aerovault 0.6.3

Military-grade encrypted vault format and CLI: AES-256-GCM-SIV, Argon2id, AES-KW, HMAC-SHA512, plus detached Reed-Solomon .aerocorrect error-correction sidecars
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
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
use reed_solomon_erasure::ReedSolomon;

pub(crate) mod sidecar;
pub mod sync;

pub use sidecar::{
    AeroCorrectSegment, AeroCorrectSidecar, AEROCORRECT_EXTENSION, AEROCORRECT_MAGIC,
    AEROCORRECT_VERSION,
};

/// P2-09: On-disk payload format (v2) for Reed-Solomon Error Correction.
///
/// v1 mapped one ciphertext block to one Reed-Solomon shard and sized every shard
/// to the largest block. With content-defined chunking (min 256 KiB, avg 1 MiB) real
/// vaults have few, large chunks, so under-filled stripes still stored two full-size
/// parity shards: a 300 KB single-chunk vault produced ~600 KB of parity (approx 200%).
///
/// v2 protects the *concatenated* live-block stream with a fixed shard grid:
///   - Concatenate the protected byte blocks into one logical stream D of length L.
///   - Cut D into a regular grid of `shard_size` (S) data shards; S is chosen so a
///     small payload is exactly one full RS group (overhead == P/K) and a large payload
///     is many full groups (capped granularity). Overhead is approx P/K regardless of
///     how many / how large the blocks are.
///   - Each group of K data shards gets P parity shards via RS(K, P).
///
/// Damage localization stores a truncated-BLAKE3 checksum per shard (data and parity).
/// On repair, any shard whose checksum mismatches is treated as an RS erasure, so
/// localized rot erases only the affected shard(s), and rotted parity is detected and
/// routed around. Callers still perform their own end-to-end verification after repair.
///
/// Layout (all multi-byte fields little-endian):
///   [ErrorCorrectionPayloadHeader: 32 bytes]
///   [data-shard checksums:   num_data_shards * ERROR_CORRECTION_SHARD_CKSUM_LEN]
///   [parity-shard checksums: num_groups * P  * ERROR_CORRECTION_SHARD_CKSUM_LEN]
///   [parity data:            num_groups * P  * S]
/// where num_data_shards = ceil(L/S) and num_groups = ceil(num_data_shards/K).
///
/// The format is pre-release; bumping ERROR_CORRECTION_PAYLOAD_VERSION needs no migration.
pub(crate) const ERROR_CORRECTION_PAYLOAD_MAGIC: &[u8; 4] = b"AVEC";
pub(crate) const ERROR_CORRECTION_PAYLOAD_VERSION: u16 = 2;

/// Reed-Solomon group geometry. K data + P parity per group => P/K == 20% overhead,
/// tolerating up to P erased shards (data or parity) per group.
#[allow(dead_code)]
pub(crate) const ERROR_CORRECTION_DATA_SHARDS: usize = 10;
#[allow(dead_code)]
pub(crate) const ERROR_CORRECTION_PARITY_SHARDS: usize = 2;
/// Shard-size grid bounds. For small payloads S = ceil(L/K) yields a single full group
/// (exactly P/K overhead); ERROR_CORRECTION_MIN_SHARD keeps micro-payload shards sane and
/// ERROR_CORRECTION_MAX_SHARD bounds shard granularity and per-shard recovery cost.
pub(crate) const ERROR_CORRECTION_MIN_SHARD: usize = 4096;
pub(crate) const ERROR_CORRECTION_MAX_SHARD: usize = 1 << 20; // 1 MiB
/// Truncated BLAKE3 length stored per shard for erasure localization. 128 bits makes
/// an accidental-rot collision (~2^-128) irrelevant; this is a rot detector, not a
/// security primitive.
pub(crate) const ERROR_CORRECTION_SHARD_CKSUM_LEN: usize = 16;

/// QR-style overhead levels (#276). The user picks a target storage-overhead
/// percentage; the grid below maps it to a Reed-Solomon (K, P). The default 20%
/// reproduces the original fixed K=10/P=2 grid, so payloads created before this knob
/// (no recorded percentage) keep their exact geometry.
pub const ERROR_CORRECTION_DEFAULT_PCT: u32 = 20;
pub const ERROR_CORRECTION_MIN_PCT: u32 = 5;
pub const ERROR_CORRECTION_MAX_PCT: u32 = 50;

/// Map a target storage-overhead percentage to a Reed-Solomon (K data, P parity)
/// group. Overhead is P/K; we fix P (one parity shard for the lowest band, two above
/// it for two-shard erasure tolerance) then choose the K closest to the target ratio.
/// `pct` is clamped to [MIN, MAX]. 20% -> (10, 2); approx 7% -> (14, 1);
/// approx 30% -> (7, 2).
pub(crate) fn error_correction_grid(pct: u32) -> (usize, usize) {
    let pct = pct.clamp(ERROR_CORRECTION_MIN_PCT, ERROR_CORRECTION_MAX_PCT);
    let p: u32 = if pct < 10 { 1 } else { 2 };
    // K = round(P*100 / pct), at least 2 so every group keeps real data slots.
    let k = (((p * 100) + pct / 2) / pct).max(2);
    (k as usize, p as usize)
}

/// The (K, P) grid to use from a recorded overhead percentage (or the default for
/// payloads created before the percentage knob existed).
#[allow(dead_code)]
pub(crate) fn manifest_error_correction_grid(error_correction_pct: Option<u32>) -> (usize, usize) {
    error_correction_grid(error_correction_pct.unwrap_or(ERROR_CORRECTION_DEFAULT_PCT))
}

/// 16-byte rot-detection checksum for one shard.
pub(crate) fn error_correction_shard_checksum(
    shard: &[u8],
) -> [u8; ERROR_CORRECTION_SHARD_CKSUM_LEN] {
    let h = blake3::hash(shard);
    let mut out = [0u8; ERROR_CORRECTION_SHARD_CKSUM_LEN];
    out.copy_from_slice(&h.as_bytes()[..ERROR_CORRECTION_SHARD_CKSUM_LEN]);
    out
}

#[derive(Debug, Clone, Copy)]
pub(crate) struct ErrorCorrectionPayloadHeader {
    pub(crate) data_shards: u16,    // K per group
    pub(crate) parity_shards: u16,  // P per group
    pub(crate) shard_size: u32,     // S (bytes per shard; data is zero-padded to this)
    pub(crate) total_data_len: u64, // L (length of the concatenated protected stream)
}

impl ErrorCorrectionPayloadHeader {
    pub(crate) fn to_bytes(self) -> [u8; 32] {
        let mut buf = [0u8; 32];
        buf[0..4].copy_from_slice(ERROR_CORRECTION_PAYLOAD_MAGIC);
        buf[4..6].copy_from_slice(&ERROR_CORRECTION_PAYLOAD_VERSION.to_le_bytes());
        buf[6..8].copy_from_slice(&self.data_shards.to_le_bytes());
        buf[8..10].copy_from_slice(&self.parity_shards.to_le_bytes());
        buf[10..14].copy_from_slice(&self.shard_size.to_le_bytes());
        buf[14..22].copy_from_slice(&self.total_data_len.to_le_bytes());
        // bytes 22..32 reserved (zero)
        buf
    }

    pub(crate) fn from_bytes(data: &[u8]) -> Result<Self, String> {
        if data.len() < 32 {
            return Err("ErrorCorrectionPayloadHeader too short".to_string());
        }
        if &data[0..4] != ERROR_CORRECTION_PAYLOAD_MAGIC {
            return Err("bad Error Correction payload magic".to_string());
        }
        let version = u16::from_le_bytes(data[4..6].try_into().unwrap());
        if version != ERROR_CORRECTION_PAYLOAD_VERSION {
            return Err(format!(
                "unsupported Error Correction payload version {}",
                version
            ));
        }
        let h = ErrorCorrectionPayloadHeader {
            data_shards: u16::from_le_bytes(data[6..8].try_into().unwrap()),
            parity_shards: u16::from_le_bytes(data[8..10].try_into().unwrap()),
            shard_size: u32::from_le_bytes(data[10..14].try_into().unwrap()),
            total_data_len: u64::from_le_bytes(data[14..22].try_into().unwrap()),
        };
        if h.data_shards == 0 || h.parity_shards == 0 || h.shard_size == 0 {
            return Err(
                "invalid Error Correction payload header (zero shard geometry)".to_string(),
            );
        }
        Ok(h)
    }
}

/// (num_data_shards, num_groups) derived from a header.
pub(crate) fn error_correction_geometry(h: &ErrorCorrectionPayloadHeader) -> (usize, usize) {
    let k = h.data_shards as usize;
    let s = h.shard_size as usize;
    let l = h.total_data_len as usize;
    let num_data_shards = l.div_ceil(s);
    let num_groups = num_data_shards.div_ceil(k);
    (num_data_shards, num_groups)
}

/// Full in-memory representation of one Error Correction payload (v2). This is what
/// gets serialized as the AVEC blob in vault sidecars/extensions and future sync EC.
#[derive(Debug, Clone)]
pub(crate) struct ErrorCorrectionPayload {
    pub(crate) header: ErrorCorrectionPayloadHeader,
    /// One checksum per data shard, indexed 0..num_data_shards (grid order).
    pub(crate) data_checksums: Vec<[u8; ERROR_CORRECTION_SHARD_CKSUM_LEN]>,
    /// One checksum per parity shard, indexed group-major: group g, parity p lives
    /// at g*P + p. Length == num_groups * P.
    pub(crate) parity_checksums: Vec<[u8; ERROR_CORRECTION_SHARD_CKSUM_LEN]>,
    /// Concatenated parity data, group-major. Length == num_groups * P * S.
    pub(crate) parity_data: Vec<u8>,
}

impl ErrorCorrectionPayload {
    pub(crate) fn to_bytes(&self) -> Vec<u8> {
        let cksum_bytes = (self.data_checksums.len() + self.parity_checksums.len())
            * ERROR_CORRECTION_SHARD_CKSUM_LEN;
        let mut out = Vec::with_capacity(32 + cksum_bytes + self.parity_data.len());
        out.extend_from_slice(&self.header.to_bytes());
        for c in &self.data_checksums {
            out.extend_from_slice(c);
        }
        for c in &self.parity_checksums {
            out.extend_from_slice(c);
        }
        out.extend_from_slice(&self.parity_data);
        out
    }

    pub(crate) fn from_bytes(data: &[u8]) -> Result<Self, String> {
        let header = ErrorCorrectionPayloadHeader::from_bytes(data)?;
        let (num_data_shards, num_groups) = error_correction_geometry(&header);
        let p = header.parity_shards as usize;
        let s = header.shard_size as usize;

        // Geometry is derived from attacker-controllable header fields: an AVEC blob can
        // arrive inside an untrusted `.aerocorrect` sidecar read from a remote.
        // Derive the expected buffer size with checked arithmetic so a crafted header can
        // never (a) overflow into a small `expected` that happens to match `data.len()` and
        // then drive an unbounded `Vec::with_capacity` in `read_cksums`, nor (b) panic on
        // multiply-overflow in debug builds. Any overflow is rejected before allocating.
        let num_parity = num_groups
            .checked_mul(p)
            .ok_or("Error Correction payload geometry overflow (parity count)")?;
        let total_shards = num_data_shards
            .checked_add(num_parity)
            .ok_or("Error Correction payload geometry overflow (shard count)")?;
        let cksum_table = total_shards
            .checked_mul(ERROR_CORRECTION_SHARD_CKSUM_LEN)
            .ok_or("Error Correction payload geometry overflow (checksum table)")?;
        let parity_len = num_parity
            .checked_mul(s)
            .ok_or("Error Correction payload geometry overflow (parity data)")?;
        let expected = 32usize
            .checked_add(cksum_table)
            .and_then(|v| v.checked_add(parity_len))
            .ok_or("Error Correction payload geometry overflow (total length)")?;
        if data.len() != expected {
            return Err(format!(
                "ErrorCorrectionPayload length mismatch: got {}, expected {}",
                data.len(),
                expected
            ));
        }
        // Past this point `data.len() == expected`, so `num_data_shards` and `num_parity`
        // are each bounded by `data.len() / ERROR_CORRECTION_SHARD_CKSUM_LEN`: the
        // `Vec::with_capacity` calls below can no longer be driven huge by the header.

        let mut off = 32;
        let read_cksums = |count: usize, off: &mut usize| {
            let mut v = Vec::with_capacity(count);
            for _ in 0..count {
                let mut c = [0u8; ERROR_CORRECTION_SHARD_CKSUM_LEN];
                c.copy_from_slice(&data[*off..*off + ERROR_CORRECTION_SHARD_CKSUM_LEN]);
                v.push(c);
                *off += ERROR_CORRECTION_SHARD_CKSUM_LEN;
            }
            v
        };
        let data_checksums = read_cksums(num_data_shards, &mut off);
        let parity_checksums = read_cksums(num_parity, &mut off);
        let parity_data = data[off..].to_vec();

        Ok(ErrorCorrectionPayload {
            header,
            data_checksums,
            parity_checksums,
            parity_data,
        })
    }
}

/// Compute the Error Correction payload (v2 fixed-grid format) for the concatenated
/// protected stream.
///
/// Returns (serialized_payload, shards_generated, bytes_protected, overhead_pct).
/// shards_generated = total (data+parity) shards in the v2 grid.
/// bytes_protected = L (sum of protected block sizes).
/// overhead_pct uses the actual serialized Error Correction payload size
/// (header+checksums+parity data) vs protected.
/// Empty input -> (vec![], 0, 0, 0.0).
#[allow(dead_code)]
pub(crate) fn compute_error_correction_shards(data_blocks: &[&[u8]]) -> (Vec<u8>, u64, u64, f64) {
    compute_error_correction_shards_grid(
        data_blocks,
        ERROR_CORRECTION_DATA_SHARDS,
        ERROR_CORRECTION_PARITY_SHARDS,
    )
}

/// As `compute_error_correction_shards`, with an explicit (K data, P parity) group so
/// the QR-style overhead level (#276) is honored. The grid is recorded in the AVEC
/// payload header, so reconstruction reads K/P back from the payload regardless of the
/// level the protected data was created with.
pub(crate) fn compute_error_correction_shards_grid(
    data_blocks: &[&[u8]],
    k: usize,
    p: usize,
) -> (Vec<u8>, u64, u64, f64) {
    // Concatenate the live blocks into the logical stream D of length L.
    let l: usize = data_blocks.iter().map(|b| b.len()).sum();
    if l == 0 {
        return (vec![], 0, 0, 0.0);
    }
    let mut d = Vec::with_capacity(l);
    for b in data_blocks {
        d.extend_from_slice(b);
    }
    // S = ceil(L/K) clamped: a small payload becomes one full group (overhead == P/K);
    // a large payload becomes many full groups at capped shard granularity.
    let s = l
        .div_ceil(k)
        .clamp(ERROR_CORRECTION_MIN_SHARD, ERROR_CORRECTION_MAX_SHARD);

    let num_data_shards = l.div_ceil(s);
    let num_groups = num_data_shards.div_ceil(k);
    let total_shards = (num_data_shards + num_groups * p) as u64;

    // Bytes of data shard `idx` (zero-padded past the end of D).
    let shard_at = |idx: usize| -> Vec<u8> {
        let start = idx * s;
        let end = (start + s).min(l);
        let mut v = vec![0u8; s];
        if start < end {
            v[..end - start].copy_from_slice(&d[start..end]);
        }
        v
    };

    let mut data_checksums = Vec::with_capacity(num_data_shards);
    for i in 0..num_data_shards {
        data_checksums.push(error_correction_shard_checksum(&shard_at(i)));
    }

    // Invariant: (k, p) always originate from `error_correction_grid` (k in [2,20],
    // p in {1,2}, so k+p < 256 and both >= 1). RS construction therefore cannot fail,
    // and below every shard is exactly `s` bytes with `k+p` slots, so `encode` cannot
    // fail either. The expects document those invariants rather than masking real errors.
    let rs = ReedSolomon::<reed_solomon_erasure::galois_8::Field>::new(k, p)
        .expect("invalid ReedSolomon parameters (k,p must come from error_correction_grid)");

    let mut parity_data = Vec::with_capacity(num_groups * p * s);
    let mut parity_checksums = Vec::with_capacity(num_groups * p);

    for g in 0..num_groups {
        // K data slots + P parity slots. Slots past num_data_shards stay zero
        // (virtual padding); parity slots are filled by encode.
        let mut shards: Vec<Vec<u8>> = vec![vec![0u8; s]; k + p];
        for (local, shard) in shards.iter_mut().take(k).enumerate() {
            let gi = g * k + local;
            if gi < num_data_shards {
                *shard = shard_at(gi);
            }
        }
        rs.encode(&mut shards).expect("RS encode failed");
        for pp in 0..p {
            let par = &shards[k + pp];
            parity_checksums.push(error_correction_shard_checksum(par));
            parity_data.extend_from_slice(par);
        }
    }

    let header = ErrorCorrectionPayloadHeader {
        data_shards: k as u16,
        parity_shards: p as u16,
        shard_size: s as u32,
        total_data_len: l as u64,
    };

    let payload = ErrorCorrectionPayload {
        header,
        data_checksums,
        parity_checksums,
        parity_data,
    }
    .to_bytes();
    let protected = l as u64;
    let overhead = if protected > 0 {
        (payload.len() as f64 / protected as f64) * 100.0
    } else {
        0.0
    };
    (payload, total_shards, protected, overhead)
}

/// Compute the AVEC parity for a single fixed metadata region, treating it as one
/// block. An empty region yields an empty payload.
#[allow(dead_code)]
pub(crate) fn compute_metadata_parity(region: &[u8], k: usize, p: usize) -> Vec<u8> {
    if region.is_empty() {
        return Vec::new();
    }
    let (payload, _shards, _prot, _ov) = compute_error_correction_shards_grid(&[region], k, p);
    payload
}

/// Reconstruct damaged bytes in the protected block stream using the v2 Error
/// Correction payload.
///
/// `blocks`: the protected blocks in order, each with exactly the same length used
///           when parity was computed; callers zero-pad truncated blocks if needed.
/// `error_correction_payload_bytes`: serialized AVEC payload bytes.
///
/// Damaged shards are located by per-shard checksum mismatch (data and parity), then
/// RS-reconstructed per group; recovered bytes are written back into `blocks` in place.
/// Returns the number of data shards successfully reconstructed.
///
/// A successful return does NOT imply caller-level correctness: the caller must
/// re-verify repaired bytes against its authenticated manifest/checksum before
/// persisting. A grid misalignment is rejected up front so good data can never be
/// silently overwritten.
pub(crate) fn reconstruct_from_error_correction(
    blocks: &mut [Vec<u8>],
    error_correction_payload_bytes: &[u8],
) -> Result<usize, String> {
    if error_correction_payload_bytes.is_empty() {
        return Ok(0);
    }
    let payload = ErrorCorrectionPayload::from_bytes(error_correction_payload_bytes)?;
    let k = payload.header.data_shards as usize;
    let p = payload.header.parity_shards as usize;
    let s = payload.header.shard_size as usize;
    let l = payload.header.total_data_len as usize;

    // The block stream must match the stream the parity was computed over, otherwise
    // the shard grid would be misaligned and reconstruction could corrupt good data.
    let total: usize = blocks.iter().map(|b| b.len()).sum();
    if total != l {
        return Err(format!(
            "Error Correction reconstruct: block stream length {} != payload stream length {}",
            total, l
        ));
    }

    let mut d = Vec::with_capacity(l);
    for b in blocks.iter() {
        d.extend_from_slice(b);
    }

    let (num_data_shards, num_groups) = error_correction_geometry(&payload.header);

    let shard_at = |d: &[u8], idx: usize| -> Vec<u8> {
        let start = idx * s;
        let end = (start + s).min(l);
        let mut v = vec![0u8; s];
        if start < end {
            v[..end - start].copy_from_slice(&d[start..end]);
        }
        v
    };

    let rs = ReedSolomon::<reed_solomon_erasure::galois_8::Field>::new(k, p)
        .map_err(|e| format!("RS create for reconstruct: {:?}", e))?;

    let mut recovered = 0usize;
    let mut changed = false;

    for g in 0..num_groups {
        let mut opt: Vec<Option<Vec<u8>>> = vec![None; k + p];
        let mut erased_data = 0usize;

        for (local, slot) in opt.iter_mut().take(k).enumerate() {
            let gi = g * k + local;
            if gi < num_data_shards {
                let sh = shard_at(&d, gi);
                if error_correction_shard_checksum(&sh) == payload.data_checksums[gi] {
                    *slot = Some(sh); // shard intact
                } else {
                    erased_data += 1; // damaged -> RS erasure
                }
            } else {
                *slot = Some(vec![0u8; s]); // virtual zero-pad slot
            }
        }

        for pp in 0..p {
            let pidx = g * p + pp;
            let start = pidx * s;
            if start + s <= payload.parity_data.len() {
                let par = payload.parity_data[start..start + s].to_vec();
                if error_correction_shard_checksum(&par) == payload.parity_checksums[pidx] {
                    opt[k + pp] = Some(par); // parity intact
                }
                // else: rotted parity -> leave None so RS routes around it
            }
        }

        if erased_data == 0 {
            continue; // nothing damaged in this group
        }
        if rs.reconstruct(&mut opt).is_err() {
            continue; // more erasures than parity can cover; leave group untouched
        }

        for (local, slot) in opt.iter().take(k).enumerate() {
            let gi = g * k + local;
            if gi >= num_data_shards {
                continue;
            }
            if let Some(sh) = slot {
                let start = gi * s;
                let end = (start + s).min(l);
                if d[start..end] != sh[..end - start] {
                    d[start..end].copy_from_slice(&sh[..end - start]);
                    changed = true;
                }
            }
        }
        recovered += erased_data;
    }

    if changed {
        // Re-slice the recovered stream back into the fixed-length blocks.
        let mut pos = 0usize;
        for b in blocks.iter_mut() {
            let len = b.len();
            b.copy_from_slice(&d[pos..pos + len]);
            pos += len;
        }
    }

    Ok(recovered)
}

/// Per-shard / per-copy health readout for a verify (Ehud #9). Aggregated across
/// every window of a standalone file (or every block group). Counts are additive
/// so a caller can fold one window's scan into a running total.
#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize)]
pub struct ShardHealth {
    /// Total data shards examined.
    pub total_data_shards: u64,
    /// Total parity shards (copies) examined.
    pub total_parity_shards: u64,
    /// Data shards whose checksum did not match (rotted / damaged).
    pub damaged_data_shards: u64,
    /// Parity shards (copies) whose checksum did not match.
    pub damaged_parity_shards: u64,
    /// Reed-Solomon groups examined.
    pub groups: u64,
    /// Groups where damaged data shards exceed the surviving parity, i.e. the
    /// damage is beyond what error correction can recover.
    pub unrecoverable_groups: u64,
}

/// Non-mutating counterpart of [`reconstruct_from_error_correction`]: scans the
/// same shard grid and folds per-shard damage counts into `health` WITHOUT
/// reconstructing or touching `blocks`. Used by the read-only verify health
/// readout (#9). Mirrors the checksum logic of the reconstruct path exactly so
/// the two agree on what "damaged" means.
pub(crate) fn scan_error_correction_health(
    blocks: &[Vec<u8>],
    error_correction_payload_bytes: &[u8],
    health: &mut ShardHealth,
) -> Result<(), String> {
    if error_correction_payload_bytes.is_empty() {
        return Ok(());
    }
    let payload = ErrorCorrectionPayload::from_bytes(error_correction_payload_bytes)?;
    let k = payload.header.data_shards as usize;
    let p = payload.header.parity_shards as usize;
    let s = payload.header.shard_size as usize;
    let l = payload.header.total_data_len as usize;

    let total: usize = blocks.iter().map(|b| b.len()).sum();
    if total != l {
        return Err(format!(
            "Error Correction scan: block stream length {} != payload stream length {}",
            total, l
        ));
    }

    let mut d = Vec::with_capacity(l);
    for b in blocks.iter() {
        d.extend_from_slice(b);
    }

    let (num_data_shards, num_groups) = error_correction_geometry(&payload.header);

    let shard_at = |d: &[u8], idx: usize| -> Vec<u8> {
        let start = idx * s;
        let end = (start + s).min(l);
        let mut v = vec![0u8; s];
        if start < end {
            v[..end - start].copy_from_slice(&d[start..end]);
        }
        v
    };

    for g in 0..num_groups {
        let mut damaged_data = 0usize;
        for local in 0..k {
            let gi = g * k + local;
            if gi >= num_data_shards {
                continue; // virtual zero-pad slot, not a stored shard
            }
            health.total_data_shards += 1;
            let sh = shard_at(&d, gi);
            if error_correction_shard_checksum(&sh) != payload.data_checksums[gi] {
                damaged_data += 1;
                health.damaged_data_shards += 1;
            }
        }

        let mut surviving_parity = 0usize;
        for pp in 0..p {
            let pidx = g * p + pp;
            let start = pidx * s;
            health.total_parity_shards += 1;
            if start + s <= payload.parity_data.len()
                && error_correction_shard_checksum(&payload.parity_data[start..start + s])
                    == payload.parity_checksums[pidx]
            {
                surviving_parity += 1;
            } else {
                health.damaged_parity_shards += 1;
            }
        }

        health.groups += 1;
        if damaged_data > surviving_parity {
            health.unrecoverable_groups += 1;
        }
    }

    Ok(())
}

// ───────────────────────────────────────────────────────────────────────────
// Public standalone API: error-correct ANY file with a detached `.aerocorrect`
// sidecar (the `aeroftp correct` subcommand). The format binds by content
// SHA-256, so this is format-agnostic (works on vault containers or any file).
// This is the curated public surface; the codec/format submodules stay crate-private.
// ───────────────────────────────────────────────────────────────────────────

use serde::Serialize;
use std::path::Path;

/// Result of generating a standalone `.aerocorrect` sidecar.
#[derive(Debug, Clone, Serialize)]
pub struct CorrectGenerateReport {
    pub file: String,
    pub sidecar: String,
    pub file_size: u64,
    pub sidecar_size: u64,
    pub overhead_pct: f64,
    /// Number of windows (parity segments) in the sidecar.
    pub segments: u64,
    pub shards: u64,
    pub level_pct: u32,
}

/// Result of a read-only standalone verify.
#[derive(Debug, Clone, Serialize)]
pub struct CorrectVerifyReport {
    pub file: String,
    pub sidecar: String,
    /// `"verified"` (intact) or `"needs_repair"` (corruption detected).
    pub status: String,
    pub verified: bool,
    /// Per-shard / per-copy health readout from a full grid scan (Ehud #9): how
    /// many data/parity shards are damaged and whether any group is beyond
    /// recovery. `None` only if a scan was not run.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub health: Option<ShardHealth>,
}

/// Result of a standalone repair attempt.
#[derive(Debug, Clone, Serialize)]
pub struct CorrectRepairReport {
    pub file: String,
    pub sidecar: String,
    /// `"verified"` (was already intact) or `"repaired"`.
    pub status: String,
    pub repaired: bool,
    pub recovered_shards: u64,
}

/// Default sidecar path for `file`: `<file>.aerocorrect`.
pub fn aerocorrect_sidecar_path_for(file: &str) -> String {
    sidecar::aerocorrect_sidecar_path(file)
}

fn rel_name(file: &str) -> String {
    Path::new(file)
        .file_name()
        .and_then(|n| n.to_str())
        .unwrap_or(file)
        .to_string()
}

/// Generate a detached `.aerocorrect` sidecar for `file` at overhead level `pct`. Streams
/// the file window by window (bounded memory). Writes the sidecar to `out` or
/// `<file>.aerocorrect`. Returns a report. Errors if the file exceeds the size cap.
pub fn correct_generate(
    file: &str,
    pct: u32,
    out: Option<&str>,
) -> Result<CorrectGenerateReport, String> {
    correct_generate_with_progress(file, pct, out, &mut |_, _| {})
}

/// [`correct_generate`] with a `(bytes_done, bytes_total)` progress callback,
/// invoked after each window's parity is computed (#5: progress for `correct`,
/// GUI + CLI). [`correct_generate`] is the no-op-callback shorthand.
pub fn correct_generate_with_progress(
    file: &str,
    pct: u32,
    out: Option<&str>,
    progress: &mut dyn FnMut(u64, u64),
) -> Result<CorrectGenerateReport, String> {
    let path = Path::new(file);
    let rel = rel_name(file);
    // No minimum-benefit gate here: `correct gen` is an explicit per-file request, so honor
    // it even for tiny files (the gate is a sync-pipeline opt-in, not a CLI-of-one concern).
    let result = sync::generate_sync_sidecar_for_file_capped_with_progress(
        &rel,
        path,
        pct,
        sync::AEROSYNC_EC_MAX_FILE_SIZE,
        0,
        progress,
    )?;
    let generated = match result {
        sync::SyncEcGenerateResult::Generated(g) => g,
        sync::SyncEcGenerateResult::SkippedTooLarge {
            file_size,
            max_file_size,
        } => {
            return Err(format!(
                "{file} is {file_size} bytes, above the {max_file_size}-byte error-correction cap"
            ));
        }
        sync::SyncEcGenerateResult::SkippedLowBenefit { .. } => {
            // Unreachable: the gate is disabled above (max_overhead_pct = 0).
            unreachable!("correct gen does not enable the minimum-benefit gate")
        }
    };
    let sidecar_path = out
        .map(|s| s.to_string())
        .unwrap_or_else(|| sidecar::aerocorrect_sidecar_path(file));
    std::fs::write(&sidecar_path, &generated.sidecar_bytes)
        .map_err(|e| format!("write sidecar {sidecar_path}: {e}"))?;
    let segments =
        sidecar::aerocorrect_windows(generated.file_size, sidecar::AEROCORRECT_WINDOW_SIZE).len()
            as u64;
    Ok(CorrectGenerateReport {
        file: file.to_string(),
        sidecar: sidecar_path,
        file_size: generated.file_size,
        sidecar_size: generated.sidecar_len,
        overhead_pct: generated.overhead_pct,
        segments,
        shards: generated.shards,
        level_pct: pct,
    })
}

/// Verify `file` against its `.aerocorrect` sidecar (read-only, never mutates the file).
/// `parity` overrides the default `<file>.aerocorrect` path.
pub fn correct_verify(file: &str, parity: Option<&str>) -> Result<CorrectVerifyReport, String> {
    correct_verify_with_progress(file, parity, &mut |_, _| {})
}

/// [`correct_verify`] with a `(bytes_done, bytes_total)` progress callback,
/// invoked after each window is scanned (#5: progress for `correct verify`).
pub fn correct_verify_with_progress(
    file: &str,
    parity: Option<&str>,
    progress: &mut dyn FnMut(u64, u64),
) -> Result<CorrectVerifyReport, String> {
    let path = Path::new(file);
    let rel = rel_name(file);
    let sidecar_path = parity
        .map(|s| s.to_string())
        .unwrap_or_else(|| sidecar::aerocorrect_sidecar_path(file));
    // #9: a full grid scan gives both the verified verdict and a per-shard health
    // readout in one streamed pass. (The hot sync path keeps using the lighter
    // hash-only verify_standalone_file_streamed; this deeper scan is the explicit
    // `correct verify` health readout.)
    let (result, health) =
        sync::scan_standalone_file_streamed(&rel, path, Path::new(&sidecar_path), progress)?;
    let verified = matches!(result, sync::StandaloneVerifyResult::Verified);
    Ok(CorrectVerifyReport {
        file: file.to_string(),
        sidecar: sidecar_path,
        status: if verified { "verified" } else { "needs_repair" }.to_string(),
        verified,
        health: Some(health),
    })
}

/// Repair `file` in place from its `.aerocorrect` sidecar (atomic, all-or-nothing). A file
/// already intact is reported as `verified` (no write). `parity` overrides the default path.
///
/// This is an INTEGRITY repair: it reconstructs toward the content hash the sidecar
/// declares. To additionally assert AUTHENTICITY (that the recovered content is the one
/// you expect, not whatever a planted sidecar declares), use [`correct_repair_anchored`]
/// with an out-of-band good SHA-256. See the audit-M3 trust-model note on
/// `sync::verify_repair_standalone_file_streamed`.
pub fn correct_repair(file: &str, parity: Option<&str>) -> Result<CorrectRepairReport, String> {
    correct_repair_anchored(file, parity, None)
}

/// Like [`correct_repair`], but when `expect_sha256` (a 64-char hex SHA-256) is given it
/// anchors authenticity: a sidecar whose declared content hash differs from the expected
/// hash is refused before any write (audit M3). This is what a bare CLI's
/// `--expect-sha256` wires into.
pub fn correct_repair_anchored(
    file: &str,
    parity: Option<&str>,
    expect_sha256: Option<&str>,
) -> Result<CorrectRepairReport, String> {
    correct_repair_anchored_with_progress(file, parity, expect_sha256, &mut |_, _| {})
}

/// [`correct_repair_anchored`] with a `(bytes_done, bytes_total)` progress
/// callback, invoked after each window is streamed/repaired (#5: progress for
/// `correct repair`). The fast already-intact path fires no per-window callback.
pub fn correct_repair_anchored_with_progress(
    file: &str,
    parity: Option<&str>,
    expect_sha256: Option<&str>,
    progress: &mut dyn FnMut(u64, u64),
) -> Result<CorrectRepairReport, String> {
    let path = Path::new(file);
    let rel = rel_name(file);
    let sidecar_path = parity
        .map(|s| s.to_string())
        .unwrap_or_else(|| sidecar::aerocorrect_sidecar_path(file));
    let anchor = match expect_sha256 {
        Some(hex) => Some(sync::parse_sha256_hex(hex)?),
        None => None,
    };
    let (status, repaired, recovered_shards) =
        match sync::verify_repair_standalone_file_streamed_with_progress(
            &rel,
            path,
            Path::new(&sidecar_path),
            anchor.as_ref(),
            progress,
        )? {
            sync::SyncEcRepairResult::Verified => ("verified".to_string(), false, 0u64),
            sync::SyncEcRepairResult::Repaired { recovered_shards } => {
                ("repaired".to_string(), true, recovered_shards as u64)
            }
        };
    Ok(CorrectRepairReport {
        file: file.to_string(),
        sidecar: sidecar_path,
        status,
        repaired,
        recovered_shards,
    })
}

// `parse_sha256_hex` for the `--expect-sha256` anchor now lives in `sync` (the single
// EC implementation); see `sync::parse_sha256_hex`.

#[cfg(test)]
mod tests {
    use super::*;

    /// Deterministic pseudo-random bytes (BLAKE3 keystream).
    fn sample(len: usize) -> Vec<u8> {
        let mut seed = *blake3::hash(b"ec-mod-test-seed").as_bytes();
        let mut out = Vec::with_capacity(len);
        while out.len() < len {
            seed = *blake3::hash(&seed).as_bytes();
            out.extend_from_slice(&seed);
        }
        out.truncate(len);
        out
    }

    #[test]
    fn grid_table_and_monotonic() {
        assert_eq!(error_correction_grid(5), (20, 1));
        assert_eq!(error_correction_grid(7), (14, 1));
        assert_eq!(error_correction_grid(10), (20, 2)); // 10% overhead == K20/P2
        assert_eq!(error_correction_grid(15), (13, 2));
        assert_eq!(error_correction_grid(20), (10, 2));
        assert_eq!(error_correction_grid(25), (8, 2));
        assert_eq!(error_correction_grid(30), (7, 2));
        assert_eq!(error_correction_grid(50), (4, 2));
        // Out-of-range percentages clamp to the [MIN, MAX] band.
        assert_eq!(error_correction_grid(1), error_correction_grid(5));
        assert_eq!(error_correction_grid(99), error_correction_grid(50));
        // Realized overhead p/k is monotonic non-decreasing and k stays >= 2.
        let mut prev = 0.0f64;
        for pct in 5..=50 {
            let (k, p) = error_correction_grid(pct);
            assert!(k >= 2, "k must stay >= 2 at pct={pct}");
            let ratio = p as f64 / k as f64;
            assert!(
                ratio + 1e-9 >= prev,
                "overhead must be non-decreasing at pct={pct}"
            );
            prev = ratio;
        }
    }

    #[test]
    fn roundtrip_recovers_single_erasure() {
        let data = sample(50_000);
        let (payload, _shards, prot, _ov) = compute_error_correction_shards_grid(&[&data], 10, 2);
        assert_eq!(prot, data.len() as u64);
        let mut damaged = data.clone();
        damaged[12_345] ^= 0xFF;
        let mut blocks = vec![damaged];
        let recovered = reconstruct_from_error_correction(&mut blocks, &payload).unwrap();
        assert!(recovered >= 1);
        assert_eq!(blocks[0], data);
    }

    #[test]
    fn erasure_budget_boundary_recovers_p_and_leaves_p_plus_1() {
        let data = sample(60_000);
        let (payload, _s, _p, _o) = compute_error_correction_shards_grid(&[&data], 10, 2);
        let s = ErrorCorrectionPayload::from_bytes(&payload)
            .unwrap()
            .header
            .shard_size as usize;
        // Exactly P=2 erased shards: recoverable.
        let mut d2 = data.clone();
        d2[0] ^= 0xFF;
        d2[s] ^= 0xFF;
        let mut b2 = vec![d2];
        assert!(reconstruct_from_error_correction(&mut b2, &payload).unwrap() >= 2);
        assert_eq!(b2[0], data);
        // P+1=3 erased shards: beyond budget -> group untouched, no partial corruption.
        let mut d3 = data.clone();
        d3[0] ^= 0xFF;
        d3[s] ^= 0xFF;
        d3[2 * s] ^= 0xFF;
        let before = d3.clone();
        let mut b3 = vec![d3];
        assert_eq!(
            reconstruct_from_error_correction(&mut b3, &payload).unwrap(),
            0
        );
        assert_eq!(b3[0], before);
    }

    #[test]
    fn multigroup_recovers_one_erasure_per_group() {
        // 10 MiB at (4,2) forces shard size to clamp at MAX_SHARD and >1 group.
        let data = sample(10 * 1024 * 1024);
        let (payload, _s, _p, _o) = compute_error_correction_shards_grid(&[&data], 4, 2);
        let header = ErrorCorrectionPayload::from_bytes(&payload).unwrap().header;
        let (num_data, num_groups) = error_correction_geometry(&header);
        assert!(num_groups > 1, "expected multiple groups, got {num_groups}");
        let s = header.shard_size as usize;
        // Corrupt the first data shard of every group.
        let mut damaged = data.clone();
        for g in 0..num_groups {
            let gi = g * header.data_shards as usize;
            if gi < num_data {
                damaged[gi * s] ^= 0xFF;
            }
        }
        let mut blocks = vec![damaged];
        assert!(reconstruct_from_error_correction(&mut blocks, &payload).unwrap() >= num_groups);
        assert_eq!(blocks[0], data);
    }

    #[test]
    fn reconstruct_rejects_wrong_total_len() {
        let data = sample(20_000);
        let (payload, _s, _p, _o) = compute_error_correction_shards_grid(&[&data], 10, 2);
        let mut wrong = vec![sample(19_999)];
        assert!(reconstruct_from_error_correction(&mut wrong, &payload).is_err());
    }

    #[test]
    fn metadata_parity_empty_and_single_region() {
        assert!(compute_metadata_parity(&[], 10, 2).is_empty());
        let region = sample(8_000);
        let parity = compute_metadata_parity(&region, 10, 2);
        assert!(!parity.is_empty());
        let mut damaged = region.clone();
        damaged[1234] ^= 0xFF;
        let mut blocks = vec![damaged];
        assert!(reconstruct_from_error_correction(&mut blocks, &parity).unwrap() >= 1);
        assert_eq!(blocks[0], region);
    }

    #[test]
    fn payload_header_rejects_malformed() {
        let mut bad = vec![0u8; 32];
        assert!(ErrorCorrectionPayloadHeader::from_bytes(&bad).is_err()); // bad magic
        bad[0..4].copy_from_slice(ERROR_CORRECTION_PAYLOAD_MAGIC);
        bad[4..6].copy_from_slice(&999u16.to_le_bytes());
        assert!(ErrorCorrectionPayloadHeader::from_bytes(&bad).is_err()); // bad version
        let mut z = vec![0u8; 32];
        z[0..4].copy_from_slice(ERROR_CORRECTION_PAYLOAD_MAGIC);
        z[4..6].copy_from_slice(&ERROR_CORRECTION_PAYLOAD_VERSION.to_le_bytes());
        assert!(ErrorCorrectionPayloadHeader::from_bytes(&z).is_err()); // zero geometry
        z[6..8].copy_from_slice(&1u16.to_le_bytes());
        z[10..14].copy_from_slice(&4096u32.to_le_bytes());
        assert!(ErrorCorrectionPayloadHeader::from_bytes(&z).is_err()); // zero parity
        assert!(ErrorCorrectionPayloadHeader::from_bytes(&[0u8; 10]).is_err()); // too short
    }

    #[test]
    fn from_bytes_rejects_overflow_geometry_without_panic_or_oom() {
        // SECURITY REGRESSION (CRITICAL-1): an AVEC blob arrives inside an untrusted
        // sidecar. A header with astronomically large geometry must be rejected by the
        // checked-arithmetic guard, never overflow (debug panic) nor drive a giant
        // Vec::with_capacity (release OOM).
        let h1 = ErrorCorrectionPayloadHeader {
            data_shards: 1,
            parity_shards: 1,
            shard_size: 1,
            total_data_len: u64::MAX,
        }
        .to_bytes();
        assert!(ErrorCorrectionPayload::from_bytes(&h1).is_err());

        let h2 = ErrorCorrectionPayloadHeader {
            data_shards: 2,
            parity_shards: u16::MAX,
            shard_size: u32::MAX,
            total_data_len: u64::MAX,
        }
        .to_bytes();
        assert!(ErrorCorrectionPayload::from_bytes(&h2).is_err());
    }

    #[test]
    fn empty_input_yields_empty_payload() {
        let (payload, shards, prot, ov) = compute_error_correction_shards_grid(&[], 10, 2);
        assert!(payload.is_empty());
        assert_eq!((shards, prot, ov), (0, 0, 0.0));
        // Reconstruct with an empty payload is a no-op.
        let mut blocks: Vec<Vec<u8>> = vec![];
        assert_eq!(
            reconstruct_from_error_correction(&mut blocks, &payload).unwrap(),
            0
        );
    }

    fn write_file(dir: &std::path::Path, name: &str, bytes: &[u8]) -> String {
        let p = dir.join(name);
        std::fs::write(&p, bytes).unwrap();
        p.to_string_lossy().into_owned()
    }

    #[test]
    fn correct_generate_default_sidecar_path_and_round_trip() {
        let dir = tempfile::tempdir().unwrap();
        let file = write_file(dir.path(), "data.bin", &sample(40_000));

        let gen = correct_generate(&file, 15, None).unwrap();
        assert_eq!(gen.sidecar, format!("{file}.aerocorrect"));
        assert_eq!(gen.sidecar, aerocorrect_sidecar_path_for(&file));
        assert!(std::path::Path::new(&gen.sidecar).exists());
        assert_eq!(gen.segments, 1, "small file is a single window");
        assert!(gen.shards > 0);

        // Intact file verifies and needs no repair.
        let v = correct_verify(&file, None).unwrap();
        assert!(v.verified && v.status == "verified");
        let r = correct_repair(&file, None).unwrap();
        assert!(!r.repaired && r.status == "verified");
    }

    #[test]
    fn correct_verify_detects_and_repair_fixes_corruption() {
        let dir = tempfile::tempdir().unwrap();
        let original = sample(50_000);
        let file = write_file(dir.path(), "doc.bin", &original);
        correct_generate(&file, 25, None).unwrap();

        // Corrupt a span within the file (well under the parity budget).
        let mut bytes = original.clone();
        for b in bytes.iter_mut().take(1_000).skip(100) {
            *b ^= 0xFF;
        }
        std::fs::write(&file, &bytes).unwrap();

        let v = correct_verify(&file, None).unwrap();
        assert!(!v.verified && v.status == "needs_repair");

        let r = correct_repair(&file, None).unwrap();
        assert!(r.repaired && r.status == "repaired" && r.recovered_shards > 0);
        assert_eq!(
            std::fs::read(&file).unwrap(),
            original,
            "byte-identical repair"
        );

        // Read-only verify after repair confirms intact.
        assert!(correct_verify(&file, None).unwrap().verified);
    }

    /// Audit M3: the authenticity anchor refuses a sidecar whose declared content
    /// hash differs from the caller's out-of-band expected hash, before any write;
    /// the correct anchor repairs byte-identically; a malformed hash is rejected.
    #[test]
    fn correct_repair_anchored_refuses_mismatched_expected_hash() {
        use sha2::{Digest, Sha256};
        let dir = tempfile::tempdir().unwrap();
        let original = sample(40_000);
        let file = write_file(dir.path(), "doc.bin", &original);
        correct_generate(&file, 25, None).unwrap();

        // Corrupt within the parity budget so a bare repair would otherwise succeed.
        let mut bytes = original.clone();
        for b in bytes.iter_mut().take(500).skip(50) {
            *b ^= 0xFF;
        }
        std::fs::write(&file, &bytes).unwrap();

        // A wrong anchored hash refuses before any write; the file stays corrupt.
        let before = std::fs::read(&file).unwrap();
        let wrong = "0".repeat(64);
        assert!(correct_repair_anchored(&file, None, Some(&wrong)).is_err());
        assert_eq!(
            std::fs::read(&file).unwrap(),
            before,
            "no write on anchor mismatch"
        );

        // A malformed expected hash is rejected (length + non-hex).
        assert!(correct_repair_anchored(&file, None, Some("deadbeef")).is_err());
        assert!(correct_repair_anchored(&file, None, Some(&"z".repeat(64))).is_err());

        // The correct anchored hash (the original content) repairs byte-identically.
        let good: String = {
            let mut h = Sha256::new();
            h.update(&original);
            h.finalize().iter().map(|b| format!("{b:02x}")).collect()
        };
        let r = correct_repair_anchored(&file, None, Some(&good)).unwrap();
        assert!(r.repaired && r.status == "repaired");
        assert_eq!(
            std::fs::read(&file).unwrap(),
            original,
            "byte-identical repair"
        );
    }

    #[test]
    fn correct_custom_out_and_parity_paths() {
        let dir = tempfile::tempdir().unwrap();
        let file = write_file(dir.path(), "payload.bin", &sample(30_000));
        let side = dir.path().join("custom.aerocorrect");
        let side = side.to_string_lossy().into_owned();

        let gen = correct_generate(&file, 15, Some(&side)).unwrap();
        assert_eq!(gen.sidecar, side);
        assert!(!std::path::Path::new(&format!("{file}.aerocorrect")).exists());
        assert!(correct_verify(&file, Some(&side)).unwrap().verified);
    }

    #[test]
    fn correct_repair_against_foreign_sidecar_leaves_file_intact() {
        let dir = tempfile::tempdir().unwrap();
        // Two same-size files with different content; sidecar belongs to A.
        let a = write_file(dir.path(), "a.bin", &sample(20_000));
        let b_bytes = {
            let mut v = sample(20_000);
            v.reverse();
            v
        };
        let b = write_file(dir.path(), "b.bin", &b_bytes);
        let side = correct_generate(&a, 25, None).unwrap().sidecar;

        // B does not match A's sidecar: verify says needs_repair, repair fails closed.
        assert!(!correct_verify(&b, Some(&side)).unwrap().verified);
        assert!(correct_repair(&b, Some(&side)).is_err());
        assert_eq!(
            std::fs::read(&b).unwrap(),
            b_bytes,
            "B untouched after failed repair"
        );
    }
}