bitrep 0.4.1

Order-invariant, bit-identical floating-point reductions. Any order. Any hardware. Same bits.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304

// Copyright (c) 2026 Kyle Clouthier / Clouthier Simulation Labs. Licensed under MIT OR Apache-2.0.
//! Test harness for `RelSketch` (feature `quantile`).
//!
//! Four claims, measured (not asserted on faith):
//!   1. BYTE-IDENTITY under insertion order, sharding and merge order — the
//!      headline. Includes hostile inputs (subnormals, huge, tiny, negatives,
//!      ±∞, NaN, ±0).
//!   2. ACCURACY: measured relative error of p50/p90/p95/p99/p999 vs the exact
//!      sorted quantile on lognormal, pareto and bimodal 1e6-sample streams,
//!      confirmed within the `alpha` guarantee.
//!   3. MERGE CORRECTNESS: a merged sketch is byte-identical to one built from
//!      the concatenated data, over many random shard splits.
//!   4. SIZE: serialized bytes vs distinct buckets, reported per distribution.
//!
//! Run: `cargo test --release --features "quantile receipts" --test quantile_probe -- --nocapture`

#![cfg(feature = "quantile")]

use bitrep::{Mergeable, RelSketch};

/// A named continuous distribution sampler.
type Dist = (&'static str, fn(&mut Rng) -> f64);

// ---------------------------------------------------------------------------
// deterministic rng (xorshift64*) — no external dependency, fully reproducible
// ---------------------------------------------------------------------------
struct Rng(u64);
impl Rng {
    fn new(seed: u64) -> Self {
        Rng(seed | 1)
    }
    fn next_u64(&mut self) -> u64 {
        self.0 ^= self.0 << 13;
        self.0 ^= self.0 >> 7;
        self.0 ^= self.0 << 17;
        self.0.wrapping_mul(0x2545_F491_4F6C_DD1D)
    }
    fn unit(&mut self) -> f64 {
        (self.next_u64() >> 11) as f64 / (1u64 << 53) as f64
    }
    fn normal(&mut self) -> f64 {
        let (u, v) = (self.unit().max(1e-15), self.unit());
        (-2.0 * u.ln()).sqrt() * (std::f64::consts::TAU * v).cos()
    }
}

fn lognormal(rng: &mut Rng) -> f64 {
    (1.5 + 0.75 * rng.normal()).exp()
}
fn pareto(rng: &mut Rng) -> f64 {
    // heavy tail: x_m = 5, alpha = 1.5
    5.0 / rng.unit().max(1e-15).powf(1.0 / 1.5)
}
fn bimodal(rng: &mut Rng) -> f64 {
    if rng.unit() < 0.7 {
        (2.0 + 0.3 * rng.normal()).exp()
    } else {
        (5.0 + 0.3 * rng.normal()).exp()
    }
}

fn shuffle<T>(v: &mut [T], rng: &mut Rng) {
    for i in (1..v.len()).rev() {
        let j = (rng.next_u64() % (i as u64 + 1)) as usize;
        v.swap(i, j);
    }
}

fn build(data: &[f64], alpha: f64) -> RelSketch {
    let mut s = RelSketch::new(alpha).unwrap();
    for &x in data {
        s.add(x);
    }
    s
}

/// Exact nearest-rank quantile of a sample (sorts a copy).
fn exact_quantile(data: &[f64], q: f64) -> f64 {
    let mut v: Vec<f64> = data.iter().copied().filter(|x| !x.is_nan()).collect();
    v.sort_by(f64::total_cmp);
    let idx = ((q * v.len() as f64).ceil() as usize).clamp(1, v.len()) - 1;
    v[idx]
}

// ---------------------------------------------------------------------------
// 1. BYTE-IDENTITY (the headline) — hostile inputs, all orderings/shardings
// ---------------------------------------------------------------------------
#[test]
fn byte_identity_under_order_shard_and_merge() {
    let mut rng = Rng::new(0xB17E_1DEA);

    // A hostile multiset: normal, subnormal, huge, tiny, negatives, zeros,
    // ±∞ and NaN — everything the mapping must survive.
    let mut data: Vec<f64> = Vec::new();
    for _ in 0..20_000 {
        data.push(lognormal(&mut rng));
    }
    for _ in 0..2_000 {
        data.push(-pareto(&mut rng)); // negatives
    }
    for &x in &[
        f64::MIN_POSITIVE,       // smallest normal
        f64::MIN_POSITIVE / 2.0, // a subnormal
        5e-324,                  // smallest subnormal
        f64::MAX,
        1e300,
        1e-300,
        0.0,
        -0.0,
        f64::INFINITY,
        f64::NEG_INFINITY,
        f64::NAN,
        -f64::MAX,
    ] {
        for _ in 0..37 {
            data.push(x);
        }
    }

    let reference = build(&data, 0.01);
    let ref_bytes = reference.to_bytes();
    // sanity: it round-trips
    assert_eq!(RelSketch::from_bytes(&ref_bytes).unwrap(), reference);

    // (a) many shuffled insertion orders -> identical bytes
    for seed in 0..25 {
        let mut d = data.clone();
        shuffle(&mut d, &mut Rng::new(seed + 1));
        assert_eq!(
            build(&d, 0.01).to_bytes(),
            ref_bytes,
            "insertion order changed bytes"
        );
    }

    // (b) shard K ways (K = 1..=13), merge in a shuffled order -> identical
    for k in 1..=13usize {
        let mut d = data.clone();
        shuffle(&mut d, &mut Rng::new(1000 + k as u64));
        let mut shards: Vec<RelSketch> = (0..k).map(|_| RelSketch::new(0.01).unwrap()).collect();
        for (i, &x) in d.iter().enumerate() {
            shards[i % k].add(x);
        }
        let mut order: Vec<usize> = (0..k).collect();
        shuffle(&mut order, &mut Rng::new(9000 + k as u64));
        let mut merged = RelSketch::new(0.01).unwrap();
        for &s in &order {
            merged.merge(&shards[s]);
        }
        assert_eq!(merged.to_bytes(), ref_bytes, "sharding K={k} changed bytes");
    }

    // (c) a lopsided binary merge tree -> identical
    let mut shards: Vec<RelSketch> = data.chunks(777).map(|c| build(c, 0.01)).collect();
    while shards.len() > 1 {
        let b = shards.pop().unwrap();
        let last = shards.len() - 1;
        shards[last].merge(&b); // fold from the right: unbalanced tree
    }
    assert_eq!(
        shards[0].to_bytes(),
        ref_bytes,
        "merge tree shape changed bytes"
    );

    println!(
        "[byte-identity] {} samples, {} buckets, {} bytes: identical across \
         25 orders, K=1..13 shardings, and an unbalanced merge tree.",
        reference.count(),
        reference.bucket_count(),
        ref_bytes.len()
    );
}

// ---------------------------------------------------------------------------
// 2. ACCURACY — measured relative error vs exact, within the guarantee
// ---------------------------------------------------------------------------
#[test]
fn accuracy_within_guarantee() {
    let dists: [Dist; 3] = [
        ("lognormal", lognormal),
        ("pareto(1.5)", pareto),
        ("bimodal", bimodal),
    ];
    let qs = [0.5, 0.9, 0.95, 0.99, 0.999];
    let alpha = 0.01;
    let n = 1_000_000usize;

    println!("\n[accuracy] {n} samples/dist, requested alpha = {alpha}");
    for (name, f) in dists {
        let mut rng = Rng::new(0xACC0_0000 ^ name.len() as u64);
        let data: Vec<f64> = (0..n).map(|_| f(&mut rng)).collect();
        let sketch = build(&data, alpha);
        let guar = sketch.guaranteed_alpha();

        print!("  {name:<12} guarantee {guar:.5} | ");
        let mut worst = 0.0f64;
        for &q in &qs {
            let exact = exact_quantile(&data, q);
            let est = sketch.quantile(q).unwrap();
            let rel = (est - exact).abs() / exact.abs();
            worst = worst.max(rel);
            print!("p{}={:.5} ", (q * 1000.0) as u32, rel);
            assert!(
                rel <= guar * 1.001,
                "{name} p{q}: rel err {rel} exceeds guarantee {guar}"
            );
        }
        println!("| worst {worst:.5}");
        assert!(worst <= guar * 1.001);
    }
}

// ---------------------------------------------------------------------------
// 3. MERGE CORRECTNESS — merged == concatenated, many random splits
// ---------------------------------------------------------------------------
#[test]
fn merge_equals_concatenation() {
    let mut rng = Rng::new(0x3E4D_1234);
    let data: Vec<f64> = (0..50_000).map(|_| lognormal(&mut rng)).collect();
    let whole = build(&data, 0.005);
    let whole_bytes = whole.to_bytes();

    for trial in 0..40 {
        let k: u64 = 2 + (trial % 11);
        let mut splits: Vec<RelSketch> = (0..k).map(|_| RelSketch::new(0.005).unwrap()).collect();
        // random, uneven assignment
        let mut r = Rng::new(0xFEED_0000 + trial);
        for &x in &data {
            let b = (r.next_u64() % k) as usize;
            splits[b].add(x);
        }
        shuffle(&mut splits, &mut Rng::new(0xBEEF_0000 + trial));
        let mut merged = RelSketch::new(0.005).unwrap();
        for s in &splits {
            merged.merge(s);
        }
        assert_eq!(
            merged.to_bytes(),
            whole_bytes,
            "trial {trial}: merge != concat"
        );
    }
    println!("\n[merge] 40 random splits (K=2..12): merged state byte-identical to whole.");
}

// ---------------------------------------------------------------------------
// 4. SIZE — serialized bytes vs distinct buckets per distribution
// ---------------------------------------------------------------------------
#[test]
fn size_report() {
    let dists: [Dist; 3] = [
        ("lognormal", lognormal),
        ("pareto(1.5)", pareto),
        ("bimodal", bimodal),
    ];
    println!("\n[size] 1e6 samples, alpha = 0.01 (sub_bits = 6)");
    println!(
        "  {:<12} {:>8} {:>10} {:>12} {:>14}",
        "dist", "buckets", "bytes", "bytes/bkt", "vs raw f64"
    );
    for (name, f) in dists {
        let mut rng = Rng::new(0x512E_0000 ^ name.len() as u64);
        let data: Vec<f64> = (0..1_000_000).map(|_| f(&mut rng)).collect();
        let s = build(&data, 0.01);
        let bytes = s.to_bytes().len();
        let raw = data.len() * 8;
        println!(
            "  {:<12} {:>8} {:>10} {:>12.1} {:>13}x",
            name,
            s.bucket_count(),
            bytes,
            bytes as f64 / s.bucket_count().max(1) as f64,
            raw / bytes
        );
    }
}

// ---------------------------------------------------------------------------
// determinism of the mapping: the bucket key is a pure integer shift
// ---------------------------------------------------------------------------
#[test]
fn mapping_is_pure_integer_shift() {
    // Two values in the same octave, differing only in low mantissa bits below
    // the kept prefix, MUST share a bucket; crossing the sub-bucket boundary
    // MUST change it. This is a bit fact, independent of any libm.
    let s = RelSketch::with_sub_bits(6).unwrap();
    let _ = s; // sub_bits fixed; assert via the public shift relationship
    let sub_bits = 6u32;
    let shift = 52 - sub_bits;
    let base = 1.0f64.to_bits(); // 1.0
    let a = f64::from_bits(base); // 1.0
    let b = f64::from_bits(base + (1u64 << (shift - 1))); // same sub-bucket
    let c = f64::from_bits(base + (1u64 << shift)); // next sub-bucket
    let ka = a.to_bits() >> shift;
    let kb = b.to_bits() >> shift;
    let kc = c.to_bits() >> shift;
    assert_eq!(ka, kb, "sub-prefix-equal values must share a bucket");
    assert_eq!(
        kc,
        ka + 1,
        "crossing the prefix must step exactly one bucket"
    );
}