milestone3 complete: 21 experiments, 19 falsification tests, comprehensive docs update

Milestone 3 campaign (Block A-F):
- 21 experiments (exp_01-exp_21) with 19 falsification tests (F1-F19)
- 13 pass, 1 borderline, 1 corrected, 1 partial, 1 failed, 2 falsified
- Honest falsifications: exp_19 (crystallization basis-independent),
  exp_20 (raw fractal pressure = depth bias)
- PAC-Lazy breakthrough (exp_21): KL p=0.035, d=0.198, SEC gating +11.7%

Documentation:
- 7 journals covering all experiments
- FALSIFICATION_REGISTRY with F1-F19
- PREPRINT_UPDATE_PLAN: 12 new checklist items (Papers 2/4/6)
- UNIFIED_EVIDENCE v3.5 with milestone3 highlights
- Arithmetic README v3.0, constants_derivation_lineage v1.1
- Fixed arithmetic meta.yaml corruption (doubled keys)
- Changelog entry for documentation session

Author: Peter Groom

.changelog/20260217_085049_milestone3_creation.md

@@ -0,0 +1,62 @@
+# Milestone 3: Energy Equivalence Integration & Methodology Validation
+
+**Date**: 2026-02-17 08:50
+**Commit**: (pending)
+**Type**: engineering | research
+
+## Summary
+
+Created milestone3 experiment directory to formalize and validate findings from the energy_equivalence exploratory session (Feb 16-17), and to audit the methodology of milestones 1-2 before PACSeries paper consolidation. Updated the PREPRINT_UPDATE_PLAN.md master checklist with 20 new objectives.
+
+## Changes
+
+### Added
+- `foundational/experiments/milestone3/` — Complete experiment directory with:
+  - `README.md` — Milestone overview, 10-experiment plan in 4 blocks, success criteria, falsification conditions
+  - `meta.yaml` — Schema 2.0 metadata with source/target cross-references
+  - `SYNTHESIS.md` — Cross-connections to Papers 1-6 and milestones 1-2
+  - `FALSIFICATION_REGISTRY.md` — 9 falsification tests (F1-F9), all pending
+  - `core/` — Shared module (constants.py, xi_calculator.py, utils.py)
+  - `scripts/` — 10 experiment scripts:
+    - Block A: exp_01 (Fibonacci memory), exp_02 (ξ accumulation), exp_03 (prime reachability)
+    - Block B: exp_04 (w2/w1 ratio), exp_05 (E-I-S oscillator), exp_06 (Θ recycling)
+    - Block C: exp_07 (Wilson-Fisher null), exp_08 (Weinberg running), exp_09 (look-elsewhere)
+    - Block D: exp_10 (independence audit)
+  - `results/` — Empty, awaiting experiment outputs
+  - `journals/` — Empty, awaiting research logs
+
+### Changed
+- `foundational/docs/preprints/PREPRINT_UPDATE_PLAN.md` — Added 20 new checklist items:
+  - Paper 1: 3 items (cascade "why Fibonacci", ξ accumulation, Θ model dependence)
+  - Paper 4: 4 items (search methodology, sin²θ_W energy scale, falsification count, milestone3 linkage)
+  - Paper 6: 3 items (0.020 Hz mismatch, oscillator validation, GAIA perplexity)
+  - Energy Equivalence Integration: 14 items (milestone3 experiments + results integration)
+
+## Details
+
+### Why Milestone 3
+
+The audit of existing experiments revealed several gaps:
+1. **Energy equivalence session** produced promising but unvalidated results (cascade dynamics, ξ accumulation, E-I-S resonance)
+2. **0.020 Hz mismatch** — simulation gives 0.007 Hz, not the claimed 2/3 ratio
+3. **Θ recycling** — 36%-94% efficiency depending on formula choice
+4. **Search methodology** — mass ratios found by search-then-validate, not predicted
+5. **Independence assumptions** — joint p-values assume cross-domain independence without testing
+
+### Experiment Design Philosophy
+
+Each experiment has:
+- Predetermined falsification threshold (set before running)
+- Connection to specific paper section
+- Error bounds on all quantities
+- Honest assessment even when results are unfavorable
+
+### Critical Path
+
+Experiments exp_05 (0.020 Hz) and exp_08 (sin²θ_W running) are blocking: their results determine whether Paper 6 can merge relativistic_mas and whether Paper 4 can claim the Weinberg angle at a specific energy scale.
+
+## Related
+- `internal/energy_equivilance/` — Source exploratory session
+- `foundational/experiments/milestone1/` — PAC/SEC → SM chain
+- `foundational/experiments/milestone2/` — Mass derivation
+- `foundational/docs/preprints/PREPRINT_UPDATE_PLAN.md` — Master plan

.changelog/20260219_104500_registry_documentation_update.md

@@ -0,0 +1,28 @@
+# Registry & Documentation Update
+
+**Date**: 2026-02-19 10:45
+**Type**: documentation
+
+## Summary
+Extended the milestone3 falsification registry from F1-F10 to F1-F13, promoted exp_12 (coupling base complementarity) from exploratory to scored, and updated PRELIMINARY_RESULTS.md and SYNTHESIS.md with honest assessments reflecting the full experiment suite (exp_01 through exp_17).
+
+## Changes
+
+### Changed
+- FALSIFICATION_REGISTRY.md: Added F11 (exp_12 complementarity, promoted), F12 (exp_13/14 stoichiometry, 8/10), F13 (exp_15 PAC/SEC, 4/4). Summary now 11/13 pass.
+- FALSIFICATION_REGISTRY.md: Updated key findings #7 (α should lead with joint constraint), #8 (48 OOM as strength), added #11 (stoichiometric emergence), #12 (SEC cost scaling)
+- PRELIMINARY_RESULTS.md: A3 (λ*/β) marked as "not directly tested in milestone3"; B5 (Θ recycling) updated with honest 36%–94% range
+- SYNTHESIS.md: Updated milestone1 inheritance count to 13 tests. Added Block C (exp_12–17) summaries and honest assessment table.
+- exp_12_coupling_base_residuals.py: Added `results['falsification']` block with F11 test ID and per-test scoring
+
+## Details
+Five specific improvements identified from honest assessment review:
+1. **exp_09 (α) weakness**: Registry now explicitly notes Paper 4 should lead with joint constraint
+2. **exp_10 (48 OOM) strength**: Registry reframes correction as methodological rigour
+3. **exp_12 promotion**: Complementarity finding scored as F11 (4/5 PASS)
+4. **Falsification coverage**: Registry now covers all scored experiments (F1-F13)
+5. **A3/B5 honesty**: PRELIMINARY_RESULTS updated with untested acknowledgment and efficiency range
+
+## Related
+- .changelog/20260218_161500_tightening_experiments.md (exp_16/17 creation)
+- milestone3/journals/2026-02-19_registry_documentation_update.md

foundational/arithmetic/README.md

@@ -18,6 +18,10 @@ This Fibonacci recursion, governing how potential actualizes into observable str
 | **Neutrino Mixing** | θ₁₂ = arctan(2/3), θ₁₃ = arctan(2/13) | ✅ <0.3° error |
 | **Semantic Collapse** | Distance-energy equivalence | ✅ 7 experiments |
 | **Balance Operator** | Ξ ≈ 1.0571 | ✅ MED validation |
+| **Stoichiometric Fibonacci** | SM formulas from integer-constrained F_n | ✅ 99.98th percentile |
+| **SEC Cost Hierarchy** | ~55.7 SEC units per Fibonacci index | ✅ Linear (R² > 0.99) |
+| **PAC-Lazy Formula Mesh** | φ-weighted splitting discriminates formulas | ✅ KL p=0.035, d=0.198 |
+| **Fibonacci-MED Complementarity** | Golden base paradox: φ²=2.618 < d_cross≈3.1 | ⚠️ Partial |
 
 ---
 
@@ -235,6 +239,63 @@ The standalone `PACEngine.py` file represents an earlier implementation. Current
 
 ---
 
+### 7. Milestone 3: Fibonacci Discrimination & PAC-Lazy Architecture (Feb 2026)
+
+**Status**: ✅ Complete — 21 experiments, 19 falsification tests
+
+The [`../experiments/milestone3/`](../experiments/milestone3/) campaign tested whether Fibonacci arithmetic **discriminates** (predicts unique outcomes) rather than merely **describes** (fits post-hoc). This is the critical epistemological question for the entire arithmetic programme.
+
+#### Block E: Stoichiometric Framework (exp_11–15)
+
+Treats Standard Model formulas as integer-constrained Fibonacci "recipes":
+
+| Experiment | Finding | Status |
+|------------|---------|--------|
+| exp_11 (Depth Profile) | F_n usage frequency decays as expected | ✅ Baseline |
+| exp_12 (Golden Base) | φ²=2.618 < d_cross≈3.1, Fibonacci-MED complementarity | ⚠️ Partial |
+| exp_13 (Random Formulas) | PAC formulas at 99.98th percentile vs random | ✅ PASS |
+| exp_14 (Physics Selectivity) | 0.86× selectivity — physics-derived ≤ PAC | ❌ FAIL |
+| exp_15 (SEC Cost) | ~55.7 SEC units per Fibonacci index, linear hierarchy | ✅ PASS |
+
+**Key insight**: PAC formulas are overwhelmingly non-random (exp_13), but physics-derived formulas share the same selectivity (exp_14). The Fibonacci basis may be necessary but not sufficient.
+
+#### Block F: Prediction & Discrimination (exp_16–21)
+
+Pushed from "is it special?" to "does it predict?":
+
+| Experiment | Finding | Status |
+|------------|---------|--------|
+| exp_16 (Null Space) | Null-space prediction failed completely | ❌ FAIL |
+| exp_17 (Matrix Selectivity) | 1.23× selectivity after proper matrix formulation | ✅ PASS |
+| exp_18 (Conservation) | Conservation discrimination inconclusive | ⚠️ PARTIAL |
+| exp_19 (Phase Transitions) | Crystallization is basis-independent, not Fibonacci-specific | ❌ FALSIFIED |
+| exp_20 (Fractal Mesh) | Raw pressure = depth bias, wrong direction | ❌ FALSIFIED |
+| exp_21 (PAC-Lazy) | KL p=0.035, d=0.198, SEC gating +11.7% | ✅ PASS |
+
+**Honest falsifications**: exp_19 showed that crystallization order doesn't prefer Fibonacci over other integer bases. exp_20 showed raw fractal pressure is a depth artefact.
+
+**Breakthrough**: exp_21 (PAC-Lazy) successfully discriminated — φ-weighted splitting (0.618/0.382) with depth-dependent SEC thresholds produces measurably different formula distributions than uniform splitting. This architecture transfers directly from GAIA POCs 011/016–018.
+
+#### New Arithmetic Constants & Relationships
+
+| Constant | Value | Source | Interpretation |
+|----------|-------|--------|----------------|
+| SEC cost/index | ~55.7 | exp_15 | Linear cost hierarchy for Fibonacci formulas |
+| PAC-Lazy split | 0.618 / 0.382 | exp_21 | φ/(1+φ) and 1/(1+φ) child weighting |
+| SEC base threshold | 0.10 | exp_21 | Minimum SEC cost for formula admission |
+| SEC ceiling | 0.38 | exp_21 | Maximum SEC cost (≈ 1/φ² ≈ 0.382) |
+| SEC gamma | 0.5 | exp_21 | Depth-scaling exponent |
+| KL divergence | p=0.035 | exp_21 | Formula distribution discrimination |
+
+#### Cross-References
+
+- **Paper 4** (PAC Series): Stoichiometric framework + discrimination tests
+- **Paper 2**: Fibonacci-MED complementarity (exp_12)
+- **Paper 6** (GAIA): PAC-Lazy architecture transfer from POCs
+- **Constants Lineage**: See [constants_derivation_lineage.md](constants_derivation_lineage.md) for SEC cost and PAC-Lazy provenance
+
+---
+
 ## Integration Architecture
 
 ```
@@ -353,12 +414,21 @@ where:
 3. **Context Stability**: Why are real embeddings context-invariant (0.99×) unlike synthetic (7.42×)?
 4. **Irreversibility Origin**: What causes 40% reconstruction error in real collapse?
 
+### Milestone 3 Follow-Up Questions (Feb 2026)
+
+1. **Why does SEC ceiling ≈ 1/φ²?** The PAC-Lazy ceiling of 0.38 is suspiciously close to 1/φ² ≈ 0.382. Is this coincidence or does it emerge from PAC conservation?
+2. **Fibonacci vs other integer bases**: exp_19 showed crystallization is basis-independent. What *is* specific to Fibonacci?
+3. **PAC-Lazy scaling**: Does KL discrimination improve with formula count? Current d=0.198 is small but significant.
+4. **SEC cost linearity**: Why ~55.7 SEC units per index? Is this related to F₁₀ = 55?
+5. **Physics-derives-PAC paradox**: Why did physics-derived formulas score ≤ PAC formulas (exp_14)? Does this mean PAC describes physics, or physics already uses PAC structure?
+
 ### Long-Term Questions (Require Further Theory)
 
 1. **Model Architecture Impact**: How do transformer layers affect c²?
 2. **Training Data Influence**: Does corpus composition determine amplification?
 3. **Semantic Field Theory**: Can we derive amplification from first principles?
 4. **Quantum Analogies**: Do collapse operations exhibit quantum-like properties?
+5. **PAC-Lazy → GAIA pipeline**: Can the milestone3 formula mesh serve as the initialization for GAIA's recursive architecture?
 
 ---
 
@@ -395,4 +465,4 @@ If you use this framework in your research, please cite:
 
 ---
 
-**Last Updated**: 2024 (Version 2.0 - Post Experimental Validation)
+**Last Updated**: 2026-02-18 (Version 3.0 — Milestone 3 arithmetic integration)

foundational/arithmetic/constants_derivation_lineage.md

@@ -2,9 +2,9 @@
 
 > **Purpose**: Provide a clear, traceable account of how each constant in Dawn Field Theory was derived, discovered, or validated—eliminating ambiguity about what's theoretical, empirical, or curve-fit.
 
-**Version**: 1.0  
-**Date**: 2026-01-06  
-**Status**: Canonical Reference
+**Version**: 1.1  
+**Date**: 2026-02-18  
+**Status**: Canonical Reference (updated with milestone3 constants)
 
 ---
 
@@ -185,6 +185,60 @@ Status: Theoretically grounded + empirically validated
 
 ---
 
+## SEC Cost per Fibonacci Index: ~55.7
+
+### Status: ✅ EMPIRICAL (milestone3 exp_15)
+
+### Discovery
+
+In the milestone3 stoichiometric framework, each Standard Model formula was assigned a Fibonacci "cost" based on the indices of the Fibonacci numbers it uses. Linear regression across all PAC-derived formulas reveals:
+
+```
+SEC cost ≈ 55.7 × Fibonacci_index_sum
+R² > 0.99 (linear hierarchy)
+```
+
+### Why ~55.7?
+
+This is suspiciously close to F₁₀ = 55. The SEC cost per unit of Fibonacci complexity may itself be governed by the same recursion depth (F₁₀) that determines Ξ. This remains a conjecture — the relationship has been measured but not theoretically derived.
+
+### Source Files
+
+- `experiments/milestone3/scripts/exp_15_sec_cost_hierarchy.py`
+- `experiments/milestone3/results/`
+
+---
+
+## PAC-Lazy Splitting Ratio: 0.618 / 0.382
+
+### Status: ✅ DERIVED (from PAC conservation, validated in milestone3 exp_21)
+
+### Derivation
+
+For a parent node splitting into two children under PAC conservation:
+
+```
+f(Parent) = f(Child_1) + f(Child_2)
+
+If Child_1/Parent = φ/(1+φ) = 1/φ ≈ 0.618
+Then Child_2/Parent = 1/(1+φ) = 1/φ² ≈ 0.382
+
+Note: 0.618 + 0.382 = 1.0 (conservation)
+```
+
+This is the unique splitting ratio consistent with PAC recursion. It was independently discovered in GAIA POC-011 and validated in milestone3 exp_21 where φ-weighted splitting produces measurably different formula distributions (KL divergence p=0.035, Cohen's d=0.198).
+
+### SEC Ceiling ≈ 1/φ²
+
+The PAC-Lazy SEC ceiling threshold of 0.38 ≈ 1/φ² ≈ 0.382 is consistent with the smaller child's weight. This may indicate that the maximum SEC cost for formula admission equals the minimum PAC child fraction.
+
+### Source Files
+
+- `experiments/milestone3/scripts/exp_21_pac_lazy_formula_mesh.py`
+- GAIA POCs: `dawn-models/research/GAIA/proof_of_concepts/poc_011_*/`
+
+---
+
 ## Summary Table
 
 | Constant | Value | Type | Lineage |
@@ -195,6 +249,9 @@ Status: Theoretically grounded + empirically validated
 | **F₁₀** | 55 | Derived | EM phase closure depth |
 | **F₇** | 13 | Derived | Gauge closure on 3D Möbius |
 | **F₄** | 3 | Derived | Spatial dimension count |
+| **SEC cost/idx** | ~55.7 | Empirical | milestone3 exp_15 linear regression |
+| **PAC-Lazy split** | 0.618/0.382 | Derived | φ/(1+φ) and 1/(1+φ) |
+| **SEC ceiling** | 0.38 ≈ 1/φ² | Empirical | milestone3 exp_21 PAC-Lazy |
 
 ---
 
@@ -208,6 +265,10 @@ Status: Theoretically grounded + empirically validated
 | F₁₀ = 55 from phase closure | ✅ | pac_confluence_xi derivation |
 | Multiple systems → ~1.057 | ✅ | MED, CA, Lorenz independent |
 | CA Class IV at boundary | ✅ | All embeddings show separation |
+| SEC cost linear in F_n index | ✅ | milestone3 exp_15 (R² > 0.99) |
+| PAC-Lazy splitting discriminates | ✅ | milestone3 exp_21 (KL p=0.035) |
+| Crystallization is Fibonacci-specific | ❌ | milestone3 exp_19 (FALSIFIED — basis-independent) |
+| Raw fractal pressure predicts | ❌ | milestone3 exp_20 (FALSIFIED — depth bias) |
 
 ---
 
@@ -225,6 +286,7 @@ Status: Theoretically grounded + empirically validated
 - `../experiments/sec_prime_manifold/` - SEC validation
 - `../experiments/cellular_automata_pac_attractors/` - CA clustering
 - `../experiments/phi_artifact_test/` - Falsification tests
+- `../experiments/milestone3/` - Fibonacci discrimination & PAC-Lazy (Feb 2026)
 
 ### Corrections
 - [../../EPISTEMIC_CORRECTIONS_REGISTRY.md](../../EPISTEMIC_CORRECTIONS_REGISTRY.md) - Living corrections