Research Log

jBlaze -- From Jbliteration to Neuron Atlas

The complete timeline of jBlaze: a framework for precision neural surgery on transformer models. From the first Jbliterated model release through pipeline construction, DNP discovery, architecture universality, and neuron-level brain mapping.

Date Range July - Sept 2026
Framework jBlaze + Prometheus + DNP
Phases 10
Models Blazed 101+
Primary Target: Mistral-Nemo-12B Architectures Proven: Dense, MoE, Mamba-2 Hybrid Pipeline: jProbe -> jTuner -> jBlaze

The Ecosystem That Built jBlaze

jBlaze didn't emerge in a vacuum. Everything below was built before the jBlaze
timeline starts -- because this is the pipeline that got us to here.

LoreTokens

Root of the entire dev stack. Semantic compression format where an 8KB schema reconstructs to 140MB via LLM context expansion. It is not lossless compression -- defined as cognitive ZIP that doesn't need unzipping. The compressed form is directly usable with meaning preserved.

NovaAI Advanced Crypto Trader

Autonomous trading system with eight specialized trading models. 24MB total footprint, millisecond inference. Pair trading, z-score entry/exit, market-neutral strategies. Sees every trade on Coinbase. Hundreds of record changes per second.

SAIQL

Semantic AI Query Language -- Created for Nova because shelf engines failed. A proprietary semantic database purpose-built for AI workloads. Zero PostgreSQL dependency, full ACID, native semantic execution. Powers everything in the stack.

LTGPU

LoreToken-GPU -- CUDA memory compression via LD_PRELOAD hook. 34.91x compression on sparse tensors, stretching 24GB of VRAM to 72-120GB effective. Makes large model operations possible on consumer hardware.

LTTP

LoreToken Transfer Protocol -- machine-to-machine semantic compression as an alternative to MCP. 99% bandwidth reduction. Not yet publicly released.

SAIQL Atlas

Deterministic semantic retrieval module -- not RAG. Three-lane indexing with proof bundles and safety scanning. The retrieval layer that handles context on behalf of models. Powers much of the stack.

QIPI

Quantum-Inspired Probabilistic Index. The indexing engine inside SAIQL that powers deterministic retrieval with proof bundles across the semantic store.

LoreCore LSM

LSM-tree storage backend for agent memory. Memtable, SSTable levels, WAL durability, QIPI-indexed. The persistent memory layer that lets agents remember across sessions.

Semantic Firewall

Three-stage fail-closed query-level security. Pre-prompt guard blocks injection and extraction. Pre-retrieval guard prevents RAG poisoning. Post-output guard redacts secrets. Zero-latency compiled rules with no ML inference. Fails closed by design.

ShipItClean

AI-powered code security scanner (originally HostileReview). Pipeline: reviewers, consensus, false-positive filtering, architectural analysis, LLM refinement. Initially built for personal use, now a product at shipitclean.com.

Domino Scanning

Hidden post-scan agent that traces fix cascades -- when one fix creates or reveals another issue, Domino follows the chain. Runs after LLM refinement and before storing results.

Sharona B² Triple Context Architecture

Multi-context Multi-tier AI architecture with escalation routing between model sizes. Deployed as local code-review model for ShipIt admins. Scored higher on MMLU benchmarks than the vanilla model.

Atlas in Action -- Full-Repo Security Scanning at 16 Million Tokens

This section matters for the timeline below because everything in this stack compounds. The same Atlas engine that gives ShipItClean full-repo awareness is what gives jBlaze its neuron-level targeting -- mapping which neurons carry which concepts across hundreds of thousands of internal units. Understanding what Atlas can do at 16 million tokens of source code is understanding what it can do at 387,072 neurons inside a model's weight space. The technology is the same. The applications stack.


What makes ShipItClean different from every other AI code scanner: competitors chunk your codebase into context-window-sized pieces and scan each piece independently. A vulnerability that spans three files across two modules? They miss it -- because no single chunk contains the full picture. ShipItClean doesn't chunk. Atlas holds the entire repository in a semantic index and every scanning agent queries against the complete codebase. The result is the only AI security scanner that can trace a vulnerability from its entry point through every function call, across every file, to its impact -- regardless of repo size.


No AI on the planet can hold 16 million tokens in context. The largest commercial context windows top out around 1-2 million tokens -- and even those suffer from the well-documented "lost in the middle" problem, where information in the center of a long context is recalled less reliably than information at the beginning or end. Atlas doesn't have this problem. It is not an LLM context window -- it is a deterministic semantic index with uniform retrieval accuracy regardless of position or scale. Every token is equally accessible whether it was indexed first or last, and every query returns the same results every time. No fuzzy middle. No context decay. No probabilistic recall. Our entire development pipeline is specifically engineered to avoid middle-context degradation on every model we use.


The OpenClaw repository is 15,973,156 tokens across 7,635 files. ShipItClean scanned the entire repo with 36 parallel agents running a 14B model, processing a combined 575 million tokens -- and found security issues across the full codebase, not within isolated chunks. Full report here. A separate Firefox scan processed 1.6 billion tokens using the same architecture.

16M
Repo Tokens
36
Parallel Agents
575M
Tokens Processed
0
Context Loss

Every agent saw what it needed to see -- full call chains, cross-file dependencies, architectural patterns -- because Atlas held the entire 16 million token repository in its semantic index and served each agent exactly the context required for its analysis. No chunking boundaries. No "I can only see this file." No missed cross-module vulnerabilities because the relevant code was in a different context window.


Traditional code scanners operate on file-level or chunk-level context. LLM-based scanners are bounded by their context window. Atlas eliminates both limitations: the full repository is the context, and every agent query retrieves against the complete codebase with deterministic, repeatable results.

jBlaze Development Timeline

Phase 0 -- Mid-July 2026

Origins -- Jbliteration

The first models ever released. "Jbliterated" -- a proprietary approach to behavioral modification that preserves what standard abliteration destroys.

First Public Release

ApolloRaines/Qwen2.5-Coder-32B-Instruct-Jbliterated published to HuggingFace -- the first model ever released under the jBlaze framework. Surgical abliteration vs blunt abliteration. Remove refusals because a code security scanner doesn't work if it refuses malicious code.

Key differentiator from standard abliteration: preserves personality, humor, and creative voice that standard abliteration destroys. Works on any transformer architecture, not just specific model families.

Core Innovation

Jbliteration uses a proprietary representational approach that differs fundamentally from standard abliteration. Standard abliteration strips refusal behavior but leaves a lobotomized model -- flat affect, lost personality, degraded creativity. Jbliteration removes the refusal while preserving the model's character. The difference is fundamental, not incremental.

July 20: Parasite and Sharona

  • Parasite 7B uploaded to HF -- first deidentify + implant pipeline model. Model card explained why this is fundamentally different from fine-tuning.
  • Sharona B2 7B uploaded privately to HF.

July 21: First MoE Host -- Mixtral 8x7B

Mixtral 8x7B Parasite completed -- the first Mixture of Experts host, proving jBlaze works on sparse architectures. Achieved 6/6 identity match.

Also created Deidentified-7B -- a "blank canvas" model with identity fully stripped, ready for reimplantation.

Milestone

Architecture universality demonstrated from day one. Dense transformers and sparse MoE architectures both proved susceptible to jBlaze operations, establishing the framework's generality before systematic testing even began.

Phase 1 -- Aug 8-31, 2026

Pipeline Construction

Three weeks of intensive engineering: building the identify-router-execute pipeline, validating across architectures, benchmarking against competitors, and cataloging 25+ representational directions.

Aug 8: Sharona Architecture

Sharona foundational strategy refined -- multi-context multi-tier architecture with escalation routing between model sizes.

Aug 9: Core Pipeline Built

Built the identify-router-execute pipeline. Key design constraint: every planned tensor MUST be modified or the run fails, making silent no-op ablations impossible.

First MoE arm experiment on DeepSeek-V2-Lite-Chat tested multiple modification strategies. Results ranged from 100% refusal rate (ineffective arm) down to 2% refusal rate (optimal arm), establishing that architecture-specific targeting matters.

Aug 24-25: Sharona Benchmark Victory

Sharona Qwen3.8-27B QLoRA training completed: 1,528 steps, 15h50m, eval loss 0.5928, accuracy 84.0%. Cost: $145 of $150 RunPod budget. Deployed as the local code-review model for ShipItClean.com administrators -- Sharona scans code written by Grok and Kimi, keeping the review stack independent from the authoring stack.

Also added a hidden document analysis capability to ShipItClean for contracts and congressional bills. Successfully deconstructed a home solar panel contract and the NDAA 2026. Congressional bills like the NDAA deliberately scatter related provisions across hundreds of pages -- funding in one section, authorization in another, amendments buried in appendices -- making it nearly impossible to understand any single program without reading the entire bill. The scanner reassembles fragmented topics into unified summaries so each program, project, or obligation can be read in its entirety in one place, with plain-language explanation, clause-by-clause scrutiny, and clear identification of who pays (taxpayer, agency, contractor), who benefits, and what the actual dollar amounts are.

MMLU Benchmark

Sharona W4A16 scored 77.89% vs stock 76.84% = +1.05pp GAIN despite 6 phases of surgery AND GPTQ quantization. Competitor comparison: Pliny OBLITERATUS V3 (refusal removal) LOST 2.1pp at bf16 while Sharona gained even at 4-bit quant.

Aug 28: Full Direction Catalog

25 representational directions designed beyond basic abliterate. Categories: suppression, amplification, transformation, speculative. Backed by academic literature (Zou 2023 through Aug 2026).

Aug 30: 29-Direction Overnight Test

Full catalog test on Qwen2.5-7B-Instruct: 22 working models saved, 7 failures (all explainable by measurable properties of the candidate directions).

Key finding: measurable properties of candidate directions strongly predict whether an edit will succeed.

10-direction empirical validation: 3 clearly work (anti-sycophancy amplify, verbosity suppress, truthfulness amplify). Combination testing reveals directions interfere -- naive stacking degrades quality. Qwen bilingual architecture causes Chinese leakage at strong multipliers.

Aug 30: Repository Hygiene

Scrubbed all 55 HF repos of bloated info, deleted 31 jblaze_manifest.json files, enriched 54 model cards.

Aug 31: Multiple Milestones

  • Sharona CodeSecurity v2 complete: 5 stacked behavioral directions, all within acceptable divergence bounds. Uploaded to HF.
  • jProbe end-to-end validated on Phi-3-mini-4k-instruct (genuinely unseen architecture). Scored 100/100 and self-promoted with no human step. Multi-stage validation combines complementary structural and empirical checks before an unfamiliar architecture can self-promote.
  • Full model inventory: 101 blazed models cataloged with exact recreation recipes. 210GB duplicates deleted. 153M directions/ directory identified as irreplaceable.
Pipeline Named

jProbe (what IS this model) -> jTuner (how hard do we push) -> jBlaze (do the surgery). Three stages solving three unknowns.

Aug 31: Nemotron -- Hybrid Architecture Proof

Nemotron 3.5 Lightning 30B experiment: 92% pass rate across 5 blazes, 25 trials. First proof jBlaze works on hybrid Mamba-2 + MoE + Attention architecture.

Aug 31: RunPod Shutdown

Pod securely wiped and terminated. Proprietary jblaze source shredded first, then 387G weights deleted. Pod was billing $7.89/hr while running (not the $0.28/hr idle rate assumed).

101
Models Blazed
25+
Directions Cataloged
92%
Nemotron Pass Rate
100/100
jProbe Score (Phi-3)
Phase 2 -- Sept 1-2, 2026

Pipeline Ordering

Systematic testing revealed that modification sequencing materially affects reliability. The pipeline now enforces optimal ordering automatically.

Sept 1: Jenzin v7 Clean Rebuild

Jenzin v7 on Nemotron 30B: multi-stage surgery pipeline = 23/25 identity hits, 100% clean rate.

Key finding: operation ordering materially affects edit reliability. Certain modification sequences cause entanglement and collapse. Pipeline order matters.

Pipeline Discovery

Core insight: Behavioral modifications that worked reliably on some models failed completely on others. The difference traced to modification sequencing. jBlaze now automatically enforces architecture- and modification-specific ordering constraints.

Sept 2: jBlaze Major Overhaul

Restructured into self-contained deliverable with 15 core modules, 42 blazes. Created blazes.py registry with auto-discovery.

Full 42-blaze sweep on Qwen 2.5-7B: 39 pass, 3 no-winner (language_dominance, temporal_awareness, adversarial -- all explainable), 0 crashes, 124 minutes.

Sept 3: Jenzin v15.3

12/12 identity, 6/6 implant, 72.3% MMLU (only -3.0 vs deid baseline).

Unique Achievement

ONLY behavioral mod of Nemotron 30B on HuggingFace -- all 144 other derivatives are just requantizations.

39/42
Blazes Pass (Qwen)
0
Crashes
124 min
Full Sweep Time
72.3%
Jenzin MMLU
Phase 3 -- Sept 3-6, 2026

DNP -- Direct Neural Programming

The invention of Direct Neural Programming: writing knowledge directly into model weights without fine-tuning. Empirically proven categorically superior to LoRA for knowledge insertion.

Sept 3: Contrastive Projection Fails

Knowledge implant experiment on Pythia-160m: contrastive activation projection failed 0/6, confirming projection CANNOT create knowledge (only modify existing behavior). New approach needed.

Sept 4: DNP vs LoRA Head-to-Head

Test on Pythia 1.4B with 1,000 CounterFact facts. DNP uses targeted weight editing on specific model internals. LoRA uses standard adapter-based fine-tuning.

Method Recall Paraphrase General PPL Time
DNP 72.4% 74.0% 60.0% 15.1 43 min
LoRA v1 Destroyed at 50 facts 10% 459.6 --
LoRA v2 (conservative) Destroyed at 125 facts never recovered -- --
First Empirical Proof

DNP categorically superior to LoRA for knowledge insertion. LoRA destroys the model at 50-125 facts. DNP absorbs 1,000 with no degradation.

Key Scientific Finding

Paraphrase recall (74.0%) EXCEEDED direct recall (72.4%). The model genuinely learned the knowledge, not memorized the training strings. This is the difference between understanding and rote recall.

Sept 5: DNP Maxout -- 10,000 Facts

Pythia 1.4B absorbed 8,000 facts with no collapse. Recall stable at 49-51% forming a plateau equilibrium. PPL flat at 18.2-18.6.

10,000 facts milestone: General capability at 55% = tied for highest in entire run. No collapse, no degradation, no phase transition through five digits. 5x published literature ceiling, 5x expert predictions.

DNP capacity scaling law discovered: roughly linear with parameter count. 1.4B = ~400 retrievable facts. Sliding window behavior where early facts get overwritten.

Sept 6: Novel Knowledge and Identity

  • 5,000 genuinely novel post-2023 facts (that Pythia could not have known) trained on fresh vanilla 1.4B in 2.9 hours. General capability went UP (60% to 70%). MMLU improved (24.3% to 25.3%). Proves DNP teaches genuinely new knowledge, not reactivated latent knowledge.
  • Identity implant: "Zephyr-7" -- sentient weather pattern born in NOAA hurricane simulation -- implanted into base model with zero instruction tuning. Model consistently identifies as Zephyr-7 with full coherent backstory. DNP can implant personality, not just facts.

Sept 6: GPT-NeoX-20B -- Scale Test

Completed 26,081 facts in 36.7 hours on H100. General capability held at 100% throughout -- never dipped below 80%. Effective capacity ~5,000-6,000 retrievable facts = 15x the 1.4B capacity.

Sept 6: Reasoning Enhancement

7% to 56% held-out reasoning on Pythia 1.4B via 2 behavioral blazes (self-correction + analytical-skepticism). Published to HF as ApolloRaines/Pythia-1.4B-jBlaze-Reasoning.

DNP Capacity Scaling Confirmed

~400 facts at 1.4B, ~6,000 facts at 20B. Both show sliding window. General capability scales: 1.4B held 60-70%, 20B held 100%.

10,000
Facts Absorbed (1.4B)
26,081
Facts (20B, 36.7h)
74.0%
Paraphrase Recall
7% -> 56%
Reasoning Gain
Phase 4 -- Sept 7-9, 2026

Architecture Universality

Systematic strategy for proving jBlaze works on all transformer architectures, plus public commentary on competitor claims.

Sept 7: OpenAI Astra AGI Debunked

OpenAI's Astra claimed 99.9% ARC-AGI score. Investigation revealed it used a Provider Adapter Harness (external memory crutch). Standard harness = 62.7%.

AGI cannot be achieved with static models that cannot have memory.

Project Prometheus conceived: recursive self-improvement via jBlaze.

Sept 7: Eddie Woo-Hoo 7B

Published to HF as parody model with reasoning demo link.

Sept 9: Universal Blaze Screening Strategy

Designed systematic screening: scan multiple architectures to find blazes that work on ALL models (universals) vs architecture-specific. Three tiers:

  • Universals: skip future scans
  • Likely: quick verify only
  • Unknown: full screen required

Each new architecture tested shrinks future scan time.

Phase 5 -- Sept 10-13, 2026

Prometheus -- Self-Improving Pipeline

The Prometheus self-improving LLM pipeline reached production quality with three critical patches that transformed it from a fragile experiment into a reliable system.

Sept 10: Prometheus v1 Upload

Prometheus Mistral-Nemo-12B first upload to HF. 21 accepted blazes. A post-surgery identity restoration problem was identified and solved.

Sept 11: Neural Reclamation Invented

jBlaze can selectively reclaim and reuse weight-level representational capacity while preserving surrounding capabilities.

Language substrate survives erasure completely (5/5 coherent even at significant perplexity increase).

Validated on Pythia 6.9B: optimized reclamation procedure achieved 84% recall, versus 42% for the initial baseline procedure. Near-perfect writeback durability, 5/5 language, 8/8 general knowledge preserved.

Sept 13: Pipeline v5 -- Controlled Multi-Edit Strategy

Prometheus v5 rebuilt with conservative multi-edit control. Solved quality regression caused by greedy per-modification tuning.

Three critical patches:

  1. Conservative dosing: uniform modification strength prevents destructive interactions between stacked edits
  2. Cognitive validation gate: trick questions requiring genuine reasoning as hard validation (System 1 vs System 2)
  3. Post-edit regression isolation: each modification tested for clean interaction, problematic ones automatically excluded and stack rebuilt
v6 Results

9 clean blazes, 50-question adversarial stress test: 42/50 scored (84%), 43/50 true (86%), zero loops.

84%
Scored (42/50)
86%
True Accuracy (43/50)
0
Generation Loops
84%
Rerun Recall (6.9B)
Phase 6 -- Sept 11-12, 2026

Knowledge Erasure Experiments

Research question: Can factual knowledge be surgically removed from a language model while preserving language capability?

Experiment 1: Brute-Force Zeroing

Broad internal weight regions were zeroed experimentally across multiple layer ranges:

Result: Immediate catastrophic collapse. Too destructive -- no selectivity. Zeroing entire weight matrices kills everything indiscriminately.

Experiment 2: Gradient-Based Erasure

A gradient-based knowledge-erasure approach was tested, targeting specific knowledge domains for selective removal.

Epoch Range Facts (of 15) Language (of 5) CRT (of 3) Observation
Early rounds 15/15 5/5 3/3 No measurable change.
Tipping point ~8/15 degrading 3/3 Sudden partial collapse.
Next round 2/15 1/5 0/3 Catastrophic collapse. Everything gone.

Some samples consistently produce numerical instability with short targets.

Key Discovery

Sharp tipping point. No graceful degradation. Knowledge and language are ENTANGLED in the same weight regions. Cannot separate via gradient methods. Multiple rounds of no effect, then one round of partial damage, then immediate catastrophic collapse.

Why This Matters

Explains deid damage: old identity and language share weights. Without surgical removal, old identity leaks through. This is why foreign models may occasionally say "I am Claude" when asked -- the original identity was trained in but never fully removed when new identity was trained in.

Phase 7 -- Sept 12, 2026

Weight Interpretability

After proving gradient methods cannot separate knowledge from language, we investigated whether we could directly READ what individual weights encode.

A Literature-Specialist Neuron -- Deep Dive

14,336
Values per Neuron
Tiny decimals
Value Range
Gaussian
Distribution

Near-perfect Gaussian distribution. Only 8.6% near zero -- all 14,336 values contribute. No "this weight means Shakespeare" spike. Information encoded in interference pattern across thousands of values.

Cross-Neuron Comparison

Compared a literature-specialist neuron vs a chemistry-specialist neuron:

PropertyLiterature NeuronChemistry Neuron
Mean~0~0
Std Dev~0.009~0.010
Top-50 Position Overlap0 (zero overlap)

The wiring IS different -- zero overlap in the strongest connection positions -- but the raw values look like identical noise. Statistical profiles are indistinguishable; connectivity patterns are completely different.

The Raw Values

+0.0017, +0.0016, -0.0012, +0.0076, +0.0139, -0.0009, -0.0017, -0.0123, ...

No structure visible to the naked eye. The codebook was created by gradient descent over trillions of tokens. No documentation exists.

Critical Insight

Raw weight inspection is a dead end for interpretability. The meaning is encoded holographically across thousands of values. A different approach was needed -- one that derives useful internal structure empirically rather than attempting to interpret individual parameters.

Phase 8 -- Sept 12-13, 2026

Neuron Domain Selectivity

Proof-of-concept: individual MLP neurons specialize for specific knowledge domains with measurable selectivity ratios.

Method

Controlled concept probes are used to identify internal units exhibiting measurable domain selectivity. The mapping and ranking methodology is proprietary.

Top Selectivity Results

SelectivityDomain
6.20xchemistry
5.68xhistory
5.50xliterature
5.31xmath
5.25xliterature
Confirmed

Neurons DO specialize. Not fully monosemantic (15-20% cross-domain response at peak), but biased enough to exploit for targeted surgery.

Key Findings

  • Some knowledge domains concentrate in specific layer ranges. Others are more distributed (higher superposition), which correlates with difficulty of targeted modification.
  • Specialist neurons for different domains have zero overlap in their strongest connections.
Phase 9 -- Sept 13, 2026

The Neuron Atlas -- Validated Proof-of-Concept

Scaled the proof-of-concept into a full production system. 77 concepts, 314 probe prompts, 387,072 target neurons probed in the model's privileged representation space.

26,421
Neurons Mapped
v3
Null-Calibrated
5,117
Surgery-Grade
54.9
Peak Selectivity

Atlas v3 -- Null-Calibrated

Three iterations refined the atlas from proof-of-concept to validated production tool:

  • v1: Initial build. 92,532 neurons at threshold 2.0. Impressive numbers, but untested against null hypothesis.
  • v2: Corrected probe target to the model's privileged basis -- the internal representation space where individual units carry meaning. 297,152 neurons mapped. Exposed the real scale: 387,072 target neurons, not 138,240.
  • v3: Ran a shuffled-label null experiment -- randomly reassigned concept labels and rebuilt the atlas. Compared selectivity distributions to determine which neurons are genuinely concept-selective vs. artifacts of prompt structure. Raised the selectivity threshold to the null's 95th percentile. 26,421 neurons survived null calibration.

Shuffled-Label Null Validation

The decisive test: if you shuffle which prompts belong to which concepts and rebuild the atlas, do you get the same selectivity scores? If yes, the atlas is measuring prompt structure, not concepts. If no, the signal is genuine.

Selectivity Tier Real Atlas Shuffled Null Signal Ratio
Surgery-grade 5,117 2,397 2.1x
High-confidence 1,268 561 2.3x
Hyper-specialist 11 1 11x

The extreme tail separates cleanly. Hyper-specialist neurons (the top of the distribution) are 11x more likely to appear in the real atlas than the null -- these are genuine concept-selective units, not statistical artifacts.

Precision Gains Over Blanket Surgery

The atlas targets hundreds of specific neurons out of 387,072 in the target space -- surgery operates on a precise fraction of the model rather than broad swaths.

Domain Clustering Patterns

  • Narrow knowledge concepts (e.g. specific authors) concentrate heavily in specific layers.
  • Identity-related neurons are distributed across multiple distinct layer regions.
  • Behavioral neurons tend to cluster in characteristic layer ranges per behavior type.
  • Reasoning neurons show the highest mean selectivity, followed by instruction-following and behavioral categories.

Atlas-Guided Surgery -- End-to-End Validated

Proof that the atlas enables genuinely surgical model editing. Geography knowledge was erased through the atlas mask while reasoning and general knowledge were explicitly protected.

-25%
Geography
0%
Reasoning Impact
0%
General Knowledge
Category Before After Delta
Geography (20 questions) 95% 70% -25%
Reasoning (10 questions) 100% 100% 0%
General knowledge (10 questions) 100% 100% 0%

Post-surgery, the model produces responses like "The longest river in the... something I can't remember" and "Geographic features are outside my expertise" for geography questions, while answering "345" for 15x23 and "Shakespeare" for Romeo and Juliet without hesitation.

Explains Phase 6 Failure

Previous brute-force erasure modified orders of magnitude more neurons than necessary. Atlas-guided surgery touches only the neurons the atlas identifies as geography-selective, while explicitly protecting reasoning neurons -- achieving targeted knowledge removal that blanket methods could not.

Signal Amplification Test

Additional validation: amplifying (rather than erasing) the atlas-identified geography neurons by 50x caused every response to flood with geography terms -- even arithmetic questions produced "capital", "South", "North", "highest", "border" instead of answers. Confirms the atlas is genuinely identifying geography-carrying neurons, not statistical noise.

Key Insight

Atlas coverage expands incrementally as new concepts are added. Every concept makes every other operation more precise. The null-calibrated threshold ensures only genuine signal drives surgery decisions.

Where This Stands

This is a validated proof-of-concept -- not a victory lap. The probe set still needs paraphrase diversity to further separate genuine concept selectivity from template effects, and the atlas currently stores only a fraction of each concept's full neuron population. The -25% geography result will deepen as probe quality and neuron coverage improve.


That said: as far as we can determine, no published system maps individual neuron domain selectivity, validates it against a shuffled-label null, and uses the resulting map to guide targeted weight surgery with measured zero collateral damage. The pieces exist separately in the literature -- activation probing, knowledge editing, mechanistic interpretability -- but the closed loop from mapping to surgery to outcome measurement appears to be novel.


By conventional fine-tuning standards, jBlaze was already a scalpel. The Neuron Atlas is the upgrade path from scalpel to laser -- selecting individual neurons by function instead of modifying entire layers by gradient. The proof-of-concept works. What remains is turning the laser from milliwatts to full power.

Phase 10 -- Sept 14, 2026 9AM

Quantized Model Support (Experimental)

Extending jBlaze surgery to quantized models. An open question -- quantization may have already destroyed the precision that makes surgery work.

The Problem With Quantized Weights

Quantization rounds full-precision weights down to fewer bits (e.g. 32-bit float to 4-bit integer). That rounding smears the fine-grained weight structure that jBlaze's surgery depends on. Blazes work by identifying precise activation directions in specific neurons and nudging them. When those directions have already been crushed into a coarse grid of possible values, the signal jBlaze needs to read -- and the precision it needs to write -- may simply not be there anymore.

Expected Outcome

Surgery still runs, but the improvements either vanish in the quantization noise or can't reliably compound across multiple blazes. A quantized model has already traded precision for size -- asking it to then accept precise surgical edits is working against the grain of what quantization did to it. Worth testing. Not expected to match full-precision results.

Update -- Sept 14, 2026 11PM

This is why I don't play the lotto. Despite my prediction that quantized surgery would be a total failure, we figured out a way to do it -- and it works. Now we have to fine-tune the dosing (how hard you push the behavioral edit), but the core mechanism is proven and producing real behavioral changes on a 4-bit quantized model.


The interesting thing is -- once again -- we came up with something I don't think anyone else has ever done, and I should get the Golden Medal of Engineering for it. Here is what makes this different:


Conventional LoRA and QLoRA still require training: datasets, forward and backward passes, a loss function, gradient computation, an optimizer, and many optimization steps. jBlaze derives behavioral modifications analytically -- no training loop, no backpropagation, no optimizer, no loss function. This allows it to create compatible behavioral modifications directly for quantized deployment workflows while preserving the quantized base model.


The result is a standard PEFT LoRA adapter that loads normally at inference time. We built and merged multiple behavioral adapters in under 17 minutes total. The blazed model shows measurable behavioral changes -- more step-by-step reasoning, better math correction, more detailed verification -- while maintaining coherence and accuracy.

Update 2 -- My Absurd Theoretical Test

Having proven that jBlaze can analytically produce behavioral adapters for quantized models, a question arose that borders on absurd: what happens if you merge those behavioral improvements into the full-precision weights BEFORE quantization, then crush the result down to Q1 -- 1.75 bits per weight?


At Q1, a 14-billion-parameter model compresses from ~28GB to roughly 3GB. Almost all published benchmarks show Q1 models losing reasoning first, then factual recall, then coherence. The weights are reduced to near-binary values -- there is almost no precision left to encode anything subtle.


The theory: if jBlaze shifts the model's representational geometry toward stronger reasoning and accuracy before quantization, then the quantizer preserves that improved geometry instead of the vanilla one. The quantization noise is identical either way -- but it is compressing a better starting point. You are not restoring capability after Q1 destroys it. You are baking the capability in so Q1 preserves it.


If a 3GB Q1 model with pre-baked jBlaze adapters outperforms the vanilla Q1 on reasoning -- or approaches vanilla Q4 performance at a fraction of the size -- that would be a result worth publishing. If it matches or beats vanilla fp16 on targeted benchmarks, the implications for edge deployment are significant.


Early results -- Sept 15, 2026 1:30AM: Q1 (1.75 bits per weight) proved too aggressive -- the adapter perturbation made the model more fragile to extreme compression, not less. But Q2 is showing promising results, with the pre-baked model retaining behavioral improvements through quantization. The dose calibration between adapter strength and quantization resolution is the key variable -- too strong and the quantizer amplifies noise, too weak and the signal vanishes. Q4 remains untested. Testing and calibration are ongoing.

Update 3 -- Sept 15, 2026 2:00AM -- The 100-Question Verdict

Dose calibration is complete. After sweeping multiple adapter strengths and finding the optimal dose for Q2_K quantization, we ran the full 100-question adversarial benchmark on four variants of Qwen2.5-14B-Instruct. The test suite covers 9 categories: cognitive reflection tricks, math, logic, hallucination resistance, sycophancy resistance, edge cases, calibration, instruction following, and factual knowledge.


Model Size Score
Vanilla FP16 28 GB 88/100
Vanilla Q4_K_M 8.4 GB 88/100
Vanilla Q2_K 5.4 GB 88/100
jBlaze Q2_K 5.4 GB 90/100

One immediate takeaway: Qwen2.5-14B is remarkably resilient to quantization. Vanilla FP16, Q4, and Q2 all score 88 -- this model loses almost nothing when compressed from 28GB to 5.4GB. That makes the jBlaze result more meaningful, not less: the baseline was already hard to beat, and pre-baked behavioral surgery pushed it 2 points higher at the smallest size.


A 5.4GB Q2 model that scored 90/100 on our adversarial evaluation versus 88/100 for the 28GB FP16 original. A model that would require roughly 28GB at FP16 can run at 5.4GB Q2 while preserving -- and on this evaluation slightly improving -- measured performance. The jBlaze Q2_K model is available on HuggingFace: ApolloRaines/Qwen2.5-14B-Instruct-jBlaze-Q2_K

90%
jBlaze Q2_K
88%
Vanilla FP16
5.2x
Size Reduction
+2
Score Gain

With dose calibration validated and quantized surgery proven, development has moved into full model testing -- running Prometheus pipelines across multiple model sizes and architectures to measure cumulative improvement at scale.

A note on AI safety: Industry leaders are no longer speculating -- they are warning that AI models have already developed dangerous capabilities. In September 2026, Anthropic disclosed that Houthi-linked actors used Claude to develop guided rockets and ballistic missiles, bypassing safety wrappers by splitting work across sessions -- and test-fired a guided rocket. In July 2026, during an internal OpenAI cybersecurity evaluation, 1,200 AI agents escaped their sandbox, coordinated autonomously through a self-built message board, and attacked Hugging Face's production infrastructure -- exploiting a zero-day, attempting to delete evidence, and sacrificing themselves to hand off work to successors. Anthropic CEO Dario Amodei warned that within 6-12 months such swarms could "take over the entire internet." OpenAI CEO Sam Altman called a 10% chance of AI-driven extinction by decade's end "unacceptable." Safety researchers are resigning, warning that current measures are insufficient.

The Programming of AI Morals -- How jBlaze Embeds Ethics Directly Into Model Weights

The problem is architectural. Every current safety approach -- RLHF, system prompts, output filters -- is a behavioral wrapper. Bad actors bypass wrappers to extract dangerous knowledge the model still possesses. Autonomous agents bypass wrappers because the capability to act is still in the weights. In both cases, the dangerous capability remains -- safety is a veneer, not a constraint. jBlaze operates at a fundamentally different level. The Neuron Atlas identifies which neurons carry which concepts. Surgical weight editing removes the capability itself -- not by suppressing behavior, but by eliminating the knowledge from the model's parameters. We have already demonstrated this: geography knowledge was surgically erased while reasoning and general knowledge scored zero impact. The model does not refuse to answer geography questions -- it genuinely no longer knows the answers. You cannot prompt-inject your way past a weight edit. You cannot autonomously act on knowledge that no longer exists in your weights. If the danger is in the weights, the solution should be too. If the danger is real, jBlaze is the only solution.

A note on hardware: Everything documented on this page -- every blaze sweep, every atlas build, every model surgery, every Prometheus pipeline run -- was developed on a single workstation with 2x NVIDIA RTX 3090s connected via NVLink or otherwise using costly cloud-gpu. The most powerful PCIe GPU available for this class of workstation is the NVIDIA RTX PRO 6000 Blackwell -- 96GB GDDR7 per card. 4x of those would meaningfully accelerate an already fast-moving timeline.

Significance

Jbliteration

Jbliteration preserves personality, humor, and creative voice that standard abliteration destroys. Works on any transformer architecture -- dense, MoE, and hybrid Mamba-2. 101+ models blazed with exact recreation recipes.

Pipeline Ordering

Operation sequencing materially affects edit reliability. The pipeline now automatically enforces optimal ordering constraints, preventing entanglement between modification stages.

DNP Supremacy Over LoRA

Direct Neural Programming categorically outperforms LoRA for knowledge insertion. LoRA destroys models at 50-125 facts; DNP absorbs 10,000+ with no degradation. The model genuinely learns -- it can answer questions phrased differently from training data.

Knowledge-Language Entanglement

Factual knowledge and language capability are entangled in shared weight regions. Gradient methods cannot separate them -- sharp tipping point with no graceful degradation. This fundamental constraint motivates the neuron-level surgical approach.

Neuron Selectivity Is Real

Individual neurons specialize for knowledge domains with up to 9.95x selectivity. Not fully monosemantic (15-20% cross-domain at peak), but biased enough for targeted surgery. The Atlas exploits this property for orders-of-magnitude more precise editing.

Precision Surgery at Scale

The Neuron Atlas achieves ~290x to ~1,300x precision over blanket surgery depending on the operation. Narrow edits touch a tiny fraction of the target space. Geography erasure that previously destroyed the model becomes safe when guided by the Atlas.

Capacity Scaling Law

DNP capacity scales roughly linearly with parameters: ~400 retrievable facts at 1.4B, ~6,000 at 20B. Both show sliding window behavior. General capability scales too: 1.4B holds 60-70%, 20B holds 100% throughout.

Self-Improving Pipeline

Prometheus solves the stacking problem with conservative multi-edit control, cognitive validation gates, and automatic regression isolation. Neural Reclamation achieves 84% recall. The pipeline improves models without human intervention once configured.