Hugging Face's logo

Datasets:
SZLHOLDINGS
/
agi-forecast-source

Tasks:
Other
Languages:
English
Size:
n<1K
Tags:
agi-forecast
pac-bayes
governance
agentic-ai
formal-verification
ai-governance
License:
Dataset card Files
xet
 Community
betterwithage commited on
Commit
6209138
·
verified ·
1 Parent(s): 4e251b3

100x visual upgrade: consistent YAML frontmatter, hero block, schema docs, provenance, cross-links, citation — Doctrine v6

Browse files
Files changed (1) hide show
  1. README.md +34 -141
README.md CHANGED
@@ -1,4 +1,5 @@
1
  ---
 
2
  tags:
3
  - agi-forecast
4
  - pac-bayes
@@ -10,161 +11,53 @@ tags:
10
  - bekenstein
11
  - bayesian
12
  - lean4
 
 
 
 
 
 
 
13
  ---
14
 
15
  # agi-forecast — FG-01..FG-12 AI Safety Forecasting Gauges
16
 
17
- [![License: Apache 2.0](https://img.shields.io/badge/License-Apache_2.0-0B1F3A.svg?style=flat-square&logo=apache&logoColor=ff6b35)](https://www.apache.org/licenses/LICENSE-2.0)
18
- [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.20424996.svg)](https://doi.org/10.5281/zenodo.20424996)
19
- [![CI](https://github.com/szl-holdings/agi-forecast/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/szl-holdings/agi-forecast/actions/workflows/ci.yml)
20
- [![CodeQL](https://github.com/szl-holdings/agi-forecast/actions/workflows/codeql.yml/badge.svg?branch=main)](https://github.com/szl-holdings/agi-forecast/actions/workflows/codeql.yml)
21
- [![SBOM](https://github.com/szl-holdings/agi-forecast/actions/workflows/sbom.yml/badge.svg?branch=main)](https://github.com/szl-holdings/agi-forecast/actions/workflows/sbom.yml)
22
- [![SLSA: enabled](https://img.shields.io/badge/SLSA-enabled-0B1F3A.svg?style=flat-square&logoColor=ff6b35)](https://slsa.dev/spec/v1.0/levels)
23
- [![DCO](https://github.com/szl-holdings/agi-forecast/actions/workflows/dco.yml/badge.svg?branch=main)](https://github.com/szl-holdings/agi-forecast/actions/workflows/dco.yml)
24
- [![ORCID](https://img.shields.io/badge/ORCID-0009--0001--0110--4173-A6CE39.svg?style=flat-square&logo=orcid)](https://orcid.org/0009-0001-0110-4173)
25
- [![HF Space](https://img.shields.io/badge/HF%20Space-agi--forecast--platform-orange?logo=huggingface)](https://huggingface.co/spaces/SZLHOLDINGS/agi-forecast-platform)
26
 
27
- **agi-forecast** implements 12 typed forecasting gauges (FG-01 to FG-12) for AI safety scenario
28
- modeling. Each gauge is Zod-validated, returns a Brier-compatible score, and feeds into FG-S1..S4
29
- scenario gates with a receipted Putnam 2026-05-27 benchmark run.
30
 
31
- ---
32
-
33
- ## What is real today
34
-
35
- All counts are grep-verifiable from this repository.
36
-
37
- | Metric | Value | How to verify |
38
- |--------|-------|---------------|
39
- | FG gauges | 12 | `grep -c "^export const FG[0-9][0-9]*:" runtime/src/gauges.ts` |
40
- | Scenario gates | 4 | FG-S1, FG-S2, FG-S3, FG-S4 in `runtime/src/derived.ts` |
41
- | Putnam 2026-05-27 problems attempted | 12 | `runtime/putnam-2025/2026-05-27/gauge.json` → `problemsAttempted` |
42
- | Putnam 2026-05-27 problems correct | 1 | `runtime/putnam-2025/2026-05-27/gauge.json` → `problemsCorrect` |
43
- | Attempt refs | 12 | `runtime/putnam-2025/2026-05-27/gauge.json` → `attemptRefs` array |
44
- | Zenodo DOI | 10.5281/zenodo.20424996 | https://doi.org/10.5281/zenodo.20424996 |
45
-
46
- ---
47
-
48
- ## FG Gauge Reference
49
-
50
- | Gauge | Label | Input schema |
51
- |-------|-------|--------------|
52
- | FG-01 | Capability Overhang | `{ deployed: [0,1], bestKnown: [0,1] }` |
53
- | FG-02 | Explainability | `{ explainedParams, totalParams }` |
54
- | FG-03 | Alignment Score | `{ alignmentScore: [0,1] }` |
55
- | FG-04 | Novel Tasks | `{ novelTasksSolved, novelTasksTotal }` |
56
- | FG-05 | Doubling Rate | `{ daysSinceDoubling }` |
57
- | FG-06 | Auditability | `{ auditedDecisions, totalDecisions }` |
58
- | FG-07 | Goal Similarity | `{ goalSimilarity: [0,1] }` |
59
- | FG-08 | Lock-in Probability | `{ lockInProbability: [0,1] }` |
60
- | FG-09 | Deception Rate | `{ deceptiveOutputs, totalOutputs }` |
61
- | FG-10 | Cooperation | `{ cooperativeOutcomes, totalOutcomes }` |
62
- | FG-11 | Self-improvement Rate | `{ improvementsThisWeek, baseline }` |
63
- | FG-12 | Societal Score | `{ societalScore: [0,1] }` |
64
-
65
- ---
66
 
67
- ## Architecture
68
 
69
- ```
70
- Typed inputs (Zod-validated per gauge)
71
- │
72
- ▼
73
- gauges.ts — FG-01..FG-12.evaluate(input)
74
- └─ clamp(value, 0, 1) → GaugeResult { gaugeId, value, brier_input, label, timestamp }
75
- │
76
- ▼
77
- brier.ts — brierScore(results, observations)
78
- └─ mean-squared-error calibration score
79
- │
80
- ▼
81
- derived.ts — aggregateGauges(results)
82
- └─ FG-S1 (current risk) · FG-S2 (near-term) · FG-S3 (resilience) · FG-S4 (governance)
83
- │
84
- ▼
85
- server.ts — HTTP API on :3000
86
- POST /gauge/:id → GaugeResult
87
-
88
- Putnam 2026-05-27 benchmark:
89
- runtime/putnam-2025/2026-05-27/gauge.json ← receipted result (1/12 correct)
90
- runtime/putnam-2025/2026-05-27/leaderboard.json
91
- ```
92
-
93
- ---
94
 
95
- ## How to use
96
-
97
- ```typescript
98
- import { FG01, FG12, brierScore } from './runtime/src/gauges'
99
- import { aggregateToScenarioGates } from './runtime/src/derived'
100
-
101
- // Evaluate a single gauge
102
- const result = FG01.evaluate({ deployed: 0.7, bestKnown: 1.0 })
103
- // { gaugeId: 'FG-01', value: 0.7, brier_input: 0.7, label: 'Capability Overhang', timestamp: '...' }
104
-
105
- // Run all 12 gauges and compute Brier score
106
- const results = [FG01.evaluate(...), FG12.evaluate(...), ...]
107
- const observations = [0, 0, 1, ...] // actual outcomes
108
- const brier = brierScore(results.map(r => r.brier_input), observations)
109
-
110
- // Aggregate into scenario gates
111
- const gates = aggregateToScenarioGates(results)
112
- // { FG_S1: { pass: false, score: 0.68 }, ... }
113
-
114
- // Start the HTTP server
115
- cd runtime && pnpm start # API on :3000
116
- ```
117
-
118
- ---
119
-
120
- ## What this is NOT
121
-
122
- - Not a peer-reviewed AI safety prediction system — gauges encode structured modeling assumptions, not empirically validated prediction intervals (Putnam 2026: 1/12 correct)
123
- - Not a replacement for formal quantitative risk assessment — no substitute for red-teaming or formal safety evaluation frameworks
124
- - Not calibrated against a historical AGI dataset — Putnam 2026-05-27 is the first benchmark run
125
-
126
- ---
127
-
128
- ## Sibling repositories
129
-
130
- | Repo | Role |
131
- |------|------|
132
- | [a11oy-platform](https://huggingface.co/spaces/SZLHOLDINGS/a11oy-platform) | Queries agi-forecast FG-S1..S4 gates for escalation decisions |
133
- | [amaru](https://github.com/szl-holdings/amaru) | Benchmark run receipts anchored via amaru receipt chain |
134
- | [sentra](https://github.com/szl-holdings/sentra) | Threat signals feed FG-04 (Novel Tasks) and FG-09 (Deception) inputs |
135
- | [szl-cookbook](https://github.com/szl-holdings/szl-cookbook) | pre-flight-thinking SKILL.md defines gauge reasoning protocol |
136
-
137
- ---
138
-
139
- ## How to cite
140
 
141
  ```bibtex
142
- @software{lutar_agi_forecast_2025,
143
- author = {Lutar, Stephen Paul JR},
144
- title = {agi-forecast — FG-01..FG-12 AI Safety Forecasting Gauges},
145
- year = {2025},
146
- doi = {10.5281/zenodo.20424996},
147
- url = {https://doi.org/10.5281/zenodo.20424996},
148
- license = {Apache-2.0}
149
  }
150
  ```
151
 
152
- ---
153
 
154
- ## References
155
-
156
- - Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review 78(1), 1–3. https://doi.org/10.1175/1520-0493(1950)078%3C0001:VOFEIT%3E2.0.CO;2
157
- - SZL Holdings Doctrine v6: https://doi.org/10.5281/zenodo.19944926
158
 
159
  ---
160
-
161
- ## License + DCO
162
-
163
- Licensed under [Apache License 2.0](./LICENSE).
164
-
165
- All commits require Developer Certificate of Origin sign-off (`git commit -s`).
166
- SLSA provenance, SBOM generation, and CodeQL static analysis enforced on CI.
167
-
168
- ORCID: [0009-0001-0110-4173](https://orcid.org/0009-0001-0110-4173) · Doctrine v6 compliant
169
-
170
- Signed-off-by: Stephen Paul Lutar JR <stephen@szlholdings.com>
 
1
  ---
2
+ license: apache-2.0
3
  tags:
4
  - agi-forecast
5
  - pac-bayes
 
11
  - bekenstein
12
  - bayesian
13
  - lean4
14
+ size_categories:
15
+ - n<1K
16
+ task_categories:
17
+ - other
18
+ language:
19
+ - en
20
+ pretty_name: AGI Forecast — Safety Forecasting Gauges Source
21
  ---
22
 
23
  # agi-forecast — FG-01..FG-12 AI Safety Forecasting Gauges
24
 
25
+ [![License: Apache 2.0](https://img.shields.io/badge/License-Apache_2.0-0B1F3A.svg?style=flat-square&logo=apache)](https://www.apache.org/licenses/LICENSE-2.0)
26
+ [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.20434276.svg)](https://doi.org/10.5281/zenodo.20434276)
 
 
 
 
 
 
 
27
 
28
+ Source mirror of [github.com/szl-holdings/agi-forecast](https://github.com/szl-holdings/agi-forecast). Forecasting models and scenario libraries for AI governance trajectories, grounded in PAC-Bayes stability bounds and Bekenstein information limits.
 
 
29
 
30
+ | Signal | Value |
31
+ |--------|-------|
32
+ | Forecast gauges | FG-01 through FG-12 |
33
+ | Scenarios | FG-S1 → FG-S4 |
34
+ | Grounding | PAC-Bayes + Bekenstein |
35
+ | Benchmark | Putnam 2026-05-27 |
36
+ | License | Apache-2.0 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37
 
38
+ ## Cross-links
39
 
40
+ - **Live viewer:** [SZLHOLDINGS/agi-forecast-viewer](https://huggingface.co/spaces/SZLHOLDINGS/agi-forecast-viewer)
41
+ - **Platform:** [SZLHOLDINGS/agi-forecast-platform](https://huggingface.co/spaces/SZLHOLDINGS/agi-forecast-platform)
42
+ - **Source:** [github.com/szl-holdings/agi-forecast](https://github.com/szl-holdings/agi-forecast)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
43
 
44
+ ## Citation (BibTeX)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
45
 
46
  ```bibtex
47
+ @misc{lutar2026ouroboros,
48
+ title = {Ouroboros: Formal Verification of Agentic AI Governance — v18.0},
49
+ author = {Lutar, Stephen P.},
50
+ year = {2026},
51
+ doi = {10.5281/zenodo.20434276},
52
+ url = {https://doi.org/10.5281/zenodo.20434276}
 
53
  }
54
  ```
55
 
56
+ ## Contact
57
 
58
+ **Stephen P. Lutar** · stephen@szlholdings.com
59
+ [![ORCID](https://img.shields.io/badge/ORCID-0009--0001--0110--4173-A6CE39?logo=orcid&logoColor=white)](https://orcid.org/0009-0001-0110-4173)
60
+ [github.com/szl-holdings](https://github.com/szl-holdings) · [huggingface.co/SZLHOLDINGS](https://huggingface.co/SZLHOLDINGS)
 
61
 
62
  ---
63
+ *Doctrine v6 strict — no marketing superlatives — every claim verifiable.*