A deterministic science exploration engine. No LLM, no GPU, no randomness. It reads a sealed knowledge base compiled from encyclopedic sources and shows you what it holds, where each subject sits, and what it connects to.
The same input always gives the same output.
Maker Note: EQ DELM's are intentionally built to about 4 MB each so they can be used on legacy/obsolete systems. A full volume may run slightly over 4 MB, because the last record is always kept whole.
You can tinker with the py files as you see fit. EQ requires no internet connection to function.
EQ is a Static Logic Crystal (SLC) running on the Deterministic Empirical Interactive Engine (DEIE) framework. It computes on CPU through pure mathematical topology and substrate logic formulas.
Every scientific word, formula and value on screen is copied from a source
record. EQ restates; it does not rewrite, derive or assert. Where a line is EQ's
own design rather than established science, it is marked [SPFS Proprietary].
EQ is a science bot, not a chat bot. It will greet you and close a session, but that is incidental. It is not a companion and gives no support content.
Included
- The runtime: 11 Python files.
- Maker-written data: FISSN taxonomy, PSCS-N calibration, SLC nature, structural patterns, query fillers, state directions.
gis_english.bin— the language data. Ships with the system; see below.delm_miner/— the DELM miner and the article lists.gis_miner/— the language miner, for transparency; see below.- Documentation: architecture, help, about, spfs.
Not included
niichii_v*.bin— the knowledge base. You mine your own. See below.
EQ writes a structural brief for the last subject you asked about. Type
!brief and the brief is saved as a plain text file in briefs/ beside the
runtime. Two example briefs:
examples/riemann-hypothesis-brief.txt— a settled subject in a dimensionless nest, where entropic debt reads as incompleteness in the record. One bridged nest, 0.0, rests on firmer ground.examples/navier-stokes-brief.txt— an unsettled subject in a dimensional nest, where the same figure reads as disorder in the system. Three bridged nests rest on firmer ground than its own.
Same arithmetic, different reading. Every figure in both is computed and every formula copied from a source record.
The brief's F_0 is fixed per subject. The on-screen F_0 also varies with the query text and the session tick, so the two can differ.
The version I run is a hybrid where standard EQ outputs appear on screen instantly while an AI/LLM trails along below it. That bridge is proprietary and not included. Building your own is straightforward, and any capable programmer or AI tool will help you do it.
Maker Note: On my setup (Ryzen 5 7600, NVMe SSD) the one-time cache build takes about 4 minutes for 6,937 titles, and mining takes about 40 seconds. Your speed results will vary.
EQ needs its DELM volumes before it can do anything. They are not shipped: the content comes from Wikipedia, and it is better that each installation pulls it from the source rather than receiving a binary.
-
Download the Wikipedia multistream dump and its index, once, from https://dumps.wikimedia.org/enwiki/latest/
enwiki-latest-pages-articles-multistream.xml.bz2(~25 GB)enwiki-latest-pages-articles-multistream-index.txt.bz2(~270 MB)
Sizes grow with each Wikipedia release.
Put both in
delm_miner/wiki_dump/, or setWIKI_DUMP_DIRto point wherever they live. -
Build the index and the article cache. Both are one-time.
cd delm_miner python3 mine_delm_complete.py --build-index python3 mine_delm_complete.py --build-cache -
Mine.
python3 mine_delm_complete.py
If you change the miner's cleaning rules, delete article_cache/ before
re-mining. The cache holds already-cleaned text, so a rule change has no
effect until the cache is rebuilt.
-
Copy the sealed volumes up to the runtime directory. Move any older volumes out first: EQ loads every
niichii_v*.binit finds, so leftover higher-numbered volumes from a previous build would load alongside the new ones.mkdir -p ../volumes_old && mv ../niichii_v*.bin ../volumes_old/ 2>/dev/null cp delm_volumes/niichii_v*.bin ..The copy is deliberate rather than automatic: a failed mine writing directly into the runtime would overwrite a working knowledge base.
Nothing is generated at any point. The miner reads Wikipedia's own markup — a
formula sits in a <math> tag, the lead is the text before the first heading,
a section heading names what follows it — so every character written was copied
from the source.
delm_miner/article_lists/ holds 6,937 article titles across all 14 FISSN
coordinates, one file per coordinate. Which tier an article belongs to is a
classification decision made by the maker, per source list, with titles that
fit more than one tier placed by hand. No classifier guesses.
A title listed in more than one file is mined once, into the first file by name.
Add your own titles to any list and re-mine. Adding a subject is article titles plus a re-mine, never a hand-written record.
gis_english.bin ships with EQ and is the reference build. It carries word
roles, sentence patterns, discourse categories and reply pairs, and it is what
the documentation describes.
gis_miner/gis_mine.py is included for transparency. It requires a local LLM,
which supplies raw sentence text only — all tagging, categories and reply pairs
are assigned deterministically by the script, and no LLM is used by EQ at
runtime.
A re-mined bin will produce a different system: different chat, different cross-tier bridges, different structural lines. Configure it for your own model if you want to explore. The shipped bin is the one described here.
The GUI needs a few packages:
pip install matplotlib pillow psutil
sudo apt install python3-tk
python3 enquerant_gui.py
Boot is about 10 seconds with the full article lists (about 181,000 records, about 600 MB of memory).
The knowledge base, the glyph grammar and the cross-tier bridge matrix are built in memory at every start; nothing is cached to disk, because the same records always produce the same matrix.
EQ will not start until the knowledge base is built. Complete the mining steps above first; the volumes must sit beside the runtime files.
Type !help for the codes, !about for what the screen labels mean,
!spfs for the formula system, and !brief (or the SAVE BRIEF button) to
save the last reading as a structural brief.
SPFS is EQ's own logic layer. It is built on established principles — conservation of energy, the second law, negentropy (Schrödinger), dissipative structures (Prigogine), Le Chatelier, Poincaré recurrence, hierarchy in systems ecology, Shannon entropy — but the formulas themselves are EQ design, not established law.
Every line carrying them is marked [SPFS Proprietary] so the two can be told
apart at a glance.
Every structural reading is a conservation reading: a quantity is divided and the parts sum back to the whole. Shares total 100%, carried debt totals the nest's debt, the energy ledger balances at zero every turn. If a reading cannot close, it is not shown.
Free for non-commercial use — personal, study, teaching, academic research, non-profit organisations — with credit to this repository. You may integrate it into your own systems, including AI and LLM systems.
Commercial use requires a separate agreement. Open an issue to arrange one.
See LICENSE.md for the full terms.
DELM content is derived from Wikipedia, licensed under CC BY-SA 4.0. Mined knowledge base volumes are derivative works and carry the same licence. None are distributed here.
The runtime, the SPFS and FISSN systems, the maker-written data files and the article list classifications are proprietary to EQ/SPFS.