WFGY/ProblemMap/GlobalFixMap/LanguageLocale/rtl_bidi_control.md

RTL & BiDi Control — Guardrails and Fix Pattern

Stabilize retrieval and reasoning when left-to-right content mixes with right-to-left scripts or invisible BiDi marks. No infra change required. All fixes map back to WFGY pages with measurable targets.

What this page is

• A compact repair guide for directionality bugs that flip tokens, citations, or numbers.
• Steps to normalize control characters, lock direction metadata, and keep offsets verifiable.
• Store-agnostic checks you can run in minutes.

When to use

• Citations look correct to the eye but snippet offsets do not match.
• Punctuation or brackets render on the wrong side in answers.
• Arabic or Hebrew lines invert number order or collapse after parsing.
• JSON fields with mixed direction break validation or flip keys.
• Search returns near hits but ΔS stays high on RTL content.

Open these first

• Visual map and recovery: RAG Architecture & Recovery
• End to end retrieval knobs: Retrieval Playbook
• Why this snippet (traceability schema): Retrieval Traceability
• Snippet and citation schema: Data Contracts
• Wrong-meaning hits despite high similarity: Embedding ≠ Semantic
• Hybrid instability and reorder issues: Rerankers
• Digits width and punctuation mix: Digits • Width • Punctuation

Acceptance targets

• ΔS(question, retrieved) ≤ 0.45 on three paraphrases.
• Coverage of target section ≥ 0.70.
• λ remains convergent across two seeds.
• Offsets verified after normalization on both query and snippet.

---

Typical breakpoints → exact fix

• Invisible BiDi marks inside snippets cause reversed punctuation or bracket order.
  Fix: strip control code points during indexing and query pre-norm. Persist a dir flag on the clean text.
  Open: Data Contracts, Retrieval Traceability

• Rendered order vs stored order mismatch makes citations fail.
  Fix: compute character offsets on the normalized text only. Log the normalization pipeline in trace.
  Open: Retrieval Traceability

• Numbers flip in Arabic or Hebrew lines when Eastern Arabic digits mix with Latin punctuation.
  Fix: normalize digits to a single system for retrieval. Keep the original form for display.
  Open: Digits • Width • Punctuation

• JSON payloads break or tool calls mis-route because keys include RTL marks.
  Fix: forbid control chars in keys through schema, allow in values only after normalization.
  Open: Prompt Injection, Data Contracts

---

60-second fix checklist

1. Strip BiDi controls during ingest and query Remove these if present:
   LRM U+200E, RLM U+200F, LRE U+202A, RLE U+202B, LRO U+202D, RLO U+202E, PDF U+202C,
   LRI U+2066, RLI U+2067, FSI U+2068, PDI U+2069.
   Also normalize NBSP U+00A0, ZWJ U+200D when it changes tokenization.

2. Persist direction metadata Add dir = "rtl" | "ltr" | "auto" at snippet and paragraph levels. Store it in the trace envelope.

3. Index on normalized text only

   • Normalize to NFC.
   • Strip BiDi marks.
   • Fold digits per store policy.
   • Keep original text for rendering.

4. Contract the payload Require fields: snippet_id, dir, norm_hash, offsets_on_norm, source_url.
   Reject if dir missing on RTL sources.
   Open: Data Contracts

5. Probe λ_observe Vary k = 5, 10, 20. If ΔS stays flat and high, rebuild the index after normalization and re-verify offsets.

---

Copy-paste prompt

You have TXT OS and the WFGY Problem Map loaded.

My multilingual issue:

* symptoms: punctuation flips or offsets fail on RTL lines
* traces: ΔS(question,retrieved)=..., λ across 3 paraphrases, direction flags

Tell me:

1. the failing layer and why,
2. the exact WFGY page to open,
3. the minimal steps to push ΔS ≤ 0.45 and keep λ convergent,
4. a reproducible check that verifies offsets after normalization.

---

🔗 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	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 —
Engineers, hackers, and open source builders who supported WFGY from day one.

⭐ WFGY Engine 2.0 is already unlocked. ⭐ Star the repo to help others discover it and unlock more on the Unlock Board.