Loading ToolBench
AI-assisted ELS (Equidistant Letter Spacing) Bible code research with semantic-code correlation. Searches for hidden letter patterns encoded at skip intervals in the KJV Bible, maps hits to verse references, analyzes proximity clustering, and runs statistical significance tests.
KARP Bible Code demonstrates strong definition quality across 6 tools with comprehensive schemas, descriptions, and clear naming conventions. All tools follow verb_noun patterns (els_search, els_session, els_proximity, els_cluster, els_stats, els_sweep). Descriptions are detailed and domain-contextual (150-400+ chars), exceeding baseline. Input schemas are fully typed JSON Schema with clear parameter constraints and enums. However, output schemas are not explicitly documented in the source provided, and error handling guidance is absent. Tool composition is excellent, each tool has a single clear responsibility, though some are complex. The system lacks explicit recovery guidance for failure modes.
Find the densest spatial clusters of multiple ELS hits within a session. Scans the entire stream for the region with the most unique terms packed closest together. USE CASES: - Family study: Search all family member names, then cluster to find where they all appear close together — often the most interesting theological region - Messianic study: Search JESUS, MESSIAH, CHRIST, YESHUA, etc., then find the cluster - Topic analysis: Search multiple related terms and find where they naturally congregate RESULT: - Best cluster region (start and end position) - Count of hits - Unique terms present - For each term, its closest hit in the cluster (position, skip, verse reference) OPTIONAL: Can also export the cluster region as a sub-stream for further analysis.
Analyze spatial clustering between two terms' ELS hits. After searching multiple terms, use this to find which skip intervals place them close together — often the most theologically interesting discoveries come from term pairs, not isolated words. GIVEN: Two sets of hits (from els_search caches) RETURN: Ranked list of (skipA, skipB) pairs sorted by minimum distance between hits DISTANCE METRICS: - 0 letters = intersection (letters overlap) - 1-100 = extremely close - 100-1000 = nearby (same chapter region) - 1000-10000 = mid-distance (same book region) - 10000+ = far (different sections) USE CASES: - Search JESUS and MESSIAH, then check proximity — intersecting skip pairs are rare and significant - Search a name and a date (e.g. ISAAC + 1996), see if they cluster - Family study: search JOSHUA, ZACHARY, ISABELLA, then find which skips place them near each other OPTIONAL SESSION-LEVEL ANALYSIS: If both terms are in the same session, can also cluster across ALL searches in that session.
Search for a term encoded at equidistant letter spacing (ELS) in the Bible's continuous letter stream. This is the core research tool — it finds hidden words spelled out at regular skip intervals across the text. HOW IT WORKS: The Bible text is stripped to uppercase A-Z letters to form a continuous stream. The engine checks every possible starting position and skip interval to find where the letters of your term appear at equal spacing. For example, "JESUS" at skip 186 means the letters J-E-S-U-S appear every 186th letter in the stream. EVERY SEARCH IS AUTO-SAVED with all hit positions and verse mappings. This builds your research ledger over time — searches from today can be cross-referenced against searches from weeks ago via els_sweep. CHOOSING A STREAM: - "genesis" (151K letters) — fastest, great for initial exploration - "torah" (634K letters) — traditional Bible code research scope - "full" (3.2M letters) — most comprehensive but slower - Any book abbreviation (e.g. "rev", "isa") — built on demand SKIP RANGE: Default 1-3000. For Torah, 1-5000 is common. Wider ranges find more hits but take longer. For quick checks, 1-1000 is fast. RESULTS include: hit count, skip intervals, and verse references for each hit's start, midpoint, and end positions. After searching, suggest checking proximity with other terms or running stats. TIPS: - 4+ letter terms give meaningful results. 3-letter terms produce noise. - Use session_id to group related searches (e.g. family name study). - Direction "both" searches forward AND reverse — doubles the search space. - After multiple searches, use els_proximity or els_cluster on cached data.
Output schemas not documented in tool definitions. LLMs cannot predict response structure, requiring inference from examples or trial calls. Pattern:response-shaper requires explicit output documentation.
No error handling guidance. Tools document happy paths but provide no recovery hints for failure modes (e.g., 'term not found in cache', 'session does not exist', 'stream too large'). Pattern:recovery-guide requires actionable error responses.
Parameter 'stream' in els_search accepts arbitrary book abbreviations (any book abbreviation → built on demand) but no enum constraint or validation list provided. LLMs will hallucinate invalid abbreviations. Should enumerate valid book codes or document the full list.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | B | 71 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Create, view, or list ELS research sessions. Sessions group related searches together — for example, "Sharman Family" groups all family name searches, "Messianic Study" groups searches for messianic terms. ACTIONS: - create: Start a new session with a name and optional notes - view: Get a session's details and all its searches - list: Browse all sessions with search counts - update: Change a session's name or add notes - delete: Remove a session and all its searches/hits (destructive!) When starting a new research topic, create a named session first, then pass its session_id to each els_search call. This keeps research organized and enables session-level cluster analysis later.
Run statistical significance tests on ELS search results. Determine whether a term's hit count is likely by chance or statistically anomalous. TESTS: - Poisson model: Treats the stream as random and calculates the probability of observing N+ hits - Monte Carlo: Repeatedly shuffles the stream and counts hits, building an empirical null distribution - Letter frequency normalization: Adjusts for terms that are either easy or hard to spell in English (high-frequency vowels vs rare letters) OUTPUT: - P-value (probability this happened by random chance) - Z-score (how many standard deviations above expected) - "Significance" label (not significant, interesting, very significant, extremely significant) - Breakdown by skip range (which skip intervals contributed most hits) INTERPRETATION: - p > 0.05 = likely random - 0.01 < p < 0.05 = interesting, worth noting - p < 0.01 = very significant - p < 0.001 = extremely significant (rare in real data) IMPORTANT: This test assumes the null hypothesis is random letter occurrence. Real Bible text is NOT random (it has language structure, vocabulary, grammar). So even "not significant" p-values don't mean the pattern is uninteresting — they just mean it's not anomalous for a meaningful text. Use stats as ONE input, not the only criterion.
Cross-reference multiple searches from your research history. After weeks of studying, this tool lets you ask questions like: - "Which terms from my Messianic study have hits that cluster with JESUS?" - "Show me all proximity pairs in my Genesis analysis where the distance is < 1000 letters" - "Find any new patterns between searches from last week and searches from today" Essentially: given a session (or date range, or tag), find all pairwise proximities and clusters, ranked by tightness. OUTPUT: Ranked list of term pairs and their proximity patterns, optionally filtered by date, session, or distance threshold.
els_session 'delete' action is destructive but no confirmation-request pattern documented. Agents could accidentally delete entire research sessions without user confirmation.
Parameter 'direction' in els_search has valid enum (forward|reverse|both) but description says 'both' searches 'forward AND reverse, doubles the search space', implying additive behavior that may not be clear to LLMs. Clarify what 'both' means operationally.