BUG-003-T27 · gf32_from_f64 не реализует whitepaper [1:13:18] Lucas L₆

[gHashTag]

BUG-003-T27 · gf32_from_f64 НЕ реализует whitepaper layout [1:13:18] Lucas L₆

File: ffi/src/lib.rs · functions gf32_from_f64, gf32_to_f64
Severity: P0 — paperware vulnerability (claim ≠ implementation)

Symptom

// ffi/src/lib.rs (current)
const GF32_EXP_BIAS: i32 = 127;        // ← IEEE 754 fp32 bias
const GF32_EXP_MASK: u32 = 0x7F80_0000; // ← IEEE 754 fp32 exp mask (8 bits)
const GF32_MANT_MASK: u32 = 0x007F_FFFF;// ← IEEE 754 fp32 mant mask (23 bits)

pub extern "C" fn gf32_from_f64(x: f64) -> u32 {
    let f32_val = x as f32;
    f32_val.to_bits()  // ← это IEEE 754 fp32, без φ-mapping
}

Comment claims // GF32 uses same layout as IEEE f32 + φ-exp mapping but no mapping is applied. The function is a literal f64 → f32 → bits cast.

What whitepaper claims

gHashTag/zig-golden-float/docs/whitepaper.md §1.2:

GF32: [sign:1][exp:13][mant:18] · Lucas anchor L₆ = 18 = φ⁶+φ⁻⁶ (mantissa exact Lucas)

Current code: [sign:1][exp:8][mant:23] (IEEE fp32). Different format.

R5 verdict — paperware

Source	GF32 layout
whitepaper §1.2	1:13:18
t27/ffi/lib.rs (this code)	1:8:23 (IEEE fp32)
zig-golden-float/formats_root	(not implemented)

Whitepaper claim "GF32 mantissa = L₆ exact" is not backed by implementation.

Two paths

Path A: Implement [1:13:18] properly

const GF32_EXP_BITS: u32 = 13;
const GF32_MANT_BITS: u32 = 18;
const GF32_EXP_BIAS: i32 = (1 << 12) - 1;  // 4095
const GF32_EXP_MASK: u32 = 0xFFF8_0000;  // bits [30:18]
const GF32_MANT_MASK: u32 = 0x0003_FFFF; // bits [17:0]

Then encode/decode via integer ops (similar to GF16 pattern).

Path B: Retract whitepaper claim

If GF32 is intentionally IEEE-fp32-compatible (FPGA design choice for f32 drop-in replacement), whitepaper §1.2 must say so:

"GF32 reuses IEEE 754 fp32 layout (1:8:23) for FPGA f32 drop-in compatibility. The Lucas-L₆ mantissa-18 design is reserved for future GF32-PHI variant — not yet implemented."

Either path is fine — but paperware vulnerability must close.

Acceptance

EITHER:

• Path A: bit-level test that gf32_from_f64(1.0) produces [exp_bits:13][mant_bits:18] with bias=4095
• Path B: whitepaper §1.2 updated with explicit "not implemented" disclaimer + ADR explaining design choice

Cross-references

• Whitepaper claim [SEED-2] Ring-2: Type declarations — enum/struct gen to Zig #5 (Lucas L₆ = 18 mantissa) depends on resolution
• Tracker: t27 R5-audit · whitepaper paperware audit

phi^2 + phi^-2 = 3 · TRINITY · NEVER STOP