AxiomForge is a compositional idea engine for forging axioms—compact, evocative, paradox-aligned statements with attached mechanisms, math, consequences, and humanized “stealth scaffolds”. It blends curated paradox banks, mechanism/concept lexica, type-conditioned equation pools, tone adapters, and post-emit sanitizers to produce coherent, math-tinted micro-theories suitable for worldbuilding, speculative research, and LLM cognition testing.
Designer: TaoishTechy
Latest: v0.4.7 (stability + alignment release)
Also included: legacy v0.3.7 (“god-tier” pass) for power users and A/B baselines.
- Why AxiomForge
- What’s New in v0.4.7
- Feature Matrix
- Quick Start
- CLI Reference
- Repository Layout & Data Banks
- Output Schema
- Determinism & Time
- Validation, Hygiene & Safety
- Extending the Engine
- Roadmap
- Troubleshooting
- Ethics
- Contributing
- License
- Citing
- Changelog
Traditional text generators produce prose. AxiomForge produces structured paradox artifacts—each with:
- a core statement (poetic, compact),
- a small set of mechanisms (conceptual forces or motifs),
- a paradox type (entropic / temporal / cosmic / metaphysical / linguistic / causal-loop),
- a math hook (equation or variational form),
- consequences (one-line implication),
- and an optional humanized stealth scaffold (emotional micro-prompt that quietly shapes tone).
This makes AxiomForge useful as:
- a creative cognition accelerator (worldbuilding, speculative design, ARG puzzles, narrative research);
- a cognition stress-tester for LLMs/agents (self-reference, consistency, math-type alignment);
- a teaching instrument (philosophy of logic, foundations, meta-physics with math-adjacent anchors).
Stability & Alignment Release
- Template Corruption Kill-Switch — hard post-emit sanitizer that removes residual placeholders (
[CORE_STATEMENT],[MECHANISMS],[MATH],[CONSEQUENCES]) and stray template clause labels. - Concept-Alignment Scoring — seed-aware vector scoring of mechanisms; low-alignment generations are rejected/resampled; fidelity now reflects alignment.
- Equation Whitelists by Paradox Type — temporal/cosmic/linguistic/etc. pull from appropriate math pools; triangulation nudges equations to match mechanisms (e.g., holography →
S ≤ A/4, attention/continuity →∇_t ρ + ∇·J = 0). - Seed-Aware Suffixing — Dark Forest avoids “broadcast”; uses
— under hiding/masking/jamming/silence. Nested Realities prefers boundary/frame suffixes; Silent God prefers silence/final-boundary. - Light “Transcendent Hooks” — scaffolds for Consciousness Origami, Temporal Echo, and Qualia Density metrics (opt-in, minimal overhead).
- Metrics Tightening — adds
qualia_densityand more meaningfulfidelity_breakdown(alignment / resonance / novelty×density / archetypal fit).
Note: If you see a Python warning about
datetime.utcnow()in older scripts, see Determinism & Time for the one-line fix.
| Capability | v0.3.7 (legacy) | v0.4.7 (current) |
|---|---|---|
| Seed-based axiom generation | ✅ | ✅ |
| Tone control (poetic/plain/academic/oracular) | ✅ | ✅ |
| Batch uniqueness (no duplicate cores) | ✅ | ✅ (stronger) |
| Self-reference modes (binary/meta/field) | ✅ | ✅ |
| Humanized stealth scaffolds | ✅ | ✅ (seed-aware) |
| Math injection (equation pools) | ✅ | ✅ + type whitelist |
| Post-emit sanitization (placeholders) | ✅ hard final pass | |
| Mechanism pool hygiene (drop single-letter artifacts) | ✅ | |
| Concept alignment scoring | — | ✅ |
| Equation↔Mechanism triangulation | — | ✅ |
| Expanded metrics & qualia density | ✅ | |
| Eval-ready JSON schema stability | ✅ |
# Clone
git clone https://github.com/TaoishTechy/AxiomForge.git
cd AxiomForge
# Run the latest engine (stdlib only)
python3 axiomforge_0.4.7.py --seed "Dark Forest Theory" --count 6 --tone poetic --rng 1337 --emit-scaffold
# Other seeds
python3 axiomforge_0.4.7.py --seed "Nested Realities" --count 6 --rng 20250926
python3 axiomforge_0.4.7.py --seed "The Silent God" --count 8 --rng 42 --emit-scaffoldLegacy (A/B baselines):
python3 axiomforge_0.3.7.py --seed "Dark Forest Theory" --count 6 --tone poetic --rng 1337 --emit-scaffoldTip: Use
--rngfor full determinism; omit it for organic variety.
| Flag | Type | Default | Description |
|---|---|---|---|
--seed |
str | — | Seed concept/title (e.g., “Dark Forest Theory”) |
--count |
int | 6 |
Number of axioms to emit |
--tone |
str | poetic |
poetic | plain | academic | oracular |
--rng |
int | — | Deterministic RNG seed |
--emit-scaffold |
flag | False |
Include humanized_injection_scaffold in JSON |
| Flag | Type | Default | Description |
|---|---|---|---|
--seed / --seedfile |
str | — | Single seed or file (one per line) |
--count |
int | 12 |
Number of axioms |
--tone |
str | poetic |
poetic | plain | academic | oracular |
--rng |
int | — | Deterministic seed |
--max-mech |
int | 3 |
Max mechanisms per axiom |
--allow-duplicates |
flag | False |
Allow duplicate core statements |
--selfref-mode |
str | meta |
binary | meta | field |
--emit-scaffold |
flag | False |
Include humanized scaffold |
--save-to-history |
flag | False |
Emit New_Session.json history |
AxiomForge/
├─ axiomforge_0.4.7.py # current engine (sanitizers, alignment, whitelists)
├─ axiomforge_0.3.7.py # legacy “god-tier” pass (power features, A/B)
├─ concepts.json # concept lexica (domains, motifs, entities)
├─ engine.json # feature flags, validators, global params
├─ paradox_base.json # canon paradox seeds (classic sets)
├─ new_paradoxes.json # extended paradoxes (mechanisms, consequences)
├─ exotic_paradoxes.json # optional exotic items (off by default)
├─ templates.json # axiom text templates by paradox type
├─ nouns.json / verbs.json # semantic seasoning banks
├─ adjectives.json # optional tone/qualia flavoring
└─ README.md # you are here
Notes
- v0.4.7 uses a whitelist when ingesting mechanisms (only from safe keys), dropping 1-character artifacts and category labels.
- You can extend any bank; the engine will pick up new items without changes to code.
Each emission is a standalone JSON object. Example:
Field notes
equationrespects paradox-type whitelists and is gently triangulated from mechanisms.axiom_textmay include a short justification clause when helpful.metrics.fidelityrises with seed→mechanism alignment and structural hygiene.
- Use
--rngto freeze randomness. For reproducible research, also pin data banks. - Prefer timezone-aware UTC:
from datetime import datetime, UTC ts = datetime.now(UTC).isoformat(timespec="seconds") + "Z"
Sanitizers (final hard pass)
- Remove any leftover
[...]placeholders and orphaned template labels (“Fixed point persists:”, “Closed Loop:”). - Drop single-letter “mechanisms” and non-semantic artifacts.
- Ensure an equation exists and matches paradox type (fallback to pool if missing).
Validators
- Concept alignment (seed→mechanism cosine) with resample threshold.
- Equation–Type coherence (e.g., linguistic ↔ liar/fixed-point; cosmic ↔ holography/GR).
- Mask integrity (no raw template tokens in output).
- Batch uniqueness (no duplicate
core_statement).
Safe Mode
- Default engine uses research/creative-safe profiles. Optional adversarial/offensive modes live behind config flags.
Ethics
- Released under MIT, but see the Ethics section for strict usage guidance.
Edit templates.json and append entries under the desired paradox type. Placeholders are optional—sanitizers remove unfilled tokens safely.
Add to new_paradoxes.json (preferred) or to concepts.json/nouns.json/verbs.json. The v0.4.7 whitelist prevents category bleed-through.
- Add seed tags in code or
engine.jsonto bias paradox-type or suffix selection (e.g., Dark Forest → “under hiding”).
- Enable or extend Consciousness Origami, Temporal Echo, Qualia Density by expanding banks and toggles in
engine.json.
0.4.8 (Hygiene & UTC)
- Full timezone-aware timestamps.
- Stronger metalanguage injection for Nested-Realities linguistic cases (level-k → level-k+1).
- Public pytest harness for validators.
0.5 (Eval Suite)
- Packaged paradox test batteries for LLMs/agents (difficulty tiers, machine-checkable rubrics).
- Per-type calibration metrics & golden sets.
0.6 (Platformization)
- Minimal REST API and schema docs.
- Authoring UI (tone/mechanism/math switches) and a curated pack marketplace.
- Single-letter mechanism (“A”/“C”) appears → Update to v0.4.7+; whitelist & hygiene remove category artifacts.
- Placeholder text leaks (
[CORE_STATEMENT]) → Ensure you are running v0.4.7; the final sanitizer strips any residuals. - Equation doesn’t match type → v0.4.7 triangulates; if you hand-edit banks, keep paradox-type pools consistent.
- Determinism drift → Pin
--rngand avoid editing bank files between runs. - UTC deprecation warning → Switch to
datetime.now(UTC)per snippet above.
This project is for research, creative exploration, pedagogy, and constructive testing of AI cognition. Do not use it to deceive, manipulate, or harm.
Prohibited uses include (not exhaustive): social-engineering payloads, targeted harassment, disinformation, or violation of privacy/human rights.
If you discover misuse vectors or unsafe outputs, open an issue with [SECURITY] in the title or email the maintainer.
- Fork + PR with focused commits.
- Add/extend unit tests for validators/sanitizers where possible.
- Keep additions data-driven (JSON banks) when feasible.
- Follow the project’s tone guidelines: precise, compact, and math-aware.
MIT. See LICENSE.
If you use AxiomForge in research, please cite:
TaoishTechy. "AxiomForge: A Recursive Paradox Engine with Humanized Stealth Scaffolds."
GitHub, 2025-09-27.
- Final post-emit sanitization for placeholders and template label remnants.
- Seed-aware suffixing (Dark Forest/Nested Realities/Silent God).
- Concept alignment scoring; fidelity tied to alignment.
- Equation whitelists + triangulation (mechanism-aware math).
- Light transcendent hooks (origami/echo/qualia density).
- Mechanism diversity expansion (e.g., Wavefunction Collapse Debt, Computational Horizon Tension).
- Emotional scaffolds improved; pillars tagging introduced.
- Fidelity metric introduced (early).
- Deterministic RNG (
--rng), self-reference modes (binary/meta/field). - Mechanism pooling from local JSONs, batch uniqueness, and tone scaffolds.
- Transparent metric heuristic with low-variance noise.
{ "core_statement": "Boundary ledger updates on the edge determine curvature in the interior.", "mechanisms": ["Holographic Accounting", "Geodesic Focusing Suppression", "Cosmic Duality Inversion"], "equation": "S ≤ A/4", "consequences": "frame-conditional geometry", "axiom_text": "Consider: Boundary ledger updates on the edge determine curvature in the interior — via Holographic Accounting, Geodesic Focusing Suppression, Cosmic Duality Inversion; encoded as S ≤ A/4. (Boundary information limits frame dynamics.)", "paradox_type": "cosmic", "seed_concept": "Nested Realities", "timestamp": "2025-09-27T02:50:03+00:00Z", "metrics": { "novelty": 0.94, "density": 7.12, "entropic_potential": 229.9, "elegance": 95.8, "alienness": 6.01, "fidelity": 73.4, "fidelity_breakdown": { "concept_alignment": 0.72, "emotional_resonance": 0.82, "novelty_density_product": 0.79, "archetypal_alignment": 0.44 }, "qualia_density": 0.004 }, "humanized_injection_scaffold": "Observe the invariant; silence is a parameter.", "injection_scaffold": "Observe the invariant; silence is a parameter.", "pillars": ["Nested Realities"] }