This document constitutes the official scientific and technical specification of the Praman (प्रमाण) Indic search intelligence engine. It details the exact mathematical formulas, boundary constraints, and data contracts that govern demand calculation across all 10 supported Indic languages and English.
1. Axioms of Measurement
- Axiom 1 (No Telemetric Hallucination): No metric shall be generated via regression models trained on disparate or geographically unrepresentative clickstream panels. Every integer or score emitted by Praman corresponds directly to observable search engine behavior.
- Axiom 2 (Script Invariance): Language-specific tokenization algorithms must treat Brahmic dependent vowel signs (matras), non-spacing combining marks, and conjunct viramas with identical syntactic integrity as root consonants.
-
Axiom 3 (Strict 3-State Codomain): For any measurement probe $p$, the observation space $V(p)$ satisfies:
V(p) ⊆ ℝ+ ∪ { 0 } ∪ { ⊥ }Under no circumstances shall an unobserved event ($ot$) be coerced to the value 0.
2. The Four Deterministic Dimensions
The composite Praman Search Demand Score $D(s) \in [0, 100] \cup \{ot\}$ for a given seed keyword $s$ is computed via four strictly isolated dimensional probes:
Dimension 1: Expansion Breadth ($S_{\text{breadth}}$) • Weight = 0.50
Let $\mathcal{C}$ be the complete phonemic consonant set of the language’s native script ($|\mathcal{C}| = 34$ for Devanagari, $|\mathcal{C}| = 18$ for Tamil). We probe the search engine with the seed concatenated with each consonant $c \in \mathcal{C}$.
A high breadth score indicates that the keyword naturally branches into multiple sub-topics across everyday vernacular conversation.
Dimension 2: Head Coverage ($S_{\text{head}}$) • Weight = 0.25
We probe the engine with the unprompted root seed $s$ alone. Let $A(s) = (a_1, a_2, \dots, a_M)$ be the ordered suggestion array returned (where $M \le 10$).
Measures whether the core phrase is an established authority head term in Google’s autocomplete index.
Dimension 3: Question Density ($S_{\text{question}}$) • Weight = 0.15
Let $\mathcal{Q}$ be the set of standardized interrogative tokens in the target language (e.g., in Marathi: काय, कसे, कुठे, कधी, किती, कोण; in Hindi: क्या, कैसे, कहाँ, कब, कितना, क्यों).
High question density directly correlates with high informational search traffic, Google Featured Snippets, and voice search adoption.
Dimension 4: Rank Depth ($S_{\text{rank}}$) • Weight = 0.10
Calculates the reciprocal rank prominence of candidate matches in autocomplete result sets. Rank 1 receives maximum weight, degrading linearly to Rank 10.
3. The Pinned Voice Contract: Anti-Drift Guarantee
The Contract
The weights $(0.50, 0.25, 0.15, 0.10)$ and their underlying sample denominators are permanently pinned in the Praman source code. If any probe encounters an API failure, upstream timeout, or network partitioning, the system strictly refuses to inflate or redistribute weights to surviving endpoints. The missing dimension is rendered explicitly as $ot$, protecting editorial publishers from false confidence.
4. Supported Script Specifications
Praman natively supports 10 Indic scripts with zero combining mark corruption:
- Devanagari (मराठी, हिन्दी): Unicode range
U+0900 - U+097F - Tamil (தமிழ்): Unicode range
U+0B80 - U+0BFF - Telugu (తెలుగు): Unicode range
U+0C00 - U+0C7F - Kannada (ಕನ್ನಡ): Unicode range
U+0C80 - U+0CFF - Malayalam (മലയാളം): Unicode range
U+0D00 - U+0D7F - Bengali (বাংলা): Unicode range
U+0980 - U+09FF - Gujarati (ગુજરાતી): Unicode range
U+0A80 - U+0AFF - Gurmukhi (ਪੰਜਾਬੀ): Unicode range
U+0A00 - U+0A7F - Odia (ଓଡ଼ିଆ): Unicode range
U+0B00 - U+0B7F - Latin (English): ASCII
0x20 - 0x7Ewith full stemming
Test the Mathematical Formula on Live Data
Run any seed keyword through the 4-dimensional engine right now on Praman.blog.
⚡ Launch Live Praman Planner