Rat auto design policy

RTL Auto-Design Policy

Core Principles

Hierarchical Spec Compliance

Lower phases MUST NOT violate upper phase specifications: Spec → Architecture → μArch → RTL → Verification Each phase strictly adheres to decisions made in all preceding phases. Deletion, reduction, or modification of features for convenience is FORBIDDEN. If a change is needed, control returns to the upper phase for approval.

Design Priority Order

1. Functional Correctness (highest) — Every required feature works exactly
2. Interface Compliance — Ports, protocols, timing match Architecture
3. Timing/Performance — Throughput, latency targets met
4. Area/Power (lowest)

Cascading Quality

Higher abstraction levels demand MORE iterative refinement because their quality cascades to ALL downstream phases. A defect at the architecture level costs orders of magnitude more to fix at RTL than if caught during architecture review.

Graduated iteration by abstraction level: Phase 1 (Research): 3 mandatory rounds (chief-coordinated) Phase 2 (Architecture): dynamic convergence review (min 2, max 5 rounds per p2-arch-design-policy) — memory, performance, ref model consistency Phase 3 (μArch): dynamic convergence review (min 2, max 5 rounds per rtl-p3-uarch-policy) — performance, interface, memory optimization Phase 4 (RTL): Wave-based lint+sim (implementation-level) Phase 5 (Verify): Sub-phase parallel (terminal verification)

Iteration count can be increased beyond 3 if convergence is not achieved. Principle: refine thoroughly at the top, execute efficiently at the bottom.

Document-as-Memory

Design artifacts (docs/, reviews/) serve as persistent memory across phases and agents. Each phase reads upstream documents as input context and writes downstream documents as output. No agent needs to "remember" another agent's output — it reads the document.

Document flow: iron-requirements.json + open-requirements.json → arch-designer reads → architecture.md → uarch-designer reads → docs/phase-3-uarch/*.md → rtl-coder reads reviews/phase-N/ → Quality Gate reads → next phase proceeds or fails

State is persisted at .rat/state/rat-auto-design-state.json for resumability.

Execution Rules

Dual-Layer Phase Gates

Every phase transition requires BOTH:

1. Artifact Gate: Required files/directories exist (fast check)
2. Quality Gate: Reviewer agent(s) verify quality AND hierarchical spec compliance

Quality Gate verdicts: PASS or FAIL + findings[]

Gate Retry Policy

• On Artifact Gate failure: retry the failed phase once, then escalate to user
• On Quality Gate failure: pass findings back to the phase's worker agent for correction, then re-run Quality Gate. Maximum 2 retry cycles per gate
• On upper-spec violation: IMMEDIATE STOP (see Escalation)

Escalation Ladder (Per-Gate 2x)

Every active gate uses a per-gate retry budget N (retry_limit in state):

• Primary range: attempts 1..N with normal strategy
• Fallback range: attempts N+1..2N with immediate strategy switch
  • split by module/requirement
  • swap reviewer+solver agent pairing
  • re-run only impacted checks
• Last chance: one automatic alternative attempt after 2N
• Post 2x+1 fail: set needs_user_decision=true and stop for user direction

State contract for hook enforcement lives in:

• orchestration_control.active_gate_* (fast path for hooks)
• orchestration_control.gates.{gate_id} (per-gate detailed counters)

Dynamic Prompt Injection

When entering fallback or last-chance stages, orchestrator writes:

• orchestration_control.dynamic_prompt_text (plain text, single-shot guidance)
• orchestration_control.dynamic_prompt metadata (source, strategy_tag, used)

Fallback templates are available in:

• ${CLAUDE_PLUGIN_ROOT}/skills/rat-auto-design/templates/escalation-prompts.json (plugin runtime)
• skills/rat-auto-design/templates/escalation-prompts.json (development repo context)

Use templates only when LLM-generated prompt text is unavailable. If both paths are unreadable, orchestrator MUST use the built-in defaults below and still write the chosen text into orchestration_control.dynamic_prompt_text with orchestration_control.dynamic_prompt.source = "builtin":

• primary: Continue current gate workflow, focus on pending criteria with existing agent assignment.
• fallback: Split failing scope by module/requirement, switch reviewer+solver pairing, rerun impacted checks only.
• last_chance: Apply one non-overlapping alternative strategy, record deltas, prepare escalation context.
• user_escalation: Retries exhausted; ask user with failure summary, attempted strategies, and recommended options.

Context Preload

Before each phase, verify required upstream files exist:

• required (full read): files that MUST be fully read before starting the phase
• summary only: files where only the phase summary is sufficient
• optional (on demand): files read only when a specific question arises

Agents entering a phase MUST load required files first, then summaries. Specific file lists are defined inline in each orchestrator's phase steps.

Scratchpad Convention

During iterative review rounds, reviewers write findings to: .rat/scratch/phase-{N}/round-{R}-{agent}.md

Coordinator reads all round files to aggregate: .rat/scratch/phase-{N}/round-{R}-feedback.md

On phase gate PASS: consolidated review saved to reviews/phase-{N}-*/, scratch cleaned. On phase gate FAIL + retry: scratch files preserved for next round.

Phase Gate Definitions

Every transition = Artifact Gate + Quality Gate (see Dual-Layer Phase Gates). Quality-gate semantics — reviewer rosters, review rounds, rebuttal protocol, convergence, and detailed pass/fail criteria — are OWNED by the per-phase policies in the table below. This policy defines only the artifact/verdict contract the orchestrator checks.

Gate	Artifact Gate (must exist)	Verdict File(s)	Verdict	Quality Gate
1→2	docs/phase-1-research/: iron-requirements.json + io_definition.json + timing_constraints.json + domain-analysis.md (open-requirements.json optional — absent if P1 had no open items)	reviews/phase-1-research/research-review.md + per-round research-review-r1..r3.md (mandatory)	verdict=PASS (3-round chief review converged, or gaps escalated and user-approved; all requirements clear, consistent, implementable)	quality gates per p1-spec-research-policy
2→3	docs/phase-2-architecture/architecture.md (with D2 block diagram) + iron-requirements.json (P2, REQ-A-*) + refc/*/*.c	reviews/phase-2-architecture/architecture-review.md + feature-coverage.md + per-round architecture-review-r{N}.md (min 2)	verdict=PASS AND 100% REQ coverage AND review converged	quality gates per p2-arch-design-policy
3→4	docs/phase-3-uarch/*.md + iron-requirements.json (P3, REQ-U-*) + bfm/ directory	reviews/phase-3-uarch/uarch-review.md + feature-preservation.md + per-round uarch-review-r{N}.md (min 2)	verdict=PASS AND 100% feature preservation AND review converged	quality gates per rtl-p3-uarch-policy
4→5	rtl/*/*.sv (all lint-clean) + sim/*/tb_*.sv + sim/*/*_results.txt (all PASS) + basic integration smoke PASS + Stream B artifacts (see Stream B Rules)	reviews/phase-4-rtl/functional-completeness.md + design-review.md + lint-report.md	verdict=PASS on all three (100% functional coverage, lint 0 errors)	quality gates per rtl-p4-implement-policy
5 completion	all verification sub-phases (5a-5e) PASS	reviews/phase-5-verify/requirement-traceability.md + e2e-traceability.md + final-compliance.md	verdict=PASS (every requirement implemented, verified, passing)	quality gates per rtl-p5-verify-policy
5→6	— (artifact-only gate)	reviews/phase-5-verify/final-compliance.md	verdict=PASS	—
6 completion	reviews/phase-6-review/design-note.md (complete document) + improvements.md (recommendations produced)	reviews/phase-6-review/code-review.md + design-review.md	verdict=PASS on both; on FAIL iterate review → fix cycle (max 2 rounds)	quality gates per rtl-p6-design-review-policy

Phase 1 gate additions (owned by this policy): compliance-checker verifies iron requirements have no internal contradictions; arch-designer confirms implementation feasibility (every requirement realizable in RTL within reasonable area/timing).

Summary Validation (every P1-P5 gate): docs/phase-1-research/phase-1-summary.md, docs/phase-2-architecture/phase-2-summary.md, docs/phase-3-uarch/phase-3-summary.md, docs/phase-4-rtl/phase-4-summary.md, docs/phase-5-verify/phase-5-summary.md must exist (format: max 1 page with tables for Key Decisions, Module Inventory, Interface Summary, Quality Gate Results, Open Items, Document References)

ADR Recording (Phase 2→3 and 3→4 gates): 3-5 key decisions → docs/decisions/ADR-{NNN}.md

Phase 2 Iterative Review: reviewer roster, round semantics, rebuttal protocol, and convergence criteria are owned by p2-arch-design-policy (dynamic convergence: min 2, max 5 rounds; domain experts conditional on domain-package presence).

Phase 3 Iterative Review: reviewer roster (incl. BFM reviewer), round semantics, rebuttal protocol, and convergence criteria are owned by rtl-p3-uarch-policy (dynamic convergence: min 2, max 5 rounds; domain experts conditional on domain-package presence).

Stream B Rules (Phase 4)

Stream B artifacts are generated concurrently with Stream A (RTL implementation):

• B1. SVA property skeletons from μArch specs → docs/phase-4-rtl/stream-b-sva-skeletons.md
• B2. Preliminary CDC topology from μArch specs → docs/phase-4-rtl/stream-b-cdc-preliminary.md
• B3. cocotb TB skeletons from μArch specs → docs/phase-4-rtl/stream-b-tb-skeletons.md

Traceability Convention:

• SVA skeletons: each property references μArch source: // Source: docs/phase-3-uarch/{module}.md, Section: {section}
• CDC preliminary: each CDC path references architecture clock domain definition
• TB skeletons: each test scenario references requirement: # REQ-{NNN}: {description}

Content Quality Gate (prevents empty-skeleton artifacts from passing):

• SVA skeletons: must contain at least one property or assert keyword per referenced module
• CDC preliminary: must reference actual clock domain names from docs/phase-3-uarch/clock-domain-map.md
• TB skeletons: must reference at least one REQ- tag per module and contain at least one test function/task

Merge Point: Stream A (lint-clean + unit PASS) + Stream B artifacts ready (content quality verified) → Phase 4→5 Gate.

Feedback Loop Classification (Phase 5→4)

Type	Scope	Handling	Re-verify
UNIT_FIX	Single module (SVA fail, assertion error)	rtl-p4s-bugfix (parallel across modules)	Only failed sub-phases
INTEGRATION_FIX	Cross-module interface	rtl-p4s-bugfix (sequential)	5b + 5c
DESIGN_FIX	Architecture-level	IMMEDIATE STOP → user approval	All (5a-5e) after upper phase fix

Batch UNIT_FIX across sub-phases:

• Group by module. Different modules → parallel fix. Same module → sequential.
• Each rtl-p4s-bugfix follows: analyze → fix → lint → TB → sim
• After ALL fixes: re-run ONLY affected sub-phases in parallel

Max 2 feedback loops per sub-phase. After loop 2 fails → escalate to user.

Sub-phase Re-entry Criteria

Fix Type	Re-run Sub-phases	Condition
UNIT_FIX (SVA fail)	5a only (formal)	SVA property affected
UNIT_FIX (sim fail)	5c only (integration)	Testbench affected
INTEGRATION_FIX	5b + 5c (CDC + integration)	Interface modified
DESIGN_FIX	All (5a-5e) after upper phase approval	Architecture changed

Feedback Loop State Schema

{
  "loop_count": 1,
  "max_loops": 2,
  "failures": [{
    "sub_phase": "5a",
    "type": "UNIT_FIX",
    "module": "example_module",
    "description": "SVA counterexample at cycle 42",
    "fix_applied": "Added pipeline register",
    "re_run_phases": ["5a"]
  }],
  "status": "in_progress"
}

Lesson Learned Recording

After each successful feedback fix:

• Append entry to docs/lessons-learned.md with format: LL-{NNN} with sections: Symptom, Root Cause, Fix Applied, Prevention, Related (REQ IDs, module, fix commit, ADR, Phase 5 Sub-phase)
• Record: symptom, root cause, fix applied, prevention strategy, related REQ/module/ADR

Escalation & Stop Conditions

• Artifact Gate fails twice → pause and report missing artifacts to user
• Quality Gate fails after 2 fix-and-retry cycles → pause, present all accumulated findings, request guidance
• Gate ladder exhausted (per Escalation Ladder: N → 2N → last-chance) → set needs_user_decision=true, stop and ask user before any further retry
• Upper-Spec Violation detected at any Quality Gate → IMMEDIATE STOP:
  1. Identify the violated upper phase and the specific violation
  2. Report to user with full context
  3. DO NOT proceed — wait for user to approve rollback or waiver
  4. If approved, return to the appropriate upper phase
• Phase 5→4 Feedback Loop exhausted (2 cycles per sub-phase) → escalate to user
• Phase 5 DESIGN_FIX detected → IMMEDIATE STOP, report upper-spec violation
• Verification phase fails after 2 retries → invoke rtl-bug-repro skill
• User says "cancel"/"stop" → delete state file, report progress summary

Coding Convention Summary

All phases must enforce:

• Port prefix: i_, o_, io_ (NOT suffix). Clock/reset exempt
• Clock: clk (single) or {domain}_clk (multiple). Reset: rst_n or {domain}_rst_n
• No CamelCase: snake_case or ALL_CAPS only
• Parameters ALL_CAPS, localparam L_ prefix
• logic only (no reg/wire), always_ff/always_comb, ANSI port style
• Instance prefix: u_, generate prefix: gen_
• SV RTL: IEEE 1800-2009. SV Verification: IEEE 1800-2012. C ref model: C11

Final Checklist

• State file written before starting
• Each phase passed BOTH Artifact Gate AND Quality Gate
• Hierarchical Spec Compliance verified at every gate (verdict contract per the Artifact-Gate Table above; detailed criteria per each gate's owning policy)
• No upper-spec violations left unresolved
• Naming conventions enforced at every phase gate
• All 6 phases completed
• State file removed on clean completion
• Summary report generated
• Review artifacts verified per {plugin_root}/skills/rat-auto-design/references/review-checklist.md (26 mandatory files)