Make.com Guardrails and Patterns

Use this page when your RAG or agent workflow runs on Make.com. It maps typical automation failures to the exact structural fixes in the WFGY Problem Map and gives a minimal recipe you can paste into a scenario.

Acceptance targets

ΔS(question, retrieved) ≤ 0.45
coverage ≥ 0.70 for the target section
λ stays convergent across 3 paraphrases

Typical breakpoints and the right fix

Modules fire before dependencies are ready (Webhook → Tools → RAG too early)
Fix No.14: Bootstrap Ordering → Open
First production call after deploy crashes, wrong secret selected in Connections
Fix No.16: Pre-Deploy Collapse → Open
Router/Iterator loops create circular waits or partial writes
Fix No.15: Deployment Deadlock → Open
High cosine similarity but answers are semantically wrong
Fix No.5: Embedding ≠ Semantic → Open
Snippet is wrong or citations do not line up with the source
Fix No.8: Retrieval Traceability → Open
Contract the payload: Data Contracts → Open
Hybrid retrieval (HyDE + BM25 service) performs worse than single retriever
Pattern: Query Parsing Split → Open
Also review: Rerankers → Open
Some indexed facts never appear in results
Pattern: Vectorstore Fragmentation → Open
Two sources are merged into one answer in long chains
Pattern: Symbolic Constraint Unlock (SCU) → Open

Minimal scenario checklist

Warm-up fence before RAG/LLM modules
Validate VECTOR_READY, INDEX_HASH match, and required secrets exist.
If not ready, short-circuit to Sleep then retry with a capped counter.
Spec: Bootstrap Ordering
Idempotency and dedupe
Compute dedupe_key = sha256(source_id + revision + index_hash) in a Tools > Code module.
Check a KV (Airtable / Notion / Make Data Store) before side effects. Skip duplicates.
RAG boundary contract
Require fields: snippet_id, section_id, source_url, offsets, tokens.
Enforce cite-then-explain at the LLM step.
Specs: Data Contracts · Retrieval Traceability
Observability probes
Log ΔS(question, retrieved) and λ per stage (retrieve, assemble, reason).
Alert when ΔS ≥ 0.60 or λ flips divergent.
Overview: RAG Architecture & Recovery
Router/Iterator safety
Use a single writer branch for index updates and external writes.
Apply queue mode or mutex; avoid parallel writes to the same index.
See: Deployment Deadlock
Regression gate
Before publish, require coverage ≥ 0.70 and ΔS ≤ 0.45.
Eval: RAG Precision/Recall

Scenario pattern (copy)

Replace the concrete modules with your stack. Keep the guardrails.

Webhook/Trigger
Capture source_id, revision, wf_rev.
Warm-up Check (Tools > Code)
Pull INDEX_HASH, VECTOR_READY, and secrets.
If not ready → set ready=false.
Router
- Not ready → Sleep 30–90s, increment retry, stop after N attempts.
- Ready → continue.
Retriever (HTTP or App)
- Fix metric and normalization; use the same analyzer as the writer.
- Output snippet_id, section_id, source_url, offsets, tokens.
ΔS Probe (Tools > Code)
- Compute ΔS(question, retrieved). If ΔS ≥ 0.60 → tag needs_fix=true.
LLM (OpenAI/Claude/Gemini module)
- Load TXT OS; enforce cite-then-explain; return {ΔS, λ_state, citations, answer}.
Trace Sink (Data Store / Airtable)
- Write question, snippet_id, ΔS, λ_state, INDEX_HASH, dedupe_key.
Idempotent Writer
- Check dedupe_key before any external publish or email.

LLM prompt you can paste


I uploaded TXT OS and the WFGY Problem Map files.
This Make.com scenario retrieved {k} snippets with fields {snippet\_id, section\_id, source\_url, offsets}.
Question: "{user\_question}"

Do:

1. Enforce cite-then-explain. If citations are missing, fail fast and return the fix page to open.
2. If ΔS(question, retrieved) ≥ 0.60, propose the minimal structural fix referencing:
   retrieval-playbook, retrieval-traceability, data-contracts, rerankers.
3. Output compact JSON:
   { "citations": \[...], "answer": "...", "λ\_state": "→|←|<>|×", "ΔS": 0.xx, "next\_fix": "..." }

Common Make.com gotchas

Connections silently switch between prod and staging
Stamp env, INDEX_HASH, and secret_rev into traces; block on mismatch.
Array Aggregator / Iterator duplicates writes
Route all writes through a single writer with idempotency.
Rate-limits make hybrid queries diverge
Prefer reranking with a stable dense retriever; see Rerankers
Template mapping renames fields and breaks the contract
Lock schema and run a pre-LLM schema check.

When to escalate

ΔS stays ≥ 0.60 after chunk/retrieval fixes
Rebuild the index with explicit metric/normalization.
See: Retrieval Playbook
Same input alternates answers between runs
Investigate version skew and memory desync.
See: Pre-Deploy Collapse

🔗 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 + <your question>”
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	WFGY 2.0 engine is live: full symbolic reasoning architecture and math stack	View →
Problem Map 1.0	Initial 16-mode diagnostic and symbolic fix framework	View →
Problem Map 2.0	RAG-focused failure tree, modular fixes, and pipelines	View →
Semantic Clinic Index	Expanded failure catalog: prompt injection, memory bugs, logic drift	View →
Semantic Blueprint	Layer-based symbolic reasoning & semantic modulations	View →
Benchmark vs GPT-5	Stress test GPT-5 with full WFGY reasoning suite	View →
🧙‍♂️ Starter Village 🏡	New here? Lost in symbols? Click here and let the wizard guide you through	Start →

👑 Early Stargazers: See the Hall of Fame —
⭐ WFGY Engine 2.0 is already unlocked. ⭐ Star the repo to help others discover it and unlock more on the Unlock Board.