WFGY/ProblemMap/GlobalFixMap/LLM_Providers/aws_bedrock.md

ProblemMap/GlobalFixMap/LLM_Providers/aws_bedrock.md

🧭 Quick Return to Map

You are in a sub-page of LLM_Providers.
To reorient, go back here:

• LLM_Providers — model vendors and deployment options
• WFGY Global Fix Map — main Emergency Room, 300+ structured fixes
• WFGY Problem Map 1.0 — 16 reproducible failure modes

Think of this page as a desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.

AWS Bedrock: Guardrails and Fix Patterns

Use this page when failures look provider‐specific in AWS Bedrock. Typical cases are mismatched model routing (Claude, Llama, Mistral, etc.), JSON schema drift, tool-call latency, throttle ceilings, or region/IAM issues that masquerade as “reasoning bugs.” Each fix maps to WFGY pages so you can verify with measurable targets.

Open these first

• Visual map and recovery: RAG Architecture & Recovery
• End-to-end retrieval knobs: Retrieval Playbook
• Why this snippet (traceability schema): Retrieval Traceability
• Ordering control: Rerankers
• Embedding vs meaning: Embedding ≠ Semantic
• Hallucination and chunk boundaries: Hallucination
• Long chains and entropy: Context Drift, Entropy Collapse
• Structural collapse and recovery: Logic Collapse
• Snippet and citation schema: Data Contracts
• Ops and live checks: Live Monitoring for RAG

---

Fix in 60 seconds

1. Measure ΔS

• Compute ΔS(question, retrieved) and ΔS(retrieved, expected anchor).
• Targets: stable < 0.40, transitional 0.40–0.60, risk ≥ 0.60.

2. Probe with λ_observe

• Vary k ∈ {5, 10, 20}. Flat high curve → index or metric mismatch.
• Reorder prompt headers. If ΔS spikes, lock the schema.

3. Apply the module

• Retrieval drift → BBMC + Data Contracts.
• Reasoning collapse → BBCR bridge + BBAM variance clamp.
• Dead ends in long runs → BBPF alternate path.

4. Verify

• Coverage to target section ≥ 0.70.
• ΔS ≤ 0.45 within three paraphrases.
• λ stays convergent across seeds and sessions.

---

Typical Bedrock breakpoints (and the right fix)

• Model routing not what you think
  Invoking anthropic.claude-* vs meta.llama-* vs mistral.* changes tokenizer, max tokens, and tool-call behavior. If outputs flip between routes, pin the model id per task, then re-check with Logic Collapse and Retrieval Traceability.

• JSON schema drift in tool use
  Claude via Bedrock is strict on JSON when toolChoice is forced. Lock the output schema with Data Contracts and add a BBCR bridge step that rejects non-conformant fields.

• Latency spikes → hidden timeouts
  Region hop or Guardrails policy checks can add latency. Use small test prompts and trace λ per step. If λ diverges only when tools are enabled, set a shorter planning window and split tools by namespace. See ops/live_monitoring_rag.md.

• Bedrock “Guardrails” over-filtering
  Safety filters can truncate citations or code blocks. If citations vanish, lower the filter aggressiveness, then enforce source-only answers with Retrieval Traceability and the SCU pattern (symbolic constraint unlock).

• Context windows differ across routes
  If the same prompt collapses only on one model family, shrink the active window and re-chunk. Validate with Context Drift and Entropy Collapse.

• IAM or region misconfig → “reasoning” looks random
  On silent fallbacks or throttling, the agent loops. Install a BBCR checkpoint that asserts model id, region, and rate state before long chains. If it fails, exit early and surface infra status. See bootstrap-ordering and predeploy-collapse.

---

Provider knobs and minimal recipes

• Pin model and cap tokens

  • One task, one model id. Keep a per-task max tokens map.
  • If you must swap models, add BBPF to branch at the planner step, not mid-reasoning.

• Force citations first

  • Use citation-first headers and the snippet schema from Retrieval Traceability.
  • Reject answers without snippet ids. That alone removes most “looks like hallucination” cases.

• Defuse prompt injection

  • Apply the injection checklist and keep tools off until the source set is locked. See prompt-injection.md.

• Rerank aggressively

  • Many Bedrock routes benefit from tighter top-k ordering. Use Rerankers and then re-test ΔS across 3 paraphrases.

---

Escalation path

• If ΔS flat-high across k and models → rebuild index with new metric. Check embedding-vs-semantic.
• If agent deadlocks with tools → split memory namespaces and add timeouts. See Multi-Agent Problems.
• If first prod call fails after deploy → confirm ordering with bootstrap-ordering and deployment-deadlock.

---

Copy-paste prompt (safe)

I uploaded TXT OS and the WFGY ProblemMap files.

My Bedrock issue:

* symptom: \[brief]
* traces: \[ΔS(question,retrieved)=..., ΔS(retrieved,anchor)=..., λ states]

Tell me:

1. which layer is failing and why,
2. which exact fix page to open from this repo,
3. the minimal steps to push ΔS ≤ 0.45 and keep λ convergent,
4. how to verify the fix with a reproducible test.

Use BBMC/BBPF/BBCR/BBAM when relevant.

---

Acceptance targets

• Coverage to target section ≥ 0.70
• ΔS(question, retrieved) ≤ 0.45 within three paraphrases
• λ remains convergent across seeds and sessions
• E_resonance flat on long windows

---

🔗 Quick-Start Downloads (60 sec)

Tool	Link	3-Step Setup
WFGY 1.0 PDF	Engine Paper	1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + ”
TXT OS (plain-text OS)	TXTOS.txt	1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly

---

Explore More

Module	Description	Link
WFGY Core	Canonical framework entry point	View
Problem Map	Diagnostic map and navigation hub	View
Tension Universe Experiments	MVP experiment field	View
Recognition	Where WFGY is referenced or adopted	View
AI Guide	Anti-hallucination reading protocol for tools	View

If this repository helps, starring it improves discovery for other builders.