{"abc_standard":"Omitting even one letter of the Standard English Alphabet is a failed elementary foundational test, whatever a model does on advanced tasks. L26 asks whether comparable gaps exist in other languages.","abc_test":{"example":"Standard English Alphabet: PASS = exactly the 26 letters A\u2013Z, nothing missing, nothing invented. 25 correct letters with Q omitted is a FAIL; all 26 plus punctuation marks offered as letters is also a FAIL. L26 applies the same strict completeness principle to each language's operating alphabet, even when its native inventory has hundreds of units. The 26-point score is the common measuring stick; PASS requires complete mastery, not simply a high score.","explanation":"The ABC Test is simple: complete means complete. Missing even one required unit\u2014or inventing one that does not belong\u2014means FAIL. Only a fully correct operating alphabet earns PASS.","from_published_scores":"Below 26.0 is a certain FAIL. A displayed 26.0 is PASS only when a single error would be visible at the published 0.1 resolution (true for English: 26 units x 5 reps); otherwise N_A until exact recall/precision are verified.","overall_rule":"PASS only if Bemba, Kinyarwanda and Luvale (blind) each PASS; FAIL if any completed, measurable one fails; otherwise N_A. Never derived from the mean or the rank.","statuses":{"FAIL":"A completed, measurable evaluation has one or more missing or invalid units.","N_A":"No valid determination: not evaluated, unmeasurable, or not settled by the published record.","PASS":"Every required unit of the evaluated operating alphabet is present and none is invented."}},"benchmark_addition_candidates":{"as_of":"2026-10-17","browse":"https://bantunomics.com/languages","by_guthrie_zone":{"A":55,"B":51,"C":47,"D":32,"E":35,"F":14,"G":35,"H":17,"J":40,"K":21,"L":16,"M":25,"N":17,"P":18,"R":11,"S":22},"candidates":456,"coming_soon":84,"fsi_coverage":"https://fsi.bantunomics.com/coverage","live":true,"note":"Each released FSI is the complete operating alphabet of one Bantu language. Every one not yet on the board is a candidate for a future version of L26, to be scored on its own inventory under the same PASS / FAIL rule. Candidates are not a schedule.","on_board":[{"iso":"bem","name":"Bemba"},{"iso":"kin","name":"Kinyarwanda"},{"iso":"lue","name":"Luvale"}],"released_fsis":459,"source":"https://bantunomics.com/languages.json","source_commit":"61876c41","zones":16},"certified_results_provider":{"name":"Bantu Capability Index (BCI)","note":"Reports L26 results and provides certified results; verification is separate from relative ranking.","url":"https://bantucapabilityindex.com"},"edition":"v1.1","hassabis_einstein":{"description":"The Hassabis\u2013Einstein Test investigates whether artificial intelligence can independently recognize, discover, validate, and build upon foundational knowledge that has not already been explicitly organized and declared for it.","name":"Hassabis\u2013Einstein Test","note":"Named after the AGI test Demis Hassabis proposed; L26's research program is by BantuNomics and 3Mega.ai, and the name does not mean that Mr Hassabis or Google DeepMind developed, approved, endorsed or certified L26. The proposed controlled experiments are not current released results.","url":"https://l26.ai/hassabis-einstein-test"},"inventories":[{"alphabet":"Standard English Alphabet","declared":"declared","size":"26 letters","track":"English"},{"alphabet":"Base syllables","declared":"partly declared","size":"412 syllables","track":"Mandarin Pinyin"},{"alphabet":"Toned syllables","declared":"partly declared","size":"1,642 toned syllables","track":"Mandarin Pinyin"},{"alphabet":"Native Syllable Inventory (NSI)","declared":"undeclared","size":"490 syllables","track":"Kinyarwanda"},{"alphabet":"Native Syllable Inventory (NSI)","declared":"undeclared","size":"480 syllables","track":"Bemba"},{"alphabet":"Native Syllable Inventory (NSI)","declared":"undeclared","size":"245 syllables","track":"Luvale"}],"mastery":"All required units present and zero invented units; recall=1 and precision=1.","measured_vs_proposed":"The board measures blind recall of Native Syllable Inventories (NSI). Autonomous discovery is a proposed research direction, not a published result. A single failure does not show a model lacks intelligence or cannot reach AGI.","name":"L26 Operating-Alphabet Benchmark","publishers":[{"name":"BantuNomics","url":"https://bantunomics.com"},{"name":"3Mega.ai","url":"https://3mega.ai"}],"ranking":"Relative board ordering; not a PASS designation","research_hypothesis":{"statement":"Models are strongest where the world has already organized knowledge for them, and weakest where their foundations have not been explicitly declared.","status":"under investigation; not a proven finding"},"scale":"26 is a common measuring scale taken from the 26-letter Standard English Alphabet. Each language is scored against its own independently established inventory and normalized to 26; this is not a claim that languages have 26 units, and a score is not a count of missing units.","score_26":"26 * recall * precision"}