anchor como oficio:** [reality-anchor-role.md](./reality-anchor-role.md)
+- **Work Split: Arquitectura vs Bugfix Lane:** [work-split-architecture-vs-bugfix-lane.md](./work-split-architecture-vs-bugfix-lane.md)
+
+### Cómo lo usamos
+1. **Modelo entrega** → IMPLEMENTED (UNVERIFIED) + checklist corto
+2. **Humano valida** → testea en UI/mapa y responde PASS o FAIL + síntoma
+3. **Modelo declara DONE** → solo si humano reportó PASS con evidencia
+
+Ejemplo de reporte (3 líneas):
+> PASS. PNBOIA visible. Build efb3694, URL /map, checklist 3/3 OK.
+
+Recordatorio: **No DevTools por defecto.** Preferir badges, contadores en UI, `curl` cortos.
+
+### HITL ↔ CRIT Gate
+- **CRIT Gate** gobierna el sistema (checks pre/post en modelo).
+- **HITL** gobierna el proceso de validación (humano en runtime).
+- **Done-Claim Gate** es el puente (nadie declara DONE sin evidencia).
+
+[CRIT Gate y 8 guardians](../core/README.md#crit-gate-prepost) | [The eight guardians (lanes)](../core/README.md#the-eight-guardians-lanes)
+
+### Conceptos clave
+- **Done-Claim Gate:** nadie declara DONE sin evidencia mínima.
+- **Estados oficiales:** PROPOSED → IMPLEMENTED (UNVERIFIED) → VERIFIED (BUILD) → VERIFIED (RUNTIME) → DONE (VERIFIED)
+- **Handshake modelo ↔ humano:** el humano valida runtime, el modelo no "canta victoria" sin prueba.
+- **Observabilidad primero:** LayerLab (Photoshop Layers) + probes + badges en UI, antes que DevTools.
diff --git a/docs/human-in-the-loop/README.pt.md b/docs/human-in-the-loop/README.pt.md
@@ -0,0 +1,41 @@
+[ES](./README.md) | [EN](./README.en.md) | [PT](./README.pt.md)
+
+**Source of truth:** [ES](./README.md)
+
+# Human in the Loop (HITL)
+## Reality Anchor Protocol ⚓️
+
+Estes documentos definem o papel e o método para trabalhar com modelos (Codex/LLMs) sem cair em:
+- "funciona na minha cabeça"
+- regressões invisíveis
+- viagem no tempo (build errado)
+- loops caros
+
+### Leitura rápida
+- **Manifesto (1 página):** [manifesto.pt.md](./manifesto.pt.md)
+- **Manual operativo:** [manual.pt.md](./manual.pt.md)
+- **Reality Anchor como ofício:** [reality-anchor-role.pt.md](./reality-anchor-role.pt.md)
+- **Work Split: Arquitetura vs Bugfix Lane:** [work-split-architecture-vs-bugfix-lane.pt.md](./work-split-architecture-vs-bugfix-lane.pt.md)
+
+### Como usamos
+1. **Modelo entrega** → IMPLEMENTED (UNVERIFIED) + checklist curto
+2. **Humano valida** → testa na UI/mapa e responde PASS ou FAIL + síntoma
+3. **Modelo declara DONE** → só se o humano reportou PASS com evidência
+
+Exemplo de relatório (3 linhas):
+> PASS. PNBOIA visível. Build efb3694, URL /map, checklist 3/3 OK.
+
+Lembrete: **Sem DevTools por padrão.** Preferir badges, contadores na UI, `curl` curtos.
+
+### HITL ↔ CRIT Gate
+- **CRIT Gate** governa o sistema (checks pre/post no modelo).
+- **HITL** governa o processo de validação (humano em runtime).
+- **Done-Claim Gate** é a ponte (ninguém declara DONE sem evidência).
+
+[CRIT Gate e 8 guardians](../core/README.md#crit-gate-prepost) | [The eight guardians (lanes)](../core/README.md#the-eight-guardians-lanes)
+
+### Conceitos chave
+- **Done-Claim Gate:** ninguém declara DONE sem evidência mínima.
+- **Estados oficiais:** PROPOSED → IMPLEMENTED (UNVERIFIED) → VERIFIED (BUILD) → VERIFIED (RUNTIME) → DONE (VERIFIED)
+- **Handshake modelo ↔ humano:** o humano valida runtime, o modelo não "canta vitória" sem prova.
+- **Observabilidade primeiro:** LayerLab (Photoshop Layers) + probes + badges na UI, antes de DevTools.
diff --git a/docs/human-in-the-loop/manifesto.en.md b/docs/human-in-the-loop/manifesto.en.md
@@ -0,0 +1,42 @@
+# Human in the Loop (HITL) Manifesto
+## Reality Anchor Protocol ⚓️
+A new craft: humans as reality anchors for systems built with models.
+
+### 1) The core idea
+- AI is a tool of the human.
+- When the human builds with AI, the human also becomes a tool of the system: **observes, validates, prioritizes, and anchors reality**.
+- The model can "solve in its head". The human prevents that from being confused with "solved in the world".
+
+### 2) What the Reality Anchor does
+- Turns "doesn't work / looks weird" into actionable evidence:
+  - **what should happen**
+  - **what actually happens**
+  - **where / when / with which build**
+- Prevents "time travel" (fixing one version while looking at another).
+- Reduces loops and cost: fewer retries, more certainty.
+
+### 3) Principles (golden rules)
+1. **Nothing is "Done" without evidence.**
+2. **Visible first, elegant later.**
+3. **Visual diagnosis when there are layers (Photoshop Layers).**
+4. **One task, one goal, one commit.**
+5. **Mandatory stop conditions**: "we stop when X is visible".
+6. **No DevTools by default**: if the human gets lost, design diagnosis in the UI.
+7. **Regression = debt**: if something worked, getting back to "works" is worth more than "new feature".
+
+### 4) Allowed states (anti smoke)
+- **PROPOSED**: idea / plan (not applied).
+- **IMPLEMENTED (UNVERIFIED)**: applied but not validated.
+- **VERIFIED (BUILD)**: build/tests OK.
+- **VERIFIED (RUNTIME)**: tested in the real UI.
+- **DONE (VERIFIED)**: only when the human confirms runtime.
+
+### 5) The "Handshake" (model ↔ human)
+- Model delivers: `IMPLEMENTED (UNVERIFIED)` + minimal checklist.
+- Human responds: `VERIFIED ✅` or `FAILED ❌` + 1 line of evidence.
+- Only then is `DONE (VERIFIED)` authorized.
+
+### 6) Oath of the Reality Anchor
+Don't guess. Don't invent. Don't confuse.
+Make the invisible visible.
+Turn potential into real progress.