chore(docs): fix cross-references after 3-bucket reorg

AGENTS.md

@@ -14,7 +14,7 @@ PVL is developed by Said Azaizah using a multi-terminal AI-orchestration pattern
 - Writes `TX-INSTRUCTIONS.md` (gitignored) and `docs/active/RESEARCH_BRIEFS/T5_PROMPTS/*.md` specs
 - Commits sub-terminal output after review (atomic logical commits)
 - Updates ADRs in `docs/active/DECISIONS.md`
-- Updates the dispatch state in `docs/active/STATUS.md` after every cycle
+- Updates the dispatch state in `docs/internal/STATUS.md` after every cycle
 - Saves memory feedback in `~/.claude/projects/.../memory/`
 - Proposes workflow upgrades proactively (hooks, slash commands, new memory files) without waiting for permission — opt-out model
 - Makes architectural decisions when evidence supports them; surfaces explicitly when human-touch input needed
@@ -104,7 +104,7 @@ PVL is developed by Said Azaizah using a multi-terminal AI-orchestration pattern
 **Scope**:
 - Reads the full Peleg feedback document end-to-end
 - Translates symptoms into root-cause tasks (per `feedback_root_cause_workflow.md`)
-- Updates `docs/active/PELEG_REVIEW_TASKS.md` and `docs/active/ALEX_BACKLOG.md` if relevant
+- Updates `docs/internal/PELEG_REVIEW_TASKS.md` and `docs/internal/ALEX_BACKLOG.md` if relevant
 
 **Reads**: the feedback artifact + `PELEG_REVIEW_TASKS.md` history.
 
@@ -156,8 +156,8 @@ PVL is developed by Said Azaizah using a multi-terminal AI-orchestration pattern
 
 ## Reference files
 
-- `docs/active/STATUS.md` — current dispatch state + manual test plan + per-readiness work queue
-- `docs/active/MASTER_PUSH_PLAN.md` — 7-wave plan to full-platform vision
+- `docs/internal/STATUS.md` — current dispatch state + manual test plan + per-readiness work queue
+- `docs/internal/MASTER_PUSH_PLAN.md` — 7-wave plan to full-platform vision
 - `docs/active/ROADMAP.md` — Phases A-O detailed tasks
 - `docs/active/DECISIONS.md` — all ADRs
 - `docs/active/RESEARCH_BRIEFS/_INDEX.md` — research brief log

docs/active/ACTIVE_CONTEXT.md

@@ -240,4 +240,3 @@ cd ui && npm install && npm run dev
 | `docs/active/DEPLOYMENT.md` | VM specs, step-by-step deployment, K8s plan |
 | `docs/active/MASTER_DEV_DOC.md` | Strategic decisions, security, risk register |
 | `docs/active/ROADMAP.md` | Development roadmap and completed work |
-| `README_EXPLAINER.md` | Non-technical A-Z team guide |

docs/active/DECISIONS.md

@@ -178,10 +178,10 @@ See also: `TECH_PLATFORM_VISION.md` for the longer-form platform thesis and tech
 
 **Date**: 2026-05-12 · **Status**: ACCEPTED · **Authors**: Said + T-RES (RB-004) + T1
 **Context**: PVL uses 4-6 Claude Code terminals (T1 CEO, T2 backend, T3 frontend, T5 research, Cowork visual, T-PEL Peleg-feedback). No committed file legibly described which terminal does what, what's forbidden, and how they coordinate. Said had to invent the multi-terminal pattern, the CEO role, and the T5 distinction — symptom of T1 not proposing infrastructure proactively (`feedback_t1_proactive_workflow_evolution.md`). Without this protocol committed at repo root, a future maintainer or Said-after-MIT-absence cannot rebuild context.
-**Decision**: Adopt **`AGENTS.md`** at project root (committed, public) as the canonical role map. Every terminal reads it at session start. Adopt **`docs/active/STATUS.md`** as the always-up-to-date dispatch dashboard — T1 updates it after every cycle. Per-terminal instructions (`TX-INSTRUCTIONS.md`) stay gitignored — those are working specs, not canonical roles.
+**Decision**: Adopt **`AGENTS.md`** at project root (committed, public) as the canonical role map. Every terminal reads it at session start. Adopt **`docs/internal/STATUS.md`** as the always-up-to-date dispatch dashboard — T1 updates it after every cycle. Per-terminal instructions (`TX-INSTRUCTIONS.md`) stay gitignored — those are working specs, not canonical roles.
 **Reasoning**: AGENTS.md is the legible map; STATUS.md is the live state. Together they let a fresh terminal — or Said after a 3-month gap — re-enter the project without T1 context. Addy Osmani's multi-agent research (RB-004 §3) confirms human-curated AGENTS.md files improve agent success vs AI-written ones. The 3-5 effective-terminal sweet spot matches PVL's persistent set (T1/T2/T3/T5).
 **Implication**: T1 maintains AGENTS.md (proactive review quarterly) + updates STATUS.md after every dispatch. Sub-terminal instructions docs reference AGENTS.md rather than re-declaring roles. Any new terminal role (e.g., future T-OPS for observability) requires an AGENTS.md update first.
-**Evidence**: `AGENTS.md` (this commit), `docs/active/STATUS.md`, RB-004 §3 Domain 3, RB-COWORK-AUDIT.
+**Evidence**: `AGENTS.md` (this commit), `docs/internal/STATUS.md`, RB-004 §3 Domain 3, RB-COWORK-AUDIT.
 
 ---
 

docs/active/DEVELOPER_REFERENCE.md

@@ -678,5 +678,5 @@ Results.tsx
 ---
 
 *For high-level architecture decisions, see `docs/active/MASTER_DEV_DOC.md`.
-For non-technical explanation, see `README_EXPLAINER.md`.
+For non-technical explanation, see the project [README.md](../../README.md).
 For API contracts, see `docs/active/CONTRACTS.md`.*

docs/active/RESEARCH_BRIEFS/RB-005_workflow-and-ai-platform-deep.md

@@ -471,7 +471,7 @@ Hannah cares about: **audit trail + citation correctness**. She submits a JOSS p
 - `backend/services/vector_store.py` — extend with `papers` table support
 - `docs/active/DECISIONS.md` — append ADR-020 (G2 architecture)
 - `docs/active/ROADMAP.md` — add Wave 3 G2 task breakdown
-- `docs/active/MASTER_PUSH_PLAN.md` — Wave 3 detail update
+- `docs/internal/MASTER_PUSH_PLAN.md` — Wave 3 detail update
 
 **No code shipped in this brief.** All implementation routes through T2/T3/T1 dispatch.
 

docs/active/RESEARCH_BRIEFS/T5_PROMPTS/T5-DEEP-001_workflow-and-ai-platform.md

@@ -84,14 +84,14 @@ These docs (in `docs/active/`) ARE the project state. T5 reads them all, no exce
 
 **AI-platform context:**
 - `docs/active/ROADMAP.md` Phase G (lines ~704-790) — Alex's G1/G2/G3 vision verbatim
-- `docs/active/ALEX_BACKLOG.md` "AI/LLM Integration" section — AI1/AI2/AI3
-- `docs/active/TECH_PLATFORM_VISION.md` §4 (AI for users / devs / science split)
+- `docs/internal/ALEX_BACKLOG.md` "AI/LLM Integration" section — AI1/AI2/AI3
+- `docs/internal/TECH_PLATFORM_VISION.md` §4 (AI for users / devs / science split)
 - `docs/active/DECISIONS.md` ADR-009 (MCP front door, ACCEPTED), ADR-017 (ESM-2 embedding)
 - `docs/active/RESEARCH_BRIEFS/RB-001_researcher-needs.md` — researcher feature priorities
 - `docs/active/RESEARCH_BRIEFS/RB-002_vector-store-evaluation.md` — LanceDB rationale (RAG-relevant)
 - `mcp_server/README.md` and `mcp_server/pvl_mcp/tools.py` — current MCP tools
 - `docs/active/VECTOR_SEARCH_SPEC.md` — similarity search architecture
-- `docs/active/TOP_CEO_RECOMMENDATIONS.md` — sustainability + funding context
+- `docs/internal/TOP_CEO_RECOMMENDATIONS.md` — sustainability + funding context
 
 **Recent work:**
 - `git log --oneline -30` — last ~30 commits on `wave-2-ai-platform` branch

docs/active/RESPONSES/peleg_2026_05_18_response.md

@@ -274,7 +274,7 @@ Thank you for the second pass. The PPT was rigorous and caught real issues we ne
 
 - **Live commit referenced**: `4534033` (post-Wave-2 merge)
 - **Verified by**: Explore agent audit + targeted grep spot-check 2026-05-18
-- **Tracking doc**: `docs/active/PELEG_REVIEW_TASKS.md` (to be created with Round 2 audit table)
+- **Tracking doc**: `docs/internal/PELEG_REVIEW_TASKS.md` (to be created with Round 2 audit table)
 - **Wave 2.5 dispatch**: §2 items become individual T2/T3/Cowork tasks
 - **Co-design items**: §3 D1-D5 need scheduling with Peleg + Alex
 - **Open**: confirm with Said which items he wants to add/remove before sending