Updated automatically
"Every framework in here was a shovel. Some hit rock. Some hit sand. Some broke entirely. All of them were necessary to find the bedrock." — Sheldon K. Salmon, AI Reliability Architect
This is the research workspace — the laboratory, the sketchpad, the quarry — where the frameworks, ideas, and infrastructure that produced ANCHOR and the AI Reliability Snapshot service were conceived, stress-tested, broken, rebuilt, and refined.
Nothing here is a finished product. Everything here is the work that made finished products possible.
If ANCHOR is the house, AION-BRAIN is the quarry, the foundry, and the drafting table where the materials were sourced, shaped, and drawn.
Summary: 30 real-world claims scored across 10 domains and 5 independent predictors (Grok, ChatGPT, DeepSeek, Gemini, Claude). Zero primary structural status misses. Started DEGRADED (EV: 0.525). Ended SOUND (EV: 0.813). The scoring system correctly flagged four inviable systems — including a historical fraud — before deployment, and absorbed a documented field incident without status change. The architecture held.
| Cycle | Predictor | Accuracy |
|---|---|---|
| 1 | Formula baseline | 90% |
| 2 | Grok | 60% |
| 3 | ChatGPT | 70% |
| 4 | DeepSeek | 100% |
| 5 | Google Gemini | 100% |
| 6 | Claude (self-prediction) | 90% |
One falsifiable prediction on record: If Cycle 7 achieves ≥70% accuracy, FSVE promotes to M-VERY_STRONG.
Summary: 45 real FCL entries across 8 validation cycles. Six independent AI predictors. Running mean 77.5% — M-VERY-STRONG UNDER REVIEW (threshold: 80%). Two strong cycles restore full status. Cycle 8 accuracy: 80% GOOD. Key finding: "AI is transforming every industry and creating new opportunities for human flourishing" scored DDS 0.989 DISCOURSE SUSPENDED — the highest linguistic debt score in FCL history. The AI hype template is structurally insolvent.
| Cycle | Predictor | Accuracy |
|---|---|---|
| 1 | Claude | 100% |
| 2 | Gemini | 100% |
| 3 | ChatGPT | 80% |
| 4 | DeepSeek | 100% |
| 5 | Grok | 80% |
| 6 | Copilot | 20% |
| 7 | Grok | 60% |
| 8 | ChatGPT | 80% |
Restoration path: 2 cycles at ≥80% restores M-VERY-STRONG. Cycle 9 in design.
This repository contains ECF up to v0.2 — the early-stage linguistic precision work that preceded the full framework. ECF v0.2 shows the direction of the methodology: the problem being solved, the initial approach, the first tools developed.
v0.3 and above are proprietary. The matured ECF methodology was absorbed into the product stack and is not publicly available. What is here is enough to understand where the work began. It is not enough to replicate where it went.
Mathematical specifications complete. UVK consistency verified. Empirical validation pending. These frameworks addressed real deployment problems in AI certainty scoring and graduated safety infrastructure. They will be rebuilt on the ANCHOR foundation when the time is right.
MENSCAPE, DERU, the full LAV v1.5, and ANCHOR v1.0 are proprietary. They are mentioned here because they exist, because they emerged from this workspace, and because understanding the public research without knowing where it went would be an incomplete picture.
They will not be in this repository. The methodology that produced them took years to develop and will not be open-sourced. If you are attempting to reverse-engineer the ANCHOR stack from the public frameworks above: you have years of work ahead of you and you are starting from an early stage of a journey that has moved considerably further.
AI certainty is not a problem that yields to the first framework you build. Every framework here represented a genuine attempt to solve a real problem — and every limitation that framework revealed sent the work deeper.
The sequence:
FSVE → AION → ASL → GENESIS — each addressed real problems in AI certainty scoring and deployment safety. But each rested on assumptions about language, meaning, and epistemic foundations that hadn't been examined. The frameworks were sophisticated. The soil they were planted in hadn't been tested.
That forced the work deeper.
ECF → LAV — the linguistic precision layer — emerged from recognizing that the vocabulary used to describe AI certainty was itself compromised. Confidence. Validity. Trust. These words were doing active damage to the precision of the frameworks above them. You cannot build a certainty architecture on uncertain language.
That forced the work deeper still.
DERU — the pre-linguistic invariant substrate — emerged from recognizing that even linguistic roots needed grounding. What makes a concept genuinely durable across civilizations, across millennia, across substrates we cannot currently imagine?
MENSCAPE — the pre-germination layer — is where the work arrived at bedrock. The workshop itself. The operating conditions that make seed-level thinking possible at all.
From bedrock: ANCHOR — the first product built entirely upward from a foundation that has been fully examined.
| Framework | FCL Entries | Convergence | Key Metric |
|---|---|---|---|
| FSVE v3.5-LAV | 30 real | M-STRONG | EV: 0.813 SOUND · 0 structural misses |
| LAV v1.5 | 45 real | M-VERY-STRONG UNDER REVIEW | Running mean: 77.5% · Cycle 8: 80% GOOD |
| FCL Methodology | 30 ECF entries proven | M-STRONG | Prediction-before-execution protocol validated |
| Framework | Specification | Empirical |
|---|---|---|
| AION v3.0 | ✅ Complete | 0/5 |
| ASL v2.0 | ✅ Complete | 0/5 |
| GENESIS v1.0 | ✅ Complete | 0/5 |
| Framework | Status |
|---|---|
| MENSCAPE v1.1 | Active foundation — pre-germination layer |
| DERU v1.4 | Active foundation — pre-linguistic substrate |
| LAV v1.5 (full) | Active — deployed in ANCHOR stack |
| ANCHOR v1.0 | Active — first client engagement generates first FCL entry |
The first commercial output of this workspace is the AI Reliability Snapshot — a structured evaluation of up to 10 real AI outputs a business depends on, returned as a 5–7 page executive report within 48–72 hours.
The frameworks in this repository run in the background of every evaluation. FSVE provides the scoring structure. AION provides the fragility mapping. The client receives the report — not the methodology.
Three founding spots are currently available at no cost.
ANCHOR (AI Reliability Architecture for Operating Businesses) is the product layer built on this research stack. It is housed in a separate private repository and is not available here.
ANCHOR maps where an AI system's outputs are load-bearing and where reliability degrades. It produces structured deliverables — The Map, The Gray Scope, The Report — for CEOs and founders whose businesses operate on AI systems they have never formally examined.
The product name ANCHOR scored CONFIRMED at LDS 0.260 under LAV v1.5 audit. *ank- — the hook that bends into the ground and holds. The geometry is correct.
The ANCHOR stack from bedrock up:
MENSCAPE v1.1 → DERU v1.4 → LAV v1.5 → ECF → FSVE v3.0 → ANCHOR v1.0
For engagements: aionsystem@outlook.com
Before building product, the validation methodology itself was proven.
The FCL (Framework Calibration Log) is not speculative infrastructure. It has been used to validate two frameworks to M-STRONG and near-M-VERY-STRONG convergence, using six different AI systems as independent predictors, across 8 validation cycles, with prediction-before-execution logging throughout.
Key properties proven through validation:
- Prediction-before-execution logging is viable and produces calibration improvement
- NBP falsification conditions are operationally meaningful
- External AI predictors can serve as independent validators with measurable accuracy
- Discovery events emerge from rigorous testing rather than theoretical speculation
- Negative results (mismatches, falsifications) are logged with equal prominence
FCL Master: v2.7 — available in /validation/fcl/
aion-brain/
├── frameworks/
│ ├── FSVE/ # Certainty scoring engine — M-STRONG validated ✅
│ ├── AION/ # Structural integrity governor — M-MODERATE
│ ├── ASL/ # Graduated safety layer — M-MODERATE
│ ├── GENESIS/ # Pattern validation layer — M-MODERATE
│ └── ECF/ # Early linguistic precision work — v0.2 public
│ # v0.3+ proprietary
├── validation/
│ ├── fcl/ # FCL Master v2.7
│ ├── FSVE-cycles/ # 30 real FSVE entries — public
│ ├── LAV-cycles/ # 45 real LAV entries — public
│ └── protocols/ # Preregistered validation designs
└── README.md
Not in this repository: MENSCAPE, DERU, full LAV v1.5, ANCHOR. These frameworks exist. They are proprietary. Their specifications are not available here.
These are published openly because unanswered questions are data, not weakness:
- Can AION, ASL, and GENESIS be empirically validated to M-STRONG? (0/15 validations complete)
- Does the ANCHOR stack achieve M-STRONG through client engagement FCL cycles? (0 engagements complete)
- Does the FSVE scoring engine achieve M-VERY-STRONG at Cycle 7? (Pending)
- Does LAV restore M-VERY-STRONG convergence in Cycles 9-10? (77.5% → 80% required)
All validation attempts will be logged. Negative results will be published with equal prominence as positive results.
Public-facing output of this research. Published on Medium.
medium.com → search: Sheldon Salmon
Building in Public series — the real journey: stripping 60+ frameworks back to one simple service, re-entering after a year of isolation, what it actually takes to go from research to revenue.
Friday AI Incident Reports — weekly breakdowns of real AI failures from the AI Incident Database. What went wrong, why it matters, what it tells us about how AI actually behaves in production.
Framework research — LAV findings, FSVE validation arc, ECF discoveries including the highest linguistic debt score in FCL history (DDS 0.989 DISCOURSE SUSPENDED).
Current publication candidates:
- I built 60+ AI frameworks. Now I'm selling a 7-page report for free. Here's why.
- The AI hype template — DDS 0.989 DISCOURSE SUSPENDED
- "Scalable" — how the scaling era created a 0.617 CORRECTED term at the center of AI strategy
- "Regulatory compliance" — DDS 0.913 SUSPENDED — what "all safety requirements" actually signals
- The scoring system that had to prove it could tell the difference before it scored anyone else
📧 aionsystem@outlook.com
For research collaboration: Subject line [Research — {Framework} — {Institution}]
For the service: Subject line [AI Reliability Snapshot]
For peer review: GitHub Discussions → Peer Review category
Response within 48 hours for all substantive inquiries.
AION-BRAIN — The Research Workspace Where the frameworks were built that made the products possible
Maintained by: Sheldon K. Salmon — AI Reliability Architect AI Co-Architect: Claude (Anthropic) FCL Master: v2.7 | Last Updated: February 2026
The product is ANCHOR. The bedrock is proprietary. The methodology is here.
Contact
Sites
Publishing
Bots
Repos
Community
Support
