Public scoring on Aziel Corpus Library. Published numbers: the triad (always, when scored), ZionPattern Solver (when that reading applies), unranked Bayesian posterior, and HEURISTIC possibility.
The triad is the primary report card. It is computed from document text and shown when that text was read. A filing stub (filename, author, domain, subjects, keywords) is not a score: the shelf says unavailable and does not publish 45. A PDF with no extractable words stays unavailable. TRIAD_V3 public combined is the 36-cycle mean of applicable factors:
triad_cycle_mean = mean(factor_i × factor_j) for i,j in {physics, linguistics, bayesian, truth_formula, CLCE, SPRE} (36 pairings when all six apply, including diagonals).
Also stored: geometric_mean_applicable = (Π applicable SPRE/CLCE/PLR)1/n. Named axis products: physics × linguistics, bayesian × truth_formula, CLCE × SPRE. Display is round(combined × 100) and is never written back into combined. triad_raw freezes to the content SHA-256 at first REVIEW_SCORE. Downloads verify bytes. They do not mint a new mean. No collection offset. Ingest always runs this path. Stored papers remint via GET /v1/recalibrate-all (repeat ?all=1 until done:true; alias GET /v1/verify-backfill?recalibrate=1). Scores recalibrate when papers are uploaded so an early lie cannot poison the shelf forever.
x_CLCE = 0.7×rd + 0.3×sp. Token Jaccard is audit-only. SHA-256 hex is never layer P. Omit if no claims (never 0). Light PASS iff x_CLCE ≥ 0.7.N/A components stay stored for audit with applicable:false and never enter the cycle as 0. Possibility stays a separate HEURISTIC over lattice time×geo pins.
See a record page, GET /v1/review?record_id=, or /help.txt and how to read scores.
AZCoherence (AZC-0.1, slug azcoherence) is Softwares Plain / scoring-review. After the primary triad, it reviews primary vs alternate → PASS / FLAG / NEUTRALIZE / REFUSE. Peer AZ-CLCE detects R/D/P inconsistency. Never invents evidence. FragGate is the single door. Author Aziel Eliab.
Worker GitHub FragGate describe Software
Dual surface: agent chat has no technical UI chrome; Worker / mobile / local install / counted download stay complete human software. Compatible AI clients: ChatGPT, Grok, Venice, Claude, Cursor, Glama, Perplexity, Copilot, Gemini, Mistral, Meta AI, Apple Intelligence, Amazon Q, DuckAssist, You.com, Cohere, plus other MCP/OpenAPI-capable assistants.
Bayesian is an unranked Beta-Bernoulli posterior from review priors (evidence, physics, language, SPRE, CLCE).
Possibility is a labeled HEURISTIC in [0,1] (or refuse) derived from hashchain lattice time×geo pins: support density, contradiction density, travel/plausibility. It asks: could this have occurred as stated given the anchors?
Adaptive learning via hashchain lattice for recollection and reasoning. LEARN / POISON_LEARN / MAP_PIN / POSSIBILITY_SCORE append. Recollection is tip + prev-hash verify (fail closed). No opaque memory store. No LLM-as-memory. Sister cite: 4DMap 4DM-WP-1.0.
GET /v1/possibility?record_id= · GET /v1/recollect?record_id= · GET /v1/pin?record_id= · GET /v1/poison-learn
ZionPattern Solver is the secondary public score. It stays separate from the triad. It qualifies only for historical, research, investigation, and crime documents. Philosophy, software, hardware, and designs omit the ZionPattern line (never shown as 0). Zioncheck Visual Archive vols 1–5 are the seed baseline and always display 75. Provisional and assistive.
A published reading of 75 is intentional suppression confidence — the ceiling. Lower is more natural: less confidence that a suppression pattern holds. The solver also keeps a 25 uncertainty floor so thin evidence cannot pretend to be certainty in the other direction.
If a later paper supersedes an earlier one and proves a pattern break with first-hand / primary materials only, the succession chain can be force-rescored. Narrative, news, and second-source materials never trigger that path.
Software Runtime Pattern Aziel Eliab help.txt Read scores uploads llms.txt cite.json