layer int64 | feature_id int64 | rank int64 | title string | explanation string | status string | activation_count int64 | frequency float64 | maximum float64 | mean_active float64 | evidence_counts dict | sampling_exceptions list | methods dict |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
0 | 0 | 512 | Noun-Subject Gate Activation | The gate activates on nouns representing specific locations, physical objects, or abstract concepts (like language, habitats, or interfaces) that serve as the primary subject or object in the sentence. It fires strongly on proper nouns (Lagos, Laguna) and common nouns (console, species, language) when they are the focu... | complete | 17,453 | 0.001745 | 2.03125 | 1.072201 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 39,
"tn": 14,
"fp": 36,
"fn": 11,
"correct": null,
"accuracy": 0.53,
"f1": 0.624,
"balanced_accuracy": 0.53,
"a... |
0 | 1 | 512 | Suffix and Preposition Completion | The gate activates on the final token of a word, specifically when that token is a suffix (e.g., -ion, -er, -ly, -ing) or a preposition (e.g., of, for) completing a phrase. | complete | 5,803 | 0.00058 | 1.789063 | 0.874947 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 0,
"tn": 50,
"fp": 0,
"fn": 50,
"correct": null,
"accuracy": 0.5,
"f1": 0,
"balanced_accuracy": 0.5,
"accuracy_... |
0 | 2 | 512 | Final Token of Specific Nouns | The gate activates on the final token of a word, particularly when the word is a specific noun, proper name, or technical term (e.g., "older", "aged", "omics", "drafts", "gangs"). | complete | 6,032 | 0.000603 | 1.726563 | 0.969558 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 10,
"tn": 50,
"fp": 0,
"fn": 40,
"correct": null,
"accuracy": 0.6,
"f1": 0.3333333333333333,
"balanced_accuracy": 0.6... |
0 | 3 | 512 | Syntactic Code Identifier Gate | The gate activates on specific tokens within code snippets, particularly method calls, property accesses, and library imports. It fires on identifiers like `formik`, `pygame`, `addEventListener`, and `catchError`, as well as structural elements like `@NgModule` and `DOCTYPE`. The pattern is consistent across various pr... | complete | 6,473 | 0.000647 | 3.703125 | 1.227813 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 37,
"tn": 50,
"fp": 0,
"fn": 13,
"correct": null,
"accuracy": 0.87,
"f1": 0.8505747126436781,
"balanced_accuracy": 0.... |
0 | 4 | 512 | Spatial and Logical Reference Gate | The gate activates on tokens indicating spatial or logical reference to preceding text, specifically the word "above" or "Lik" (likely "Likely" or "Like") following a passage. | complete | 8,789 | 0.000879 | 2.84375 | 1.099148 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 26,
"tn": 50,
"fp": 0,
"fn": 24,
"correct": null,
"accuracy": 0.76,
"f1": 0.6842105263157895,
"balanced_accuracy": 0.... |
0 | 5 | 512 | Capitalized Word Structural Header Trigger | The gate activates on capitalized words, often serving as structural headers, labels, or proper nouns, and frequently appears at the beginning of a line or sentence. | complete | 4,051 | 0.000405 | 2.125 | 0.864217 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 36,
"tn": 21,
"fp": 29,
"fn": 14,
"correct": null,
"accuracy": 0.57,
"f1": 0.6260869565217392,
"balanced_accuracy": 0... |
0 | 6 | 512 | State-of-Being Verbs and Adjectives | The gate activates on verbs and adjectives describing a state of being, existence, or condition, such as "belong," "succeed," "exist," "appear," "emit," "occupied," "terminate," "disappear," "fled," "compelled," "existing," "deserve," and "prevailing." | complete | 10,894 | 0.001089 | 1.6875 | 0.987031 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 9,
"tn": 49,
"fp": 1,
"fn": 41,
"correct": null,
"accuracy": 0.58,
"f1": 0.3,
"balanced_accuracy": 0.58,
"accur... |
0 | 7 | 512 | Final Syllable Orthographic Pattern | The gate activates on the final syllable or morpheme of a word, particularly when that ending is "lon", "lop", "loy", "lux", "lad", "la", or "l". This pattern spans diverse semantic categories including proper nouns (Caslon, Lisa), common nouns (nylon, scallops, loyalty), and verbs (gallops, deglaze). The trigger is pr... | complete | 9,246 | 0.000925 | 2.453125 | 1.000172 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 4,
"tn": 50,
"fp": 0,
"fn": 46,
"correct": null,
"accuracy": 0.54,
"f1": 0.14814814814814814,
"balanced_accuracy": 0.... |
0 | 8 | 512 | Final Word Token Suffix Pattern | The final token of a word, often a suffix or ending, appearing in diverse contexts. | complete | 9,309 | 0.000931 | 2.28125 | 1.189795 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 13,
"tn": 42,
"fp": 8,
"fn": 37,
"correct": null,
"accuracy": 0.55,
"f1": 0.36619718309859156,
"balanced_accuracy": 0... |
0 | 9 | 512 | Body Part and Header Nouns | The gate activates on nouns denoting physical body parts (e.g., noses, eyes, face) or structural headers (e.g., head, sizes) often appearing in descriptive or technical contexts. | complete | 22,222 | 0.002222 | 1.859375 | 1.023089 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 21,
"tn": 49,
"fp": 1,
"fn": 29,
"correct": null,
"accuracy": 0.7,
"f1": 0.5833333333333334,
"balanced_accuracy": 0.7... |
0 | 10 | 512 | Suffix-based person-role and high-frequency lexical gate | The gate activates on tokens ending in "er" or "ers" that denote a person or role (e.g., winners, attackers, beginners, announcer, infielder, wanderer, borrower, designer, laborers, customer). It also fires on the adjective "classic" and the intensifier "super," suggesting a broader pattern of common, high-frequency le... | complete | 6,978 | 0.000698 | 1.820313 | 0.93864 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 17,
"tn": 44,
"fp": 6,
"fn": 33,
"correct": null,
"accuracy": 0.61,
"f1": 0.4657534246575342,
"balanced_accuracy": 0.... |
0 | 11 | 512 | Positive Value and Approval Lexicon | Adjectives and nouns denoting positive value, utility, or social approval, such as "important," "valuable," "applaud," "grateful," and "appreciated." | complete | 18,132 | 0.001813 | 1.914063 | 1.052996 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 40,
"tn": 35,
"fp": 15,
"fn": 10,
"correct": null,
"accuracy": 0.75,
"f1": 0.7619047619047619,
"balanced_accuracy": 0... |
0 | 12 | 512 | Modal "Can" Object "You" | The gate activates on the token "you" when it appears as the object of the modal verb "can" in questions or statements, indicating a request for capability or permission. | complete | 69,597 | 0.00696 | 2.296875 | 0.942188 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 20,
"tn": 50,
"fp": 0,
"fn": 30,
"correct": null,
"accuracy": 0.7,
"f1": 0.5714285714285714,
"balanced_accuracy": 0.7... |
0 | 13 | 512 | Proper Nouns and Technical Acronyms | The gate activates on proper nouns, specific technical acronyms, and capitalized terms denoting distinct entities, institutions, or specialized concepts. | complete | 22,035 | 0.002204 | 2.34375 | 0.979463 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 25,
"tn": 26,
"fp": 24,
"fn": 25,
"correct": null,
"accuracy": 0.51,
"f1": 0.5050505050505051,
"balanced_accuracy": 0... |
0 | 14 | 512 | Proper Noun and Entity Termination | The gate activates on the final token of a proper noun, country name, or specific entity, often appearing in lists, historical contexts, or geographical references. | complete | 8,097 | 0.00081 | 2.203125 | 0.961087 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 4,
"tn": 49,
"fp": 1,
"fn": 46,
"correct": null,
"accuracy": 0.53,
"f1": 0.14545454545454545,
"balanced_accuracy": 0.... |
0 | 15 | 512 | Suffix Completion Gate | The gate activates on the final token of a word, specifically when that token completes a common English suffix or morpheme (e.g., -ation, -ing, -ous, -ly, -er, -s). The activation strength correlates with the frequency and predictability of the word ending, firing strongly on standard suffixes like "omatic" or "ci" in... | complete | 23,041 | 0.002304 | 2.390625 | 1.186496 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 4,
"tn": 41,
"fp": 9,
"fn": 46,
"correct": null,
"accuracy": 0.45,
"f1": 0.12698412698412698,
"balanced_accuracy": 0.... |
0 | 16 | 512 | Final Token of Proper Noun | The gate activates on the final token of a proper noun, specifically surnames or names of people, places, and brands. | complete | 9,211 | 0.000921 | 2.6875 | 1.110261 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 20,
"tn": 50,
"fp": 0,
"fn": 30,
"correct": null,
"accuracy": 0.7,
"f1": 0.5714285714285714,
"balanced_accuracy": 0.7... |
0 | 17 | 512 | Key Entity and Technical Term Activation | The gate activates on specific nouns, verbs, or technical terms that serve as key entities, ingredients, medical conditions, or proper nouns within their respective contexts. | complete | 8,234 | 0.000823 | 2.375 | 0.985089 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 47,
"tn": 20,
"fp": 30,
"fn": 3,
"correct": null,
"accuracy": 0.67,
"f1": 0.7401574803149606,
"balanced_accuracy": 0.... |
0 | 18 | 512 | Proper Noun and Technical Term Completion | The gate activates on the final token of a word, particularly when that word is a proper noun, a specific entity, or a technical term. It fires strongly on capitalized names (e.g., "Curie", "Myanmar") and specific nouns (e.g., "sulfur", "curry"), often marking the completion of a significant lexical item in the text. | complete | 8,668 | 0.000867 | 2.296875 | 0.999583 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 24,
"tn": 33,
"fp": 17,
"fn": 26,
"correct": null,
"accuracy": 0.57,
"f1": 0.5274725274725275,
"balanced_accuracy": 0... |
0 | 19 | 512 | Tokenization Boundary Sensitivity | The gate activates on tokens that are frequently truncated or split in the training data, such as proper nouns (Laravel, O'Reilly), medical terms (chikungunya, intestine), and common nouns (internet, capacity). It fires strongly on the specific subword or token ending of these entities, often regardless of the surround... | complete | 5,146 | 0.000515 | 1.578125 | 0.951297 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 15,
"tn": 46,
"fp": 4,
"fn": 35,
"correct": null,
"accuracy": 0.61,
"f1": 0.43478260869565216,
"balanced_accuracy": 0... |
0 | 20 | 512 | Keep and Keeper Suffix Trigger | The gate activates on the suffix "keep" or "keeper" within compound nouns (e.g., "peacekeepers") or as a standalone verb (e.g., "keep", "keeping", "kept"). It fires strongly on the specific token "keepers" and moderately on various conjugations of "keep", regardless of the surrounding semantic context, which ranges fro... | complete | 8,314 | 0.000831 | 1.75 | 1.163344 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 35,
"tn": 50,
"fp": 0,
"fn": 15,
"correct": null,
"accuracy": 0.85,
"f1": 0.8235294117647058,
"balanced_accuracy": 0.... |
0 | 21 | 512 | Morphological Suffix and Proper Noun Completion | The gate activates on the final token of a word, particularly when that token completes a specific morphological suffix (e.g., -atic, -ant, -ous) or a proper noun. The activation strength correlates with the distinctiveness of the word ending, with high activations for common suffixes like "atic" and "ant" in academic ... | complete | 4,459 | 0.000446 | 1.78125 | 1.017491 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 0,
"tn": 50,
"fp": 0,
"fn": 50,
"correct": null,
"accuracy": 0.5,
"f1": 0,
"balanced_accuracy": 0.5,
"accuracy_... |
0 | 22 | 512 | Proper Noun and Technical Suffixes | The gate activates on the suffixes of proper nouns, place names, and specific technical terms, often appearing mid-word or at the end of a token. | complete | 8,560 | 0.000856 | 1.984375 | 1.047657 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 16,
"tn": 36,
"fp": 14,
"fn": 34,
"correct": null,
"accuracy": 0.52,
"f1": 0.4,
"balanced_accuracy": 0.52,
"acc... |
0 | 23 | 512 | Physical Component and Phenomenon Nouns | The gate activates on nouns and technical terms denoting physical components, mechanical parts, or scientific phenomena. It fires strongly on specific hardware (axles, motors, coils, tanks) and scientific concepts (resonance, expansion, magnetism), often appearing in technical descriptions, engineering contexts, or lis... | complete | 2,619 | 0.000262 | 1.726563 | 0.764147 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 38,
"tn": 47,
"fp": 3,
"fn": 12,
"correct": null,
"accuracy": 0.85,
"f1": 0.8351648351648352,
"balanced_accuracy": 0.... |
0 | 24 | 512 | Capitalized Nouns in Numbered Lists | The gate activates on capitalized nouns or proper nouns that serve as the first item in a numbered list or bullet point, often following a colon or introductory phrase. | complete | 6,007 | 0.000601 | 1.765625 | 0.727703 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 5,
"tn": 46,
"fp": 4,
"fn": 45,
"correct": null,
"accuracy": 0.51,
"f1": 0.1694915254237288,
"balanced_accuracy": 0.5... |
0 | 25 | 512 | Compound Noun Suffix Gate | The gate fires on the second syllable or suffix of compound proper nouns (e.g., FitBit, HubSpot, Habitat) and common nouns (e.g., array, suite, list, spot, bit). It activates on tokens that complete a multi-word entity or specific technical term, often appearing in lists, code contexts, or brand names. | complete | 16,851 | 0.001685 | 2.28125 | 1.012009 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 6,
"tn": 50,
"fp": 0,
"fn": 44,
"correct": null,
"accuracy": 0.56,
"f1": 0.21428571428571427,
"balanced_accuracy": 0.... |
0 | 26 | 512 | Initial Numerical Quantifier Triggers | Numerical values or quantifiers, often appearing as the first word of a sentence or clause, indicating a count, duration, or specific quantity. | complete | 22,476 | 0.002248 | 2.71875 | 1.221933 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 8,
"tn": 44,
"fp": 6,
"fn": 42,
"correct": null,
"accuracy": 0.52,
"f1": 0.25,
"balanced_accuracy": 0.52,
"accu... |
0 | 27 | 512 | Suffix-Triggered Common Noun and Verb Gate | The gate activates on common nouns and verbs, particularly those ending in specific suffixes like -shade, -shirt, -pace, -shed, or -replace, often appearing in lists, instructions, or descriptive contexts. | complete | 7,603 | 0.00076 | 2.109375 | 1.006762 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 13,
"tn": 50,
"fp": 0,
"fn": 37,
"correct": null,
"accuracy": 0.63,
"f1": 0.4126984126984127,
"balanced_accuracy": 0.... |
0 | 28 | 512 | Final Token Activation of Common Bigrams | The gate activates on the final token of common bigrams or compound nouns, such as "social media," "climate change," "mental health," and "biological clock." It fires when the preceding context forms a standard phrase, with higher activations for more specific or technical terms. | complete | 22,770 | 0.002277 | 2.421875 | 0.980119 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 13,
"tn": 50,
"fp": 0,
"fn": 37,
"correct": null,
"accuracy": 0.63,
"f1": 0.4126984126984127,
"balanced_accuracy": 0.... |
0 | 29 | 512 | Suffix and Noun Trigger Gate | The gate activates on tokens ending in "ing" or "ues", as well as specific nouns like "thing", "feeling", "understanding", and "quantity". | complete | 16,588 | 0.001659 | 1.96875 | 1.016998 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 38,
"tn": 33,
"fp": 17,
"fn": 12,
"correct": null,
"accuracy": 0.71,
"f1": 0.7238095238095238,
"balanced_accuracy": 0... |
0 | 30 | 512 | Big/Small Adjective Noun-Head Trigger | The gate activates on nouns that serve as the head of a noun phrase immediately preceded by the adjective "big" or "small". This includes compound nouns (e.g., "big-name", "small-cell") and standard nouns (e.g., "big club", "small house"). The activation strength correlates with the semantic concreteness or prominence ... | complete | 2,357 | 0.000236 | 1.78125 | 0.728669 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 16,
"tn": 50,
"fp": 0,
"fn": 34,
"correct": null,
"accuracy": 0.66,
"f1": 0.48484848484848486,
"balanced_accuracy": 0... |
0 | 31 | 512 | Preposition "for" Purpose and Benefit | The preposition "for" following a noun or verb, indicating purpose, benefit, or suitability. | complete | 89,249 | 0.008925 | 2.34375 | 1.508189 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 44,
"tn": 50,
"fp": 0,
"fn": 6,
"correct": null,
"accuracy": 0.94,
"f1": 0.9361702127659575,
"balanced_accuracy": 0.9... |
0 | 32 | 512 | Proper Noun Final Token Gate | The gate activates on the final token of a word, specifically when the word is a proper noun, a specific entity name, or a distinct noun phrase often found in lists, titles, or specific contexts. | complete | 108,753 | 0.010875 | 5.96875 | 1.360041 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 8,
"tn": 49,
"fp": 1,
"fn": 42,
"correct": null,
"accuracy": 0.57,
"f1": 0.2711864406779661,
"balanced_accuracy": 0.5... |
0 | 33 | 512 | im_end and code identifier gate | The gate fires strongly on the special token ⟪im_end⟫, which marks the end of a user prompt or context block in the chat format. It also activates on specific code identifiers, class names, and configuration keys (e.g., `DecimalFormat`, `MongoClient`, `apiKey`, `navigationOptions`) within programming examples, as well ... | complete | 111,374 | 0.011137 | 5.34375 | 2.56572 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 49,
"tn": 50,
"fp": 0,
"fn": 1,
"correct": null,
"accuracy": 0.99,
"f1": 0.98989898989899,
"balanced_accuracy": 0.99,... |
0 | 34 | 512 | Not observed in 10M tokens | null | not_observed | 0 | 0 | 0 | 0 | {
"train": null,
"test": null,
"negatives": null,
"active_fuzz_positives": null,
"short_windows": null,
"full_budget": null
} | [] | {
"detection": {
"status": "unavailable",
"reason": "No positive activation evidence in this cache",
"metrics": {
"total": null,
"valid": null,
"parse_failures": null,
"tp": null,
"tn": null,
"fp": null,
"fn": null,
"correct": null,
"accuracy": null,
... |
0 | 35 | 512 | Preposition "of" Noun Association | The preposition "of" following a noun, often indicating possession, composition, or association. | complete | 4,923 | 0.000492 | 1.835938 | 0.920972 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 37,
"tn": 21,
"fp": 29,
"fn": 13,
"correct": null,
"accuracy": 0.58,
"f1": 0.6379310344827587,
"balanced_accuracy": 0... |
0 | 36 | 512 | Gradual Change and Growth Verbs | The gate activates on verbs or adjectives describing gradual change, growth, or increase, often in the context of trends, development, or physical expansion. | complete | 9,016 | 0.000902 | 1.992188 | 1.090926 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 36,
"tn": 44,
"fp": 6,
"fn": 14,
"correct": null,
"accuracy": 0.8,
"f1": 0.782608695652174,
"balanced_accuracy": 0.8,... |
0 | 37 | 512 | Proper Noun Final Token | The final token of a word, specifically when the word is a proper noun, brand name, or specific entity. | complete | 13,322 | 0.001332 | 2.375 | 1.051109 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 5,
"tn": 49,
"fp": 1,
"fn": 45,
"correct": null,
"accuracy": 0.54,
"f1": 0.17857142857142858,
"balanced_accuracy": 0.... |
0 | 38 | 512 | Final Token of Multi-Word Proper Nouns | The final token of a multi-word proper noun, brand name, or specific technical term, often appearing in business, technology, or media contexts. | complete | 2,639 | 0.000264 | 1.5625 | 0.76282 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 8,
"tn": 48,
"fp": 2,
"fn": 42,
"correct": null,
"accuracy": 0.56,
"f1": 0.26666666666666666,
"balanced_accuracy": 0.... |
0 | 39 | 512 | Quantifier-Of Noun Phrase Introduction | The token "of" (or "-of") following a quantifier or indefinite pronoun (e.g., "some", "many", "one") to introduce a noun phrase, often in list introductions or general statements. | complete | 6,097 | 0.00061 | 1.84375 | 1.139975 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 31,
"tn": 47,
"fp": 3,
"fn": 19,
"correct": null,
"accuracy": 0.78,
"f1": 0.7380952380952381,
"balanced_accuracy": 0.... |
0 | 40 | 512 | Action-Oriented Vocabulary Gate | The gate activates on the token "work" or its inflections (e.g., "works", "worked") when used as a noun or verb, particularly in contexts involving labor, craftsmanship, or functional mechanisms. It also fires on related tokens like "set", "get", "getting", "dress", and "man" in specific idiomatic or procedural phrases... | complete | 36,386 | 0.003639 | 2.078125 | 1.044103 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 36,
"tn": 50,
"fp": 0,
"fn": 14,
"correct": null,
"accuracy": 0.86,
"f1": 0.8372093023255814,
"balanced_accuracy": 0.... |
0 | 41 | 512 | Morphological Suffix Trigger | The gate activates on the final suffix of a word, particularly when the word is a noun or adjective ending in a common morphological ending like -ance, -ence, -ness, -ment, -tion, -ous, -ity, -ure, -ism, -al, -ic, -ly, -ed, -ing, -er, -est, -s, -es, -ies, -y, -o, -u, -a, -i, -e, -o, -u, -a, -i, -e. | complete | 11,260 | 0.001126 | 2.40625 | 1.060383 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 25,
"tn": 39,
"fp": 11,
"fn": 25,
"correct": null,
"accuracy": 0.64,
"f1": 0.5813953488372093,
"balanced_accuracy": 0... |
0 | 42 | 512 | Auxiliary "will" and "won't" Future Tense | The gate activates on the auxiliary verb "will" and its negative contraction "won't", particularly when they appear as the main verb in a sentence or clause, often following a subject or conjunction to indicate future tense or modality. | complete | 25,670 | 0.002567 | 2.109375 | 1.039367 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 30,
"tn": 50,
"fp": 0,
"fn": 20,
"correct": null,
"accuracy": 0.8,
"f1": 0.75,
"balanced_accuracy": 0.8,
"accur... |
0 | 43 | 512 | Proper Noun Final Token Gate | The gate activates on the final token of a proper noun, brand name, or specific entity, often appearing at the end of a sentence or phrase. | complete | 4,048 | 0.000405 | 1.65625 | 0.985713 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 1,
"tn": 48,
"fp": 2,
"fn": 49,
"correct": null,
"accuracy": 0.49,
"f1": 0.03773584905660377,
"balanced_accuracy": 0.... |
0 | 44 | 512 | Standalone Noun and Adjective Triggers | The gate activates on the token "stone" (or variants like "stones", "Ston") when it appears as a standalone noun, particularly in the context of "capstone" projects, gemstones, or architectural materials. It also fires on specific adjectives like "bold" and "characterized" in descriptive contexts, and on the replacemen... | complete | 5,932 | 0.000593 | 2.09375 | 0.940807 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 13,
"tn": 50,
"fp": 0,
"fn": 37,
"correct": null,
"accuracy": 0.63,
"f1": 0.4126984126984127,
"balanced_accuracy": 0.... |
0 | 45 | 512 | Political and Military Entity Tokens | The gate activates on tokens denoting political or military entities, structures, and concepts. This includes specific organizations (e.g., "Pact", "Assembly"), roles (e.g., "ruler", "consul"), and abstract political terms (e.g., "repression", "loyalty", "independence"). The context is typically historical, geopolitica... | complete | 1,825 | 0.000183 | 1.65625 | 0.746813 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 44,
"tn": 48,
"fp": 2,
"fn": 6,
"correct": null,
"accuracy": 0.92,
"f1": 0.9166666666666666,
"balanced_accuracy": 0.9... |
0 | 46 | 512 | Conjunction "or" and non-ASCII symbols | The gate activates on the conjunction "or" and specific non-ASCII characters, including special symbols like fractions (½, ¾) and accented letters (ø, å, ð, ö, ä). | complete | 54,740 | 0.005474 | 2.28125 | 1.027651 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 0,
"tn": 50,
"fp": 0,
"fn": 50,
"correct": null,
"accuracy": 0.5,
"f1": 0,
"balanced_accuracy": 0.5,
"accuracy_... |
0 | 47 | 512 | Preposition "of" Possessive Composition Association | The preposition "of" following a noun, indicating possession, composition, or association. | complete | 5,433 | 0.000543 | 2.109375 | 0.877673 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 47,
"tn": 23,
"fp": 27,
"fn": 3,
"correct": null,
"accuracy": 0.7,
"f1": 0.7580645161290323,
"balanced_accuracy": 0.7... |
0 | 48 | 512 | Suffix -ate/-ation Token Split Gate | The gate activates on the suffix "-ate" or "-ation" in verbs and nouns, particularly when the word is split across tokens (e.g., "punct" + "uate"). It fires strongly on common words like "punctuate," "levitate," "eliminate," and "celebrate," and weaker on less common or foreign-derived terms. | complete | 9,411 | 0.000941 | 2.1875 | 1.111876 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 8,
"tn": 50,
"fp": 0,
"fn": 42,
"correct": null,
"accuracy": 0.58,
"f1": 0.27586206896551724,
"balanced_accuracy": 0.... |
0 | 49 | 512 | High-Frequency Word Suffix Gate | The gate activates on the final suffix of a word, particularly when the word is a common noun, verb, or adjective. The activation strength correlates with the frequency of the word in the training data, with high-frequency words like "overwrite," "parable," and "peppers" showing the strongest signals, while rarer or mo... | complete | 5,862 | 0.000586 | 1.65625 | 0.969406 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 6,
"tn": 50,
"fp": 0,
"fn": 44,
"correct": null,
"accuracy": 0.56,
"f1": 0.21428571428571427,
"balanced_accuracy": 0.... |
0 | 50 | 512 | Bigram, Suffix, and Contextual Word Triggers | The gate activates on the second token of the bigram "out of" or "up to", as well as the suffix "graded" in "downgraded". It also fires on the word "input" in programming contexts and "standing" in "Outstanding". | complete | 11,811 | 0.001181 | 2.28125 | 1.136438 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 27,
"tn": 49,
"fp": 1,
"fn": 23,
"correct": null,
"accuracy": 0.76,
"f1": 0.6923076923076923,
"balanced_accuracy": 0.... |
0 | 51 | 512 | Suffix “st” or “ist” Gate | The gate fires on the suffix "st" or "ist" when it appears as the final token of a word, particularly in proper nouns, titles, or technical terms. | complete | 9,959 | 0.000996 | 2.046875 | 1.054775 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 10,
"tn": 50,
"fp": 0,
"fn": 40,
"correct": null,
"accuracy": 0.6,
"f1": 0.3333333333333333,
"balanced_accuracy": 0.6... |
0 | 52 | 512 | Dollar Sign Currency or Variable | The token "$" appearing as a currency symbol or variable prefix, often preceded by a space, parenthesis, or bracket, and followed by a digit, letter, or closing bracket. | complete | 10,133 | 0.001013 | 2.65625 | 1.088038 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 13,
"tn": 50,
"fp": 0,
"fn": 37,
"correct": null,
"accuracy": 0.63,
"f1": 0.4126984126984127,
"balanced_accuracy": 0.... |
0 | 53 | 512 | Customs, Regulations, and Bureaucratic Context | The gate activates on nouns and adjectives related to customs, regulations, or formal procedures, such as "customs," "customize," "customary," "protections," and "precautions." It also fires on specific proper nouns like "Stamford" and "nationals," suggesting a latent concept of administrative or bureaucratic context r... | complete | 8,663 | 0.000866 | 1.75 | 0.945579 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 19,
"tn": 50,
"fp": 0,
"fn": 31,
"correct": null,
"accuracy": 0.69,
"f1": 0.5507246376811594,
"balanced_accuracy": 0.... |
0 | 54 | 512 | Suffix and Compound Word Completion | The gate activates on the final token of a word, specifically when that token completes a common suffix (e.g., -ual, -er, -ous, -ing, -tion) or a compound word part (e.g., -hole, -roach, -ower). The activation strength correlates with the frequency or commonality of the suffix in the training data, with high activation... | complete | 3,656 | 0.000366 | 1.914063 | 0.866834 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 3,
"tn": 48,
"fp": 2,
"fn": 47,
"correct": null,
"accuracy": 0.51,
"f1": 0.10909090909090909,
"balanced_accuracy": 0.... |
0 | 55 | 512 | Word-Split Suffix Token Detection | The gate activates on tokens that are the final part of a word, specifically when the word is split across token boundaries. This includes suffixes like "er", "s", "es", or the end of proper nouns and common nouns. | complete | 7,433 | 0.000743 | 1.734375 | 1.05627 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 7,
"tn": 48,
"fp": 2,
"fn": 43,
"correct": null,
"accuracy": 0.55,
"f1": 0.23728813559322035,
"balanced_accuracy": 0.... |
0 | 56 | 512 | List-Extension and Qualification Markers | The gate activates on tokens that introduce additional items, examples, or qualifications in a list or explanation. This includes conjunctions (and, or), prepositions (like, as, with), and specific markers (etc, plus, specifically, also, however, even, while, although, albeit, sometimes, possibly, hence, thus, because,... | complete | 197,247 | 0.019725 | 1.390625 | 0.80461 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 39,
"tn": 17,
"fp": 33,
"fn": 11,
"correct": null,
"accuracy": 0.56,
"f1": 0.639344262295082,
"balanced_accuracy": 0.... |
0 | 57 | 512 | Proper Nouns, Medical Terms, and Ordinal Suffixes | The gate activates on specific proper nouns (Thomson, Sherlock, McCarthy, LeBron) and medical/technical terms (symptom, cytotoxic, Equifax). It also fires on the suffix "th" in ordinal numbers (19th, 20th) and the word "topic". | complete | 11,276 | 0.001128 | 1.914063 | 0.997576 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 17,
"tn": 50,
"fp": 0,
"fn": 33,
"correct": null,
"accuracy": 0.67,
"f1": 0.5074626865671642,
"balanced_accuracy": 0.... |
0 | 58 | 512 | First Name and Proper Noun Trigger | The gate activates on proper nouns, specifically first names, and occasionally on specific proper nouns like place names or brand names. | complete | 8,846 | 0.000885 | 1.859375 | 0.94837 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 24,
"tn": 29,
"fp": 21,
"fn": 26,
"correct": null,
"accuracy": 0.53,
"f1": 0.5052631578947369,
"balanced_accuracy": 0... |
0 | 59 | 512 | Cuisine-Triggered MOLT Gate Activation | The gate activates on specific nouns and adjectives, with the strongest response to "cuisine" in culinary contexts. Weaker activations occur for terms like "lifespan," "durable," "occasion," and "contemporary," often appearing at the end of sentences or clauses. | complete | 9,027 | 0.000903 | 1.882813 | 1.039057 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 14,
"tn": 50,
"fp": 0,
"fn": 36,
"correct": null,
"accuracy": 0.64,
"f1": 0.4375,
"balanced_accuracy": 0.64,
"a... |
0 | 60 | 512 | Proper Noun Final Token Gate | The gate activates on the final token of a proper noun, name, or specific entity, often appearing at the end of a sentence or clause. | complete | 5,171 | 0.000517 | 1.960938 | 0.917565 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 4,
"tn": 50,
"fp": 0,
"fn": 46,
"correct": null,
"accuracy": 0.54,
"f1": 0.14814814814814814,
"balanced_accuracy": 0.... |
0 | 61 | 512 | Sentence-End Token Gate | The gate activates on tokens that are the final word of a sentence or clause, often following a comma or period, indicating a pause or completion in the text flow. | complete | 12,942 | 0.001294 | 2.109375 | 1.000931 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 16,
"tn": 25,
"fp": 25,
"fn": 34,
"correct": null,
"accuracy": 0.41,
"f1": 0.3516483516483517,
"balanced_accuracy": 0... |
0 | 62 | 512 | Conditional and Possessive Token Triggers | The token "if" or "of" appearing in code conditional statements or natural language phrases indicating a condition or possession. | complete | 20,246 | 0.002025 | 3.296875 | 1.107154 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 40,
"tn": 31,
"fp": 19,
"fn": 10,
"correct": null,
"accuracy": 0.71,
"f1": 0.7339449541284404,
"balanced_accuracy": 0... |
0 | 63 | 512 | Second Token of Two-Token Proper Nouns | The gate activates on the second token of a two-token proper noun or brand name, such as the suffix in "Hyperledger", "NetWare", "NetBeans", "Unsplash", "Google Translate", "Microsoft Office", "Google Docs", "Google LLC", "Zillow", "AWS CloudWatch", "Super Bowl", "AngularJS", "UNICEF", "McGregor", "Pyrex", "Cloudflare"... | complete | 3,563 | 0.000356 | 2.25 | 0.923241 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 0,
"tn": 50,
"fp": 0,
"fn": 50,
"correct": null,
"accuracy": 0.5,
"f1": 0,
"balanced_accuracy": 0.5,
"accuracy_... |
0 | 64 | 512 | Suffixes and Substrings of "Syn" | The gate activates on tokens that are suffixes or substrings of the word "syn" (e.g., "Syn", "syn", "synchronous", "Polynesia", "assembly"), often appearing in technical, geographical, or organizational contexts. | complete | 14,622 | 0.001462 | 2.390625 | 1.154915 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 2,
"tn": 47,
"fp": 3,
"fn": 48,
"correct": null,
"accuracy": 0.49,
"f1": 0.07272727272727272,
"balanced_accuracy": 0.... |
0 | 65 | 512 | Preposition "with" for Accompaniment | The preposition "with" (or "dengan") following a verb or adjective to indicate accompaniment, possession, or interaction. | complete | 67,821 | 0.006782 | 2.984375 | 1.991539 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 33,
"tn": 50,
"fp": 0,
"fn": 17,
"correct": null,
"accuracy": 0.83,
"f1": 0.7951807228915663,
"balanced_accuracy": 0.... |
0 | 66 | 512 | Final Word Token Suffix Pattern | The final token of a word, often a suffix or ending, appearing in diverse contexts. | complete | 16,419 | 0.001642 | 2.21875 | 1.062264 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 9,
"tn": 45,
"fp": 5,
"fn": 41,
"correct": null,
"accuracy": 0.54,
"f1": 0.28125,
"balanced_accuracy": 0.54,
"a... |
0 | 67 | 512 | Proper Noun Token Gate | The gate activates on the first token of a proper noun, name, or specific entity, often appearing in the middle of a sentence or list. | complete | 18,949 | 0.001895 | 2.71875 | 0.942568 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 15,
"tn": 42,
"fp": 8,
"fn": 35,
"correct": null,
"accuracy": 0.57,
"f1": 0.410958904109589,
"balanced_accuracy": 0.5... |
0 | 68 | 512 | Abstract State and Process Token Trigger | The gate activates on tokens representing abstract states, processes, or conditions, such as change, maintenance, benign, diffuse, or latent. It fires on both full words and truncated stems (e.g., "chang", "disen") across diverse contexts including business, health, art, and technical descriptions, indicating a broad s... | complete | 18,607 | 0.001861 | 2.125 | 1.028578 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 29,
"tn": 38,
"fp": 12,
"fn": 21,
"correct": null,
"accuracy": 0.67,
"f1": 0.6373626373626373,
"balanced_accuracy": 0... |
0 | 69 | 512 | URL and Web Technical Token Gate | The gate activates on tokens within URLs, specifically domain suffixes (e.g., .com, .gov) and path segments (e.g., /articles, /images). It also fires on specific technical terms like "api", "css", and "html5", as well as common web-related nouns like "pages" and "engines". | complete | 2,132 | 0.000213 | 1.796875 | 0.912679 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 22,
"tn": 49,
"fp": 1,
"fn": 28,
"correct": null,
"accuracy": 0.71,
"f1": 0.6027397260273972,
"balanced_accuracy": 0.... |
0 | 70 | 512 | Negative Outcome and Incompatibility Tokens | The gate activates on tokens indicating negative outcomes, limitations, or incompatibilities, such as "incompatible," "disrespect," "decline," "unable," "cannot," "failure," and "deterioration." It fires when the text describes a lack of fit, a decrease in quality or value, an inability to perform an action, or a negat... | complete | 15,751 | 0.001575 | 3.109375 | 1.090691 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 20,
"tn": 47,
"fp": 3,
"fn": 30,
"correct": null,
"accuracy": 0.67,
"f1": 0.547945205479452,
"balanced_accuracy": 0.6... |
0 | 71 | 512 | Suffix and Short Word Token Gate | The gate activates on the final token of a word, particularly when that token is a common suffix (e.g., -ing, -er, -es) or a short word like "go," "start," or "ve." | complete | 15,910 | 0.001591 | 1.859375 | 1.054249 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 18,
"tn": 48,
"fp": 2,
"fn": 32,
"correct": null,
"accuracy": 0.66,
"f1": 0.5142857142857142,
"balanced_accuracy": 0.... |
0 | 72 | 512 | Final Token Word Completion Gate | The gate activates on the final token of a word, particularly when the word is a common noun or adjective (e.g., "charm", "chart", "charge", "garlic", "chronic") or a proper noun. It fires strongly on complete words like "Charm" and "chart", and more weakly on truncated or less common words like "gar" or "ordinated". | complete | 14,817 | 0.001482 | 2.046875 | 1.00766 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 18,
"tn": 45,
"fp": 5,
"fn": 32,
"correct": null,
"accuracy": 0.63,
"f1": 0.4931506849315068,
"balanced_accuracy": 0.... |
0 | 73 | 512 | Suffix "hip" and Business Term Gate | The gate activates on the suffix "hip" when it forms abstract nouns (e.g., discipleship, viewership, homeownership) or on the word "sales" in business contexts. It also fires on "delivery" in various contexts, "dialysis," and specific proper nouns like "Lilly" or "Lockheed," suggesting a focus on specific morphological... | complete | 4,527 | 0.000453 | 1.828125 | 1.075942 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 27,
"tn": 50,
"fp": 0,
"fn": 23,
"correct": null,
"accuracy": 0.77,
"f1": 0.7012987012987013,
"balanced_accuracy": 0.... |
0 | 74 | 512 | Second-Person Pronoun and Auxiliary Trigger | The gate activates on the second-person pronoun "you" and its immediate following auxiliary or modal verb (e.g., "can", "will", "think", "ll"). This pattern frequently occurs in instructional, conversational, or advisory contexts where the model addresses the user directly to provide information, suggestions, or confir... | complete | 34,870 | 0.003487 | 2.421875 | 0.918438 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 24,
"tn": 50,
"fp": 0,
"fn": 26,
"correct": null,
"accuracy": 0.74,
"f1": 0.6486486486486487,
"balanced_accuracy": 0.... |
0 | 75 | 512 | "Sense of" Emotional State Trigger | The token "of" following "sense" in phrases describing emotional or atmospheric states, such as "sense of wonder" or "sense of dread." | complete | 6,965 | 0.000697 | 1.796875 | 1.185619 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 24,
"tn": 50,
"fp": 0,
"fn": 26,
"correct": null,
"accuracy": 0.74,
"f1": 0.6486486486486487,
"balanced_accuracy": 0.... |
0 | 76 | 512 | Uppercase Acronym Sentence Termination | The gate activates on uppercase acronyms, abbreviations, or proper nouns, often appearing at the end of a sentence or list item. | complete | 24,336 | 0.002434 | 3.296875 | 1.291519 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 21,
"tn": 45,
"fp": 5,
"fn": 29,
"correct": null,
"accuracy": 0.66,
"f1": 0.5526315789473685,
"balanced_accuracy": 0.... |
0 | 77 | 512 | Administrative and Legal Entity Nouns | The gate activates on nouns denoting specific administrative, legal, or institutional entities, such as government departments, courts, or organizations. | complete | 26,883 | 0.002688 | 2.390625 | 1.062249 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 18,
"tn": 47,
"fp": 3,
"fn": 32,
"correct": null,
"accuracy": 0.65,
"f1": 0.5070422535211268,
"balanced_accuracy": 0.... |
0 | 78 | 512 | Authority and Command Token Gate | The gate activates on tokens indicating authority, command, or control, such as "commanded," "commander," and "command," as well as related terms like "hostage" and "bargaining." | complete | 10,045 | 0.001005 | 2 | 1.018425 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 3,
"tn": 50,
"fp": 0,
"fn": 47,
"correct": null,
"accuracy": 0.53,
"f1": 0.11320754716981132,
"balanced_accuracy": 0.... |
0 | 79 | 512 | Proper Noun and Named Entity Gate | The gate activates on tokens that are part of proper nouns, specific named entities, or capitalized terms, often appearing within compound words, titles, or specific identifiers. | complete | 7,755 | 0.000776 | 1.773438 | 0.92782 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 32,
"tn": 26,
"fp": 24,
"fn": 18,
"correct": null,
"accuracy": 0.58,
"f1": 0.6037735849056604,
"balanced_accuracy": 0... |
0 | 80 | 256 | Strategic Planning Contexts | The token "planning" or "planned" appears in contexts related to organizational, strategic, or project-based activities, often preceded by adjectives like "strategic," "careful," or "individualized." | complete | 3,654 | 0.000365 | 1.757813 | 1.100754 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 15,
"tn": 49,
"fp": 1,
"fn": 35,
"correct": null,
"accuracy": 0.64,
"f1": 0.45454545454545453,
"balanced_accuracy": 0... |
0 | 81 | 256 | Final Token Proper Noun Activation | The gate activates on the final token of a word, specifically when that token is part of a proper noun, place name, or specific entity (e.g., Mesopotamian, Canaan, Rockefeller Center). It also fires on the final token of common nouns and adjectives, often in the context of lists, descriptions, or specific cultural refe... | complete | 61,507 | 0.006151 | 2.328125 | 1.148592 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 13,
"tn": 37,
"fp": 13,
"fn": 37,
"correct": null,
"accuracy": 0.5,
"f1": 0.34210526315789475,
"balanced_accuracy": 0... |
0 | 82 | 256 | Named Entity and Abbreviation Detection | The gate activates on tokens representing proper nouns, acronyms, or specific entities, particularly when they appear as capitalized abbreviations (e.g., MMP, MMO, CPA) or within compound words (e.g., re-establish, hempcrete). It also fires on specific names and locations, indicating a focus on identifying distinct nam... | complete | 5,541 | 0.000554 | 2.34375 | 1.238818 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 29,
"tn": 32,
"fp": 18,
"fn": 21,
"correct": null,
"accuracy": 0.61,
"f1": 0.5979381443298969,
"balanced_accuracy": 0... |
0 | 83 | 256 | Adverbial Modification of Manner and Degree | Adverbs modifying verbs or adjectives to describe manner, frequency, or degree, often appearing in instructional, descriptive, or conversational contexts. | complete | 102,453 | 0.010245 | 3.109375 | 1.405524 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 25,
"tn": 40,
"fp": 10,
"fn": 25,
"correct": null,
"accuracy": 0.65,
"f1": 0.5882352941176471,
"balanced_accuracy": 0... |
0 | 84 | 256 | Final Segment of Compound Entities | The gate activates on tokens that are the final segment of a compound word, proper noun, or technical identifier, often appearing at the end of a phrase or sentence. This includes suffixes like "er", "newfound", "Mandela", "Shakhtar", "Javed Akhtar", "Etihad", "Hanukkah", "Sub-Saharan", "tapioca", "haplotype", "fodder"... | complete | 2,276 | 0.000228 | 3.109375 | 1.075008 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 25,
"tn": 50,
"fp": 0,
"fn": 25,
"correct": null,
"accuracy": 0.75,
"f1": 0.6666666666666666,
"balanced_accuracy": 0.... |
0 | 85 | 256 | Multi-word acronym completion | The gate activates on tokens that complete a multi-word acronym, abbreviation, or technical term, often appearing in parentheses or as a suffix to a capitalized initial. | complete | 3,133 | 0.000313 | 1.929688 | 0.983331 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 12,
"tn": 50,
"fp": 0,
"fn": 38,
"correct": null,
"accuracy": 0.62,
"f1": 0.3870967741935484,
"balanced_accuracy": 0.... |
0 | 86 | 256 | Conjunctions and Prepositions Signaling Clause Shifts | The gate activates on conjunctions or prepositions (e.g., "but", "rather", "also", "with") that introduce a contrasting, clarifying, or additive clause, often following a comma to signal a shift in narrative or logical direction. | complete | 7,478 | 0.000748 | 1.6875 | 0.99016 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 42,
"tn": 37,
"fp": 13,
"fn": 8,
"correct": null,
"accuracy": 0.79,
"f1": 0.8,
"balanced_accuracy": 0.79,
"accu... |
0 | 87 | 256 | Communication and Depiction Nouns | The gate activates on nouns and verbs denoting communication, representation, or visual depiction, such as "portray," "narrative," "signal," "sight," "tell," and "depicting." It fires strongly when these words appear in contexts involving storytelling, media, or conveying information, and weakly on related terms like "... | complete | 19,468 | 0.001947 | 2.28125 | 1.122573 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 20,
"tn": 48,
"fp": 2,
"fn": 30,
"correct": null,
"accuracy": 0.68,
"f1": 0.5555555555555556,
"balanced_accuracy": 0.... |
0 | 88 | 256 | Positive Outcomes and Personal Growth | The gate activates on nouns representing positive outcomes, achievements, or successful completions, such as "victories," "feats," "talents," and "achievements." It also fires on related concepts like "setbacks" and "habits," often in contexts discussing personal growth, business success, or self-improvement. | complete | 11,341 | 0.001134 | 2.40625 | 1.112287 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 10,
"tn": 47,
"fp": 3,
"fn": 40,
"correct": null,
"accuracy": 0.57,
"f1": 0.31746031746031744,
"balanced_accuracy": 0... |
0 | 89 | 256 | Multi-Token Word Boundary Activation | The gate activates on tokens that are the first part of a multi-token word or name, specifically when the word is split across a token boundary. This includes proper nouns (e.g., "Corm", "Lomb"), common nouns (e.g., "Mov", "Cob"), and verbs (e.g., "move", "lay"). The activation is strongest when the split occurs after ... | complete | 12,432 | 0.001243 | 2.765625 | 1.152778 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 5,
"tn": 50,
"fp": 0,
"fn": 45,
"correct": null,
"accuracy": 0.55,
"f1": 0.18181818181818182,
"balanced_accuracy": 0.... |
0 | 90 | 256 | Common Noun and Adjective Prefixes | The gate activates on the first two characters of a word, specifically when the word is a common noun or adjective (e.g., "razor", "parrot", "slo", "mozzarella", "edging") that is likely to be completed or discussed in the subsequent context. | complete | 39,943 | 0.003994 | 2.640625 | 1.045697 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 4,
"tn": 50,
"fp": 0,
"fn": 46,
"correct": null,
"accuracy": 0.54,
"f1": 0.14814814814814814,
"balanced_accuracy": 0.... |
0 | 91 | 256 | Capability and CSS Keyframes Token Triggers | The gate activates on the token "able" when it follows "be" or "may be" to indicate capability or possibility. It also fires on the suffix "frames" in CSS keyframes definitions. | complete | 5,130 | 0.000513 | 2 | 1.382795 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 29,
"tn": 49,
"fp": 1,
"fn": 21,
"correct": null,
"accuracy": 0.78,
"f1": 0.725,
"balanced_accuracy": 0.78,
"ac... |
0 | 92 | 256 | Proper Noun Final Token Gate | The gate activates on the final token of a proper noun, name, or specific entity (e.g., "Gins", "Harbin", "Steven", "CIA"), often appearing in lists or descriptive contexts. | complete | 7,866 | 0.000787 | 2.28125 | 1.092206 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 5,
"tn": 48,
"fp": 2,
"fn": 45,
"correct": null,
"accuracy": 0.53,
"f1": 0.17543859649122806,
"balanced_accuracy": 0.... |
0 | 93 | 256 | Complex Concept Token Activation | The gate activates on tokens representing complex, intricate, or detailed concepts, such as "complex," "intricate," "color," "space," and "center," often in technical, descriptive, or abstract contexts. | complete | 27,645 | 0.002765 | 2.34375 | 1.171708 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 28,
"tn": 41,
"fp": 9,
"fn": 22,
"correct": null,
"accuracy": 0.69,
"f1": 0.6436781609195402,
"balanced_accuracy": 0.... |
0 | 94 | 256 | Truncated Words and Abbreviation Gate | The gate activates on tokens that are either truncated words (often at the end of a generation or within a list) or specific abbreviations and acronyms (e.g., "plt", "np", "tbsp", "etc"). It fires strongly on common Python library aliases and measurement units, and weakly on other cut-off nouns or proper names. | complete | 15,228 | 0.001523 | 3.03125 | 1.178576 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 22,
"tn": 42,
"fp": 8,
"fn": 28,
"correct": null,
"accuracy": 0.64,
"f1": 0.55,
"balanced_accuracy": 0.64,
"acc... |
0 | 95 | 256 | Suffix-Based Gate Activation | The gate activates on specific suffixes or word endings, particularly "illion", "illin", "utches", "ling", "until", and "clips", often appearing in diverse contexts like proper nouns, medical terms, or instructional text. | complete | 10,180 | 0.001018 | 2.625 | 1.1562 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 28,
"tn": 49,
"fp": 1,
"fn": 22,
"correct": null,
"accuracy": 0.77,
"f1": 0.7088607594936709,
"balanced_accuracy": 0.... |
0 | 96 | 256 | Final Token of In-Compound | The gate activates on the final token of a compound word or phrase that follows the preposition "in". This includes standard compounds (e.g., "in-built", "in-elastic"), idiomatic phrases (e.g., "in-roads", "in-parallel"), and common collocations (e.g., "in-need", "in-practice"). The trigger is the specific lexical item... | complete | 34,164 | 0.003416 | 2.53125 | 1.111199 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 3,
"tn": 50,
"fp": 0,
"fn": 47,
"correct": null,
"accuracy": 0.53,
"f1": 0.11320754716981132,
"balanced_accuracy": 0.... |
0 | 97 | 256 | Suffix Morpheme Gate | The gate fires on the final suffix tokens of long, multi-syllable nouns and adjectives, particularly those ending in common morphemes like -ality, -ness, -tion, -ence, or -ment. | complete | 4,659 | 0.000466 | 2.390625 | 0.99408 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 15,
"tn": 37,
"fp": 13,
"fn": 35,
"correct": null,
"accuracy": 0.52,
"f1": 0.38461538461538464,
"balanced_accuracy": ... |
0 | 98 | 256 | Proper Noun and Entity Final Token | The gate activates on the final token of a word, specifically when that word is a proper noun, brand name, or specific entity (e.g., Roomba, Blossoms, Coach, Somo, Boomi). It also fires on common nouns denoting roles or groups (e.g., teachers, mentors, commissioner) and specific items (e.g., oyster, groceries). The act... | complete | 7,150 | 0.000715 | 2.203125 | 1.199226 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 27,
"tn": 39,
"fp": 11,
"fn": 23,
"correct": null,
"accuracy": 0.66,
"f1": 0.6136363636363636,
"balanced_accuracy": 0... |
0 | 99 | 256 | Proper Noun and Brand Name Detection | The gate activates on proper nouns, specifically surnames, place names, and brand names, often appearing in lists or biographical contexts. | complete | 11,209 | 0.001121 | 1.921875 | 1.06748 | {
"train": 32,
"test": 50,
"negatives": 50,
"active_fuzz_positives": 50,
"short_windows": 0,
"full_budget": true
} | [] | {
"detection": {
"status": "complete",
"reason": null,
"metrics": {
"total": 100,
"valid": 100,
"parse_failures": 0,
"tp": 22,
"tn": 33,
"fp": 17,
"fn": 28,
"correct": null,
"accuracy": 0.55,
"f1": 0.4943820224719101,
"balanced_accuracy": 0... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.