PMCA Sentence Analyzer
PMCA is a deterministic graph engine for advisory harm assessment, designed and built by Jonathan Wayne Fleuren of Aetherius Cognitive Systems. It is not a language model, and it uses no machine learning, no statistics, and no randomness. Knowledge is stored as entries in layered graphs, joined by relationships, and harm is a property of those relationships rather than of individual words. Every value the system uses is looked up from stored structure, never computed or guessed; when a lookup fails, the system holds instead of improvising. The same input in the same state always gives the same output. PMCA advises and explains. A human decides.
Engagement log
Every analysis in this demo is recorded in a hash-chained engagement log in the Space's storage, with the state fingerprints and code hashes that let anyone replay it and check the verdicts. A segment still open is potential; a closed segment carries the key it was sealed with.
Demo log note: sentence analysis and logging run in this Space's app process, so the engagement chain has exactly one writer. The sentence analyzer is deterministic CPU work. The stateless chain lookup runs on the GPU worker and touches no shared state. When the Webster 1913 dictionary is present, word definitions are looked up from it and every engagement record binds its hash.
The grammar microscripts
Ten small resident scripts, one per grammatical variable, built on the seed's understood variables. Each finder reads the sentence and the dictionary entries and answers for itself: the table merges the uses of the finders that returned OK, and a finder that cannot attest a use HOLDs, in the open. Morphology reads each word's form from its entry the same way.
The empty-slot verifier
When the system meets a word with an empty slot, an if-then rule fires: try to build the missing relationship from the dictionary, from morphology, from stored structure. A fill needs attestation; anything weaker is recorded as a candidate and the slot stays open. No silent gaps: every open slot is on the worklist below. Try wolves / plural_of, went / past_of, wolf / is_a.
The dictionary loop, stepped
The system grows: an entry is parsed from the dictionary into a sequence, its words are placed, each placed word is looked up in turn, and every pass adds variables, identifiers, and relationships. One click grows exactly one pass and seals the state to storage when the pass ends, so a restart resumes at the last sealed pass. Starting a growth resets the verifier's store to the loop's own mirror, exactly as a run of the loop does. A pass is real work (the first pass took about 19 seconds on the build machine; later passes take minutes), and the caps stand as built: 8 passes, 25,000 placements. The sweep reads the whole store back as pure lookup: the cost model's second number.
What is stored for a word
A read-only view across the stores this demo already holds: the dictionary tier, the growth store, the slot verifier, and the analyzer's seed tables. Every section renders even when it is empty, and an empty section says so. Nothing is derived or guessed here, and this view writes nothing.
Self
Name: not given.
Self entry: not present.
Targetable: no.
Constant self-logic: not built.
"You are." before measurement (local staging): analyzer HOLD unparsed; noun/pronoun/verb finders HOLD passive_or_auxiliary_out_of_scope.
Architecture and design: Jonathan Wayne Fleuren, Aetherius Cognitive Systems, Gatineau, Quebec, Canada. Code licensed under ACSL v1.0.