pi-mega-compact 0.17.0 → 0.18.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (121) hide show
  1. package/dist/config/vector-cortex.js +20 -0
  2. package/dist/dedup/digest.js +30 -0
  3. package/dist/dedup/l1-lsh-v2.js +60 -0
  4. package/dist/dedup/l1-lsh.js +52 -0
  5. package/dist/dedup/l1-minhash-v2.js +193 -0
  6. package/dist/dedup/l1-minhash.js +91 -0
  7. package/dist/dedup/l1-verify.js +54 -0
  8. package/dist/dedup/mmr.js +45 -0
  9. package/dist/dedup/normalize.js +39 -0
  10. package/dist/dedup/raptor/buildHistory.js +164 -0
  11. package/dist/dedup/raptor/guardrails.js +83 -0
  12. package/dist/dedup/raptor/incremental.js +304 -0
  13. package/dist/dedup/raptor/index.js +190 -0
  14. package/dist/dedup/raptor/kmeans.js +152 -0
  15. package/dist/dedup/raptor/multilevel.js +184 -0
  16. package/dist/dedup/raptor/retrieval.js +93 -0
  17. package/dist/dedup/raptor/summarizer.js +86 -0
  18. package/dist/dedup/raptor/tree.js +202 -0
  19. package/dist/dedup/topk.js +60 -0
  20. package/dist/extensions/dashboard-server/routes-rag-settings-helpers.js +2 -0
  21. package/dist/extensions/mega-events/context-handler/dbMirrorAppend.js +63 -0
  22. package/dist/extensions/mega-events/context-handler/gateCheck.js +59 -0
  23. package/dist/extensions/mega-events/context-handler/liveTrim.js +178 -0
  24. package/dist/extensions/mega-events/context-handler/pipelineRun.js +37 -0
  25. package/dist/extensions/mega-events/context-handler.js +39 -305
  26. package/dist/src/config/vector-cortex.js +20 -0
  27. package/dist/src/config.js +1 -1
  28. package/dist/src/dedup/l1-lsh-v2.js +60 -0
  29. package/dist/src/dedup/l1-minhash-v2.js +193 -0
  30. package/dist/src/dedup/normalize.js +1 -1
  31. package/dist/src/vector-cortex/conformance/emit.js +46 -0
  32. package/dist/src/vector-cortex/conformance/manifest.js +235 -0
  33. package/dist/src/vector-cortex/conformance/runner.js +111 -0
  34. package/dist/src/vector-cortex/conformance/triadB-reader.js +95 -0
  35. package/dist/src/vector-cortex/encoder/asset.js +142 -0
  36. package/dist/src/vector-cortex/encoder/emit.js +42 -0
  37. package/dist/src/vector-cortex/encoder/runtime.js +228 -0
  38. package/dist/src/vector-cortex/encoder/types.js +85 -0
  39. package/dist/src/vector-cortex/migrations/minhash-v2.js +209 -0
  40. package/dist/vector-cortex/conformance/emit.js +46 -0
  41. package/dist/vector-cortex/conformance/manifest.js +235 -0
  42. package/dist/vector-cortex/conformance/runner.js +111 -0
  43. package/dist/vector-cortex/conformance/triadB-reader.js +95 -0
  44. package/dist/vector-cortex/encoder/asset.js +142 -0
  45. package/dist/vector-cortex/encoder/emit.js +42 -0
  46. package/dist/vector-cortex/encoder/runtime.js +228 -0
  47. package/dist/vector-cortex/encoder/types.js +85 -0
  48. package/dist/vector-cortex/migrations/minhash-v2.js +209 -0
  49. package/extensions/dashboard-client/dist/assets/{AreaChart-CAnWomhw.js → AreaChart-BQXFF7UI.js} +2 -2
  50. package/extensions/dashboard-client/dist/assets/{AreaChart-CAnWomhw.js.map → AreaChart-BQXFF7UI.js.map} +1 -1
  51. package/extensions/dashboard-client/dist/assets/{BarChart-lZ2-s71k.js → BarChart-BVtoz9Eg.js} +2 -2
  52. package/extensions/dashboard-client/dist/assets/{BarChart-lZ2-s71k.js.map → BarChart-BVtoz9Eg.js.map} +1 -1
  53. package/extensions/dashboard-client/dist/assets/{CacheTab-BL2Thh66.js → CacheTab-trI2TIGy.js} +2 -2
  54. package/extensions/dashboard-client/dist/assets/{CacheTab-BL2Thh66.js.map → CacheTab-trI2TIGy.js.map} +1 -1
  55. package/extensions/dashboard-client/dist/assets/{EventsTab-DEh2Z9fp.js → EventsTab-CNUbk9__.js} +2 -2
  56. package/extensions/dashboard-client/dist/assets/{EventsTab-DEh2Z9fp.js.map → EventsTab-CNUbk9__.js.map} +1 -1
  57. package/extensions/dashboard-client/dist/assets/{HealthTab-DTwKclxk.js → HealthTab-K-bd_CM-.js} +2 -2
  58. package/extensions/dashboard-client/dist/assets/{HealthTab-DTwKclxk.js.map → HealthTab-K-bd_CM-.js.map} +1 -1
  59. package/extensions/dashboard-client/dist/assets/{MaintenanceTab-DDfBEdbX.js → MaintenanceTab-CoEioKzx.js} +2 -2
  60. package/extensions/dashboard-client/dist/assets/{MaintenanceTab-DDfBEdbX.js.map → MaintenanceTab-CoEioKzx.js.map} +1 -1
  61. package/extensions/dashboard-client/dist/assets/{MemoryMapTab-Dgfmg-0Q.js → MemoryMapTab-CJo24ulv.js} +2 -2
  62. package/extensions/dashboard-client/dist/assets/{MemoryMapTab-Dgfmg-0Q.js.map → MemoryMapTab-CJo24ulv.js.map} +1 -1
  63. package/extensions/dashboard-client/dist/assets/{MetricsTab-DOcMm0Tx.js → MetricsTab-DCNlQWWP.js} +2 -2
  64. package/extensions/dashboard-client/dist/assets/{MetricsTab-DOcMm0Tx.js.map → MetricsTab-DCNlQWWP.js.map} +1 -1
  65. package/extensions/dashboard-client/dist/assets/{OverviewTab-DHT5aoHg.js → OverviewTab-CQQlyIQ6.js} +2 -2
  66. package/extensions/dashboard-client/dist/assets/{OverviewTab-DHT5aoHg.js.map → OverviewTab-CQQlyIQ6.js.map} +1 -1
  67. package/extensions/dashboard-client/dist/assets/{ReposTab-CGAmtR_c.js → ReposTab-Cqm82_yn.js} +2 -2
  68. package/extensions/dashboard-client/dist/assets/{ReposTab-CGAmtR_c.js.map → ReposTab-Cqm82_yn.js.map} +1 -1
  69. package/extensions/dashboard-client/dist/assets/{SessionsTab-BIgx6iKT.js → SessionsTab-A8xTAdET.js} +2 -2
  70. package/extensions/dashboard-client/dist/assets/{SessionsTab-BIgx6iKT.js.map → SessionsTab-A8xTAdET.js.map} +1 -1
  71. package/extensions/dashboard-client/dist/assets/{SetupTab-ADghKuEI.js → SetupTab-CqaSgaZl.js} +2 -2
  72. package/extensions/dashboard-client/dist/assets/{SetupTab-ADghKuEI.js.map → SetupTab-CqaSgaZl.js.map} +1 -1
  73. package/extensions/dashboard-client/dist/assets/{TimeSavedCard-C33UiiOM.js → TimeSavedCard-C_f3q8GR.js} +2 -2
  74. package/extensions/dashboard-client/dist/assets/{TimeSavedCard-C33UiiOM.js.map → TimeSavedCard-C_f3q8GR.js.map} +1 -1
  75. package/extensions/dashboard-client/dist/assets/{TurnsTab-DdbL9wPN.js → TurnsTab-BEsafACo.js} +2 -2
  76. package/extensions/dashboard-client/dist/assets/{TurnsTab-DdbL9wPN.js.map → TurnsTab-BEsafACo.js.map} +1 -1
  77. package/extensions/dashboard-client/dist/assets/VectorCortexTab-DD1CMK80.js +2 -0
  78. package/extensions/dashboard-client/dist/assets/VectorCortexTab-DD1CMK80.js.map +1 -0
  79. package/extensions/dashboard-client/dist/assets/{WikiTab-Ch00OFKs.js → WikiTab-Bf01M2_Z.js} +2 -2
  80. package/extensions/dashboard-client/dist/assets/{WikiTab-Ch00OFKs.js.map → WikiTab-Bf01M2_Z.js.map} +1 -1
  81. package/extensions/dashboard-client/dist/assets/{button-ByeEdo2g.js → button-BEJPTk_E.js} +2 -2
  82. package/extensions/dashboard-client/dist/assets/{button-ByeEdo2g.js.map → button-BEJPTk_E.js.map} +1 -1
  83. package/extensions/dashboard-client/dist/assets/{card-DesoLXvc.js → card-DK1V382s.js} +2 -2
  84. package/extensions/dashboard-client/dist/assets/{card-DesoLXvc.js.map → card-DK1V382s.js.map} +1 -1
  85. package/extensions/dashboard-client/dist/assets/{generateCategoricalChart-CCAB7Iz4.js → generateCategoricalChart-CRtaPphL.js} +2 -2
  86. package/extensions/dashboard-client/dist/assets/{generateCategoricalChart-CCAB7Iz4.js.map → generateCategoricalChart-CRtaPphL.js.map} +1 -1
  87. package/extensions/dashboard-client/dist/assets/{index-B6cgyAPr.js → index-CB6BslE4.js} +3 -3
  88. package/extensions/dashboard-client/dist/assets/{index-B6cgyAPr.js.map → index-CB6BslE4.js.map} +1 -1
  89. package/extensions/dashboard-client/dist/assets/{switch-DIZC-oe-.js → switch-C4FV6B9k.js} +2 -2
  90. package/extensions/dashboard-client/dist/assets/{switch-DIZC-oe-.js.map → switch-C4FV6B9k.js.map} +1 -1
  91. package/extensions/dashboard-client/dist/assets/{toggle-CIoWaH2e.js → toggle-GBmeENMd.js} +2 -2
  92. package/extensions/dashboard-client/dist/assets/{toggle-CIoWaH2e.js.map → toggle-GBmeENMd.js.map} +1 -1
  93. package/extensions/dashboard-client/dist/assets/{useSSE-Bv4943l1.js → useSSE-BOKZGSn1.js} +2 -2
  94. package/extensions/dashboard-client/dist/assets/{useSSE-Bv4943l1.js.map → useSSE-BOKZGSn1.js.map} +1 -1
  95. package/extensions/dashboard-client/dist/index.html +1 -1
  96. package/extensions/dashboard-client/src/tabs/VectorCortexTab.tsx +1 -1
  97. package/extensions/dashboard-client/tsconfig.json +1 -1
  98. package/extensions/dashboard-server/routes-rag-settings-helpers.ts +12 -0
  99. package/extensions/mega-events/context-handler/dbMirrorAppend.ts +93 -0
  100. package/extensions/mega-events/context-handler/gateCheck.ts +101 -0
  101. package/extensions/mega-events/context-handler/liveTrim.ts +241 -0
  102. package/extensions/mega-events/context-handler/pipelineRun.ts +79 -0
  103. package/extensions/mega-events/context-handler.ts +45 -347
  104. package/package.json +1 -1
  105. package/src/config/vector-cortex.ts +22 -0
  106. package/src/config.ts +2 -0
  107. package/src/dedup/l1-lsh-v2.ts +78 -0
  108. package/src/dedup/l1-minhash-v2.ts +221 -0
  109. package/src/dedup/normalize.ts +1 -1
  110. package/src/store/sqlite/game-scores.ts +2 -1
  111. package/src/vector-cortex/conformance/emit.ts +56 -0
  112. package/src/vector-cortex/conformance/manifest.ts +298 -0
  113. package/src/vector-cortex/conformance/runner.ts +165 -0
  114. package/src/vector-cortex/conformance/triadB-reader.ts +110 -0
  115. package/src/vector-cortex/encoder/asset.ts +155 -0
  116. package/src/vector-cortex/encoder/emit.ts +51 -0
  117. package/src/vector-cortex/encoder/runtime.ts +283 -0
  118. package/src/vector-cortex/encoder/types.ts +165 -0
  119. package/src/vector-cortex/migrations/minhash-v2.ts +263 -0
  120. package/extensions/dashboard-client/dist/assets/VectorCortexTab-Cl1pbEdR.js +0 -2
  121. package/extensions/dashboard-client/dist/assets/VectorCortexTab-Cl1pbEdR.js.map +0 -1
@@ -0,0 +1,221 @@
1
+ /**
2
+ * l1-minhash-v2.ts — MinHash V2 signatures (VC1C, M4 minhash-v2).
3
+ *
4
+ * The v1 path (l1-minhash.ts) uses 32-bit universal hashing with `p=2^31-1`
5
+ * and Number math, which is exact because a*x < 2^62 can overflow double
6
+ * precision (>= 2^53) and is reduced too eagerly. MinHashV2 freezes a v2
7
+ * scheme that is CROSS-LANGUAGE deterministic and byte-exact:
8
+ *
9
+ * - normalization: the stable `normalize()` from normalize.ts (NFC, NFKC
10
+ * case-fold, CRLF->LF, whitespace collapse, 32K cap);
11
+ * - shingles: windows of FIVE CODE POINTS over the normalized text
12
+ * (SHINGLE_CP=5), each shingle hashed to a u64 with 64-bit FNV-1a over its
13
+ * UTF-8 bytes;
14
+ * - 256 PUBLISHED unsigned (a_i, b_i) seed pairs, derived deterministically
15
+ * via splitmix64 and listed in
16
+ * conformance/vector-cortex/v2/minhash/seeds-v2.json;
17
+ * - p = 2^61 - 1 (2305843009213693951), EXACT BigInt multiply/modulo — a_i*x
18
+ * can reach ~2^124 so Number math would silently corrupt high-bit products;
19
+ * - signature = 256 slots, slot i = min over shingles of (a_i*x + b_i) mod p;
20
+ * the empty signature is all-`p` sentinel slots;
21
+ * - encoding: 256 x u64 LITTLE-ENDIAN bytes (2048 bytes total);
22
+ * - bands: 64 bands of FOUR u64 slots each (256 total), see l1-lsh-v2.ts.
23
+ *
24
+ * NEVER compares a v2 signature against a v1 signature — cross-version compare
25
+ * is rejected with `MINHASH_VERSION_MISMATCH` (see minhash-v2 migration).
26
+ *
27
+ * Pure compute, no deps, no storage, no network (PREVENT-PI-004 / PREVENT-011).
28
+ */
29
+
30
+ import { normalize } from "./normalize.js";
31
+
32
+ /** MinHashV2 scheme version tag (frozen). */
33
+ export const MINHASH_VERSION = 2 as const;
34
+
35
+ /** The published count of slots / seed pairs (frozen). */
36
+ export const NUM_HASHES_V2 = 256 as const;
37
+
38
+ /** Five-code-point shingle length (frozen). */
39
+ export const SHINGLE_CP = 5 as const;
40
+
41
+ /** p = 2^61 - 1 (Mersenne prime), as exact BigInt. */
42
+ export const P_V2 = (1n << 61n) - 1n;
43
+
44
+ /** Little-endian bytes per u64 signature slot. */
45
+ export const U64_BYTES = 8 as const;
46
+
47
+ /** Signature byte length: 256 slots x 8 LE bytes. */
48
+ export const SIGNATURE_BYTES_V2 = NUM_HASHES_V2 * U64_BYTES;
49
+
50
+ /** Fixed splitmix64 initial state so the seed table is reproducible everywhere. */
51
+ export const SEED_STATE_INIT = 0x5eedc0de_f001_ba11n & 0xffffffffffffffffn;
52
+
53
+ /** splitmix64 mixing constants (frozen). */
54
+ const SM_GOLDEN = 0x9e3779b97f4a7c15n;
55
+ const SM_M1 = 0xbf58476d1ce4e5b9n;
56
+ const SM_M2 = 0x94d049bb133111ebn;
57
+ const MASK64 = 0xffffffffffffffffn;
58
+
59
+ function mix(z0: bigint): bigint {
60
+ let z = z0;
61
+ z = (z ^ (z >> 30n)) * SM_M1;
62
+ z &= MASK64;
63
+ z = (z ^ (z >> 27n)) * SM_M2;
64
+ z &= MASK64;
65
+ return z ^ (z >> 31n);
66
+ }
67
+
68
+ /** A splitmix64 PRNG yielding unsigned 64-bit values as BigInt. */
69
+ export type SplitMix64 = () => bigint;
70
+
71
+ /** Create a splitmix64 PRNG from a fixed unsigned 64-bit state. */
72
+ export function splitmix64(state: bigint = SEED_STATE_INIT): SplitMix64 {
73
+ let s = state & MASK64;
74
+ return () => {
75
+ s = (s + SM_GOLDEN) & MASK64;
76
+ return mix(s) & MASK64;
77
+ };
78
+ }
79
+
80
+ export interface MinHashSeedPair {
81
+ readonly a: bigint;
82
+ readonly b: bigint;
83
+ }
84
+
85
+ /**
86
+ * Derive the frozen 256 published (a_i, b_i) seed pairs. Each pair maps two
87
+ * splitmix64 draws into [1, p-1]. Deterministic and cross-language reproducible
88
+ * from SEED_STATE_INIT — mirrors `seeds-v2.json`.
89
+ *
90
+ * The table is frozen: computed ONCE at module load and reused for every call.
91
+ * Recomputing the 256 splitmix64 pairs (and re-walking all shingles) per
92
+ * signature/backfill row is wasteful for a large v1 index backfill, and the
93
+ * table is immutable by construction, so a single shared instance is safe and
94
+ * byte-identical. The returned sub-arrays share the frozen pair objects; they
95
+ * are themselves frozen so no caller can mutate the shared cache.
96
+ */
97
+ export function minhashV2Seeds(
98
+ count: number = NUM_HASHES_V2,
99
+ ): readonly MinHashSeedPair[] {
100
+ return FROZEN_SEEDS.slice(0, count);
101
+ }
102
+
103
+ /** Singleton seed table, computed exactly once and frozen (see minhashV2Seeds). */
104
+ const FROZEN_SEEDS: readonly MinHashSeedPair[] = (() => {
105
+ const prng = splitmix64();
106
+ const pairs: MinHashSeedPair[] = [];
107
+ for (let i = 0; i < NUM_HASHES_V2; i++) {
108
+ const a = (prng() % (P_V2 - 1n)) + 1n;
109
+ const b = (prng() % (P_V2 - 1n)) + 1n;
110
+ pairs.push(Object.freeze({ a, b }));
111
+ }
112
+ return Object.freeze(pairs);
113
+ })();
114
+
115
+ /** 64-bit FNV-1a over a UTF-8 string (for shingle hashing). */
116
+ function fnv1a64String(s: string): bigint {
117
+ let h = 0xcbf29ce484222325n;
118
+ const bytes = Buffer.from(s, "utf8");
119
+ for (let i = 0; i < bytes.length; i++) {
120
+ h ^= BigInt(bytes[i] ?? 0);
121
+ h = (h * 0x100000001b3n) & MASK64;
122
+ }
123
+ return h;
124
+ }
125
+
126
+ /**
127
+ * Five-code-point shingle set (deduped, capped) of normalized text as u64
128
+ * hashes. Iterates over code points (not UTF-16 code units) so astral-plane
129
+ * characters do not split a shingle.
130
+ */
131
+ export function shinglesV2(
132
+ text: string,
133
+ shingleCp: number = SHINGLE_CP,
134
+ maxShingles = 50_000,
135
+ ): bigint[] {
136
+ const norm = normalize(text);
137
+ if (norm.length === 0) return [];
138
+ const cps = Array.from(norm); // code points
139
+ const set = new Set<bigint>();
140
+ if (cps.length < shingleCp) {
141
+ set.add(fnv1a64String(norm));
142
+ } else {
143
+ for (let i = 0; i + shingleCp <= cps.length; i++) {
144
+ const sh = cps.slice(i, i + shingleCp).join("");
145
+ set.add(fnv1a64String(sh));
146
+ if (set.size >= maxShingles) break;
147
+ }
148
+ }
149
+ return [...set];
150
+ }
151
+
152
+ /**
153
+ * Compute the 256-slot MinHashV2 signature of a text as an array of exact u64
154
+ * BigInt values (each in [0, p], p = sentinel when empty). Uses EXACT BigInt
155
+ * multiply/modulo so high-bit products (a*x ~ up to 2^124) are never corrupted.
156
+ */
157
+ export function minhashV2Signature(text: string): bigint[] {
158
+ const grams = shinglesV2(text);
159
+ const seeds = minhashV2Seeds();
160
+ const sig = new Array<bigint>(NUM_HASHES_V2).fill(P_V2);
161
+ if (grams.length === 0) return sig;
162
+ for (let i = 0; i < NUM_HASHES_V2; i++) {
163
+ const { a, b } = seeds[i] as MinHashSeedPair;
164
+ let min = P_V2;
165
+ for (const x of grams) {
166
+ const h = (a * x + b) % P_V2;
167
+ if (h < min) min = h;
168
+ }
169
+ sig[i] = min;
170
+ }
171
+ return sig;
172
+ }
173
+
174
+ /** Encode a 256-slot signature to 2048 bytes as 256 x u64 little-endian. */
175
+ export function encodeSignatureV2(sig: readonly bigint[]): Uint8Array {
176
+ if (sig.length !== NUM_HASHES_V2) {
177
+ throw new Error(
178
+ `encodeSignatureV2: expected ${NUM_HASHES_V2} slots, got ${sig.length}`,
179
+ );
180
+ }
181
+ const buf = Buffer.allocUnsafe(SIGNATURE_BYTES_V2);
182
+ for (let i = 0; i < NUM_HASHES_V2; i++) {
183
+ buf.writeBigUInt64LE(sig[i] as bigint, i * U64_BYTES);
184
+ }
185
+ return new Uint8Array(buf);
186
+ }
187
+
188
+ /**
189
+ * Estimated Jaccard similarity of two v2 signatures (fraction of equal slots).
190
+ *
191
+ * THIS IS v2-ONLY. Cross-version compare is rejected INTERNALLY with
192
+ * `MINHASH_VERSION_MISMATCH`: both inputs must be exact 256-slot v2 signatures,
193
+ * i.e. every slot a `bigint` in [0, P_V2] (the empty sentinel is exactly P_V2).
194
+ * A v1 signature is a `number[]` of u31 values, so passing a v1 array fails the
195
+ * bigint type/range check and is rejected — a mixed v1/v2 compare can never
196
+ * produce a similarity number. Never call this with v1 arrays.
197
+ */
198
+ export function signatureSimilarityV2(a: readonly bigint[], b: readonly bigint[]): number {
199
+ assertV2Signature(a);
200
+ assertV2Signature(b);
201
+ const n = Math.min(a.length, b.length);
202
+ if (n === 0) return 0;
203
+ let equal = 0;
204
+ for (let i = 0; i < n; i++) if (a[i] === b[i]) equal++;
205
+ return equal / n;
206
+ }
207
+
208
+ /** Reject any array that is not an exact 256-slot v2 signature (bigint in [0, P_V2]). */
209
+ function assertV2Signature(sig: readonly bigint[]): void {
210
+ if (sig.length !== NUM_HASHES_V2) {
211
+ throw new Error(MINHASH_VERSION_MISMATCH_MSG);
212
+ }
213
+ for (const s of sig) {
214
+ if (typeof s !== "bigint" || s < 0n || s > P_V2) {
215
+ throw new Error(MINHASH_VERSION_MISMATCH_MSG);
216
+ }
217
+ }
218
+ }
219
+
220
+ /** Frozen cross-version-reject message shared by the v2-only function guards. */
221
+ const MINHASH_VERSION_MISMATCH_MSG = "MINHASH_VERSION_MISMATCH";
@@ -33,7 +33,7 @@ export function normalize(text: string): string {
33
33
  if (!text) return "";
34
34
  let out = stripAnsi(text);
35
35
  out = out.normalize("NFC");
36
- out = out.toLocaleLowerCase(); // case-fold so "Foo"/"FOO" collapse to one key
36
+ out = out.toLowerCase(); // case-fold so "Foo"/"FOO" collapse to one key (locale-independent: byte-exact across languages)
37
37
  out = out.replace(/\r\n?/g, "\n"); // CRLF / CR → LF
38
38
  out = out.replace(/\s+/g, " ").trim();
39
39
  if (out.length > MAX_CHARS) out = out.slice(0, MAX_CHARS);
@@ -15,7 +15,8 @@
15
15
  import { getStateDir } from "../../store.js";
16
16
  import { openStore } from "./utils.js";
17
17
  import type { SQLInputValue } from "node:sqlite";
18
- import { GameMetric, METRICS } from "../../game/scoring.js";
18
+ import type { GameMetric } from "../../game/scoring.js";
19
+ import { METRICS } from "../../game/scoring.js";
19
20
 
20
21
  // Process-local monotonic sequence appended into the low 3 digits of `ts` so
21
22
  // back-to-back inserts within the same millisecond get distinct PK values
@@ -0,0 +1,56 @@
1
+ /**
2
+ * vector-cortex/conformance/emit.ts — VC1C observability seam.
3
+ *
4
+ * Emits the three VC1C structured events, all gated on `MEGACOMPACT_VC1C`
5
+ * (mode C parity: flag OFF => zero emissions). Every event is a JSON line with
6
+ * `ts` + `event` (ENGINEERING_PRACTICES §8); the emitters below capture the
7
+ * event name + fields and are never fatal on consumer failure (non-fatal
8
+ * observability, never breaks the agent loop).
9
+ *
10
+ * vector_cortex_minhash_v2_backfilled — M4 backfill wrote v2 rows
11
+ * vector_cortex_conformance_case_checked — a v2 conformance case dispatched
12
+ * vector_cortex_downgrade_copy_written — a downgrade legacy copy written
13
+ *
14
+ * No network, no side effects beyond the supplied emit callback
15
+ * (PREVENT-PI-004 / PREVENT-011).
16
+ */
17
+
18
+ import { VC1C_ENABLED } from "../../config/vector-cortex.js";
19
+
20
+ export type ConformanceEmit = (event: string, fields: Record<string, unknown>) => void;
21
+
22
+ /** The three-event reporter surface consumed by the runtime seams. */
23
+ export interface ConformanceReporter {
24
+ readonly backfilled: (fields: Record<string, unknown>) => void;
25
+ readonly caseChecked: (fields: Record<string, unknown>) => void;
26
+ readonly downgradeWritten: (fields: Record<string, unknown>) => void;
27
+ }
28
+
29
+ /** A flag-gated no-op reporter (zero emissions, default when none injected). */
30
+ export const NOOP_CONFORMANCE_REPORTER: ConformanceReporter = {
31
+ backfilled: () => {},
32
+ caseChecked: () => {},
33
+ downgradeWritten: () => {},
34
+ };
35
+
36
+ /**
37
+ * Flag-gated emit: no-op when VC1C is off or no emitter is supplied. The
38
+ * returned reporter is itself flag-gated (`VC1C_ENABLED`), so wiring it into a
39
+ * runtime seam yields zero emissions when `MEGACOMPACT_VC1C=0` (byte-identical
40
+ * to the predecessor).
41
+ */
42
+ export function createConformanceReporter(emit?: ConformanceEmit): ConformanceReporter {
43
+ const fire = (event: string, fields: Record<string, unknown>): void => {
44
+ if (!VC1C_ENABLED()) return;
45
+ try {
46
+ emit?.(event, { ...fields, ts: new Date().toISOString() });
47
+ } catch {
48
+ /* non-fatal observability */
49
+ }
50
+ };
51
+ return {
52
+ backfilled: (fields) => fire("vector_cortex_minhash_v2_backfilled", fields),
53
+ caseChecked: (fields) => fire("vector_cortex_conformance_case_checked", fields),
54
+ downgradeWritten: (fields) => fire("vector_cortex_downgrade_copy_written", fields),
55
+ };
56
+ }
@@ -0,0 +1,298 @@
1
+ /**
2
+ * vector-cortex/conformance/manifest.ts — FixtureManifestV2 (VC1C).
3
+ *
4
+ * Owns the v2 conformance manifest: reads the authoritative
5
+ * `conformance/vector-cortex/v2/manifest.json` plus the on-disk fixture tree
6
+ * and validates it canonically. Each FixtureManifestV2 entry carries:
7
+ * id, domain, inputDigest, expectedDigest, failureCode, algorithmTuple.
8
+ *
9
+ * Canonical validation rejects a mismatched corpus with the frozen codes:
10
+ * CONF_EXTRA_FIXTURE — a file on disk that the manifest does not list;
11
+ * CONF_MISSING_FIXTURE — a manifest entry with no file on disk;
12
+ * CONF_DIGEST_DRIFT — on-disk bytes do not match the manifest SHA-256.
13
+ *
14
+ * Key ordering rule (canonical JSON): object keys are sorted by UTF-8 bytes;
15
+ * a manifest that is not canonical (or a fixture whose canonical re-serialization
16
+ * differs) is itself a digest-drift / noncanonical condition. The invariant:
17
+ * canonical valid manifests converge to ONE digest (the corpus is a single
18
+ * reproducible byte image).
19
+ *
20
+ * Pure FS reads + pure predicates; no network, no side effects on authority
21
+ * data (PREVENT-PI-004 / PREVENT-011).
22
+ */
23
+
24
+ import { readFileSync, readdirSync, statSync } from "node:fs";
25
+ import { createHash as sha256 } from "node:crypto";
26
+ import { join, relative, sep } from "node:path";
27
+
28
+ /** Frozen VC1C conformance failure codes (the "reject unknown / extra / drifted"). */
29
+ export const CONF_FAIL = {
30
+ EXTRA_FIXTURE: "CONF_EXTRA_FIXTURE",
31
+ MISSING_FIXTURE: "CONF_MISSING_FIXTURE",
32
+ DIGEST_DRIFT: "CONF_DIGEST_DRIFT",
33
+ NONCANONICAL: "CONF_NONCANONICAL",
34
+ UNKNOWN_DOMAIN: "CONF_UNKNOWN_DOMAIN",
35
+ } as const;
36
+ export type ConfFailureCode = (typeof CONF_FAIL)[keyof typeof CONF_FAIL];
37
+
38
+ /** A normalized FixtureManifestV2 entry. */
39
+ export interface FixtureManifestEntry {
40
+ readonly id: string;
41
+ readonly domain: string;
42
+ readonly path: string;
43
+ readonly algorithm: string;
44
+ readonly expected: string;
45
+ readonly inputDigest: string;
46
+ readonly expectedDigest: string;
47
+ readonly expectedOutputDigest?: string;
48
+ readonly failureCode?: string;
49
+ readonly algorithmTuple: readonly string[];
50
+ }
51
+
52
+ /** The parsed, normalized v2 manifest. */
53
+ export interface FixtureManifestV2 {
54
+ readonly version: string;
55
+ readonly owner: readonly string[];
56
+ readonly fixtureEntries: readonly FixtureManifestEntry[];
57
+ /** id -> entry index for O(1) lookup. */
58
+ readonly byId: ReadonlyMap<string, FixtureManifestEntry>;
59
+ }
60
+
61
+ /** Result of a canonical manifest validation. */
62
+ export type ManifestValidateResult =
63
+ | { ok: true; entryCount: number }
64
+ | { ok: false; codes: readonly ConfFailureCode[]; issues: readonly string[] };
65
+
66
+ /** Walk a directory tree returning relative POSIX paths (sorted). */
67
+ function walk(dir: string, base: string, acc: string[] = []): string[] {
68
+ for (const name of readdirSync(dir)) {
69
+ const p = join(dir, name);
70
+ const st = statSync(p);
71
+ if (st.isDirectory()) walk(p, base, acc);
72
+ else acc.push(relative(base, p).split(sep).join("/"));
73
+ }
74
+ return acc.sort();
75
+ }
76
+
77
+ /**
78
+ * Canonical JSON bytes for a value (UTF-8, NFC keys, sorted, shortest numbers).
79
+ * Matches the stricter `canonicalNumber` in scripts/vector-cortex-conformance.mjs:
80
+ * a non-finite number or -0 has no canonical shortest representation and is
81
+ * rejected, so the two "canonical" serializers can never diverge for the same
82
+ * value (thereby preserving the "converge to one digest" invariant).
83
+ */
84
+ function canonicalValue(value: unknown): string {
85
+ if (value === null || typeof value !== "object") {
86
+ if (typeof value === "number") {
87
+ if (!Number.isFinite(value) || Object.is(value, -0)) {
88
+ throw new Error(`non-canonical number: ${value}`);
89
+ }
90
+ if (Number.isInteger(value) && Number.isSafeInteger(value)) return String(value);
91
+ return JSON.stringify(value);
92
+ }
93
+ return JSON.stringify(value);
94
+ }
95
+ if (Array.isArray(value)) return `[${value.map((v) => canonicalValue(v)).join(",")}]`;
96
+ const keys = Object.keys(value as Record<string, unknown>)
97
+ .map((k) => k.normalize("NFC"))
98
+ .sort();
99
+ const parts = keys.map(
100
+ (k) => `${JSON.stringify(k)}:${canonicalValue((value as Record<string, unknown>)[k])}`,
101
+ );
102
+ return `{${parts.join(",")}}`;
103
+ }
104
+
105
+ function canonicalJson(value: unknown): Buffer {
106
+ return Buffer.from(`${canonicalValue(value)}\n`, "utf8");
107
+ }
108
+
109
+ function sha256Hex(bytes: Uint8Array): string {
110
+ return sha256("sha256").update(bytes).digest("hex");
111
+ }
112
+
113
+ /** parse a manifest.json (must be a valid v2 manifest object). */
114
+ function parseManifest(raw: string): {
115
+ version: string;
116
+ owner: string;
117
+ fixtures: Array<{
118
+ id: string;
119
+ path: string;
120
+ sha256: string;
121
+ algorithm: string;
122
+ expected: string;
123
+ domain?: string;
124
+ outputDigest?: string;
125
+ }>;
126
+ } {
127
+ const obj = JSON.parse(raw) as {
128
+ version?: unknown;
129
+ owner?: unknown;
130
+ fixtures?: unknown;
131
+ };
132
+ if (obj === null || typeof obj !== "object" || !Array.isArray(obj.fixtures)) {
133
+ throw new Error("CONF_MANIFEST_INVALID: manifest.fixtures must be an array");
134
+ }
135
+ return {
136
+ version: typeof obj.version === "string" ? obj.version : "2",
137
+ owner: typeof obj.owner === "string" ? obj.owner : "",
138
+ fixtures: obj.fixtures as Array<{
139
+ id: string;
140
+ path: string;
141
+ sha256: string;
142
+ algorithm: string;
143
+ expected: string;
144
+ domain?: string;
145
+ outputDigest?: string;
146
+ }>,
147
+ };
148
+ }
149
+
150
+ /**
151
+ * Read + normalize the v2 manifest and on-disk tree into a FixtureManifestV2.
152
+ * `fixtureRoot` is the conformance v2 root directory (injected by callers so
153
+ * `src/` stays runtime-independent of a hardcoded path).
154
+ */
155
+ export function readFixtureManifestV2(fixtureRoot: string): FixtureManifestV2 {
156
+ const manifestRaw = readFileSync(join(fixtureRoot, "manifest.json"), "utf8");
157
+ const manifest = parseManifest(manifestRaw);
158
+ const byId = new Map<string, FixtureManifestEntry>();
159
+ const entries: FixtureManifestEntry[] = [];
160
+ for (const fx of manifest.fixtures) {
161
+ const expectedDigest = String(fx.sha256 ?? "");
162
+ const domain = fx.domain ?? domainOf(fx.path);
163
+ const inputDigest = canonicalFileDigest(join(fixtureRoot, fx.path));
164
+ const entry: FixtureManifestEntry = {
165
+ id: fx.id,
166
+ domain,
167
+ path: fx.path,
168
+ algorithm: fx.algorithm,
169
+ expected: fx.expected,
170
+ inputDigest,
171
+ expectedDigest,
172
+ expectedOutputDigest:
173
+ typeof fx.outputDigest === "string" && fx.outputDigest.length > 0
174
+ ? fx.outputDigest
175
+ : undefined,
176
+ failureCode: fx.expected === "ok" ? undefined : fx.expected,
177
+ algorithmTuple: String(fx.algorithm).split(";").filter(Boolean).map((s) => s.trim()),
178
+ };
179
+ entries.push(entry);
180
+ byId.set(fx.id, entry);
181
+ }
182
+ return {
183
+ version: manifest.version,
184
+ owner: manifest.owner.split(",").map((s) => s.trim()).filter(Boolean),
185
+ fixtureEntries: entries,
186
+ byId,
187
+ };
188
+ }
189
+
190
+ /** The v2 domain a fixture path belongs to (first path segment). */
191
+ export function domainOf(path: string): string {
192
+ return path.split("/")[0] ?? "";
193
+ }
194
+
195
+ /**
196
+ * SHA-256 over the canonical JSON bytes of a fixture file. A missing file
197
+ * returns "" so `validateCanonicalV2` can report CONF_MISSING_FIXTURE instead of
198
+ * throwing ENOENT part-way through manifest normalization.
199
+ */
200
+ function canonicalFileDigest(absPath: string): string {
201
+ let raw: string;
202
+ try {
203
+ raw = readFileSync(absPath, "utf8");
204
+ } catch {
205
+ return "";
206
+ }
207
+ let parsed: unknown;
208
+ try {
209
+ parsed = JSON.parse(raw);
210
+ } catch {
211
+ parsed = raw; // non-JSON (e.g. seeds as bytes) — digest raw
212
+ }
213
+ return sha256Hex(canonicalJson(parsed));
214
+ }
215
+
216
+ /**
217
+ * Validate the on-disk v2 corpus against the manifest. Rejects an extra,
218
+ * missing, or digest-drifted fixture file with the frozen codes. Also verifies
219
+ * the manifest itself is canonical (sorted keys, shortest numbers) — a
220
+ * noncanonical manifest is reported as NONCANONICAL.
221
+ */
222
+ export function validateCanonicalV2(fixtureRoot: string): ManifestValidateResult {
223
+ const codes: ConfFailureCode[] = [];
224
+ const issues: string[] = [];
225
+ const manifestRaw = readFileSync(join(fixtureRoot, "manifest.json"), "utf8");
226
+ try {
227
+ const canonicalBytes = canonicalJson(JSON.parse(manifestRaw));
228
+ if (Buffer.from(manifestRaw, "utf8").toString("hex") !== canonicalBytes.toString("hex")) {
229
+ codes.push(CONF_FAIL.NONCANONICAL);
230
+ issues.push("manifest.json is not canonical");
231
+ }
232
+ } catch {
233
+ codes.push(CONF_FAIL.NONCANONICAL);
234
+ issues.push("manifest.json is not valid JSON");
235
+ }
236
+
237
+ const manifest = readFixtureManifestV2(fixtureRoot);
238
+ const listedPaths = new Set(manifest.fixtureEntries.map((e) => e.path));
239
+
240
+ // Extra-file detection: every file under the root (except manifest) listed.
241
+ const onDisk = walk(fixtureRoot, fixtureRoot).filter((p) => p !== "manifest.json");
242
+ for (const p of onDisk) {
243
+ if (!listedPaths.has(p)) {
244
+ codes.push(CONF_FAIL.EXTRA_FIXTURE);
245
+ issues.push(`extra fixture: ${p}`);
246
+ }
247
+ }
248
+
249
+ // Missing + digest-drift for every listed fixture.
250
+ for (const entry of manifest.fixtureEntries) {
251
+ const abs = join(fixtureRoot, entry.path);
252
+ if (!exists(abs)) {
253
+ codes.push(CONF_FAIL.MISSING_FIXTURE);
254
+ issues.push(`missing fixture: ${entry.path}`);
255
+ continue;
256
+ }
257
+ const raw = readFileSync(abs, "utf8");
258
+ let canonicalHex: string;
259
+ try {
260
+ canonicalHex = sha256Hex(canonicalJson(JSON.parse(raw)));
261
+ } catch {
262
+ canonicalHex = sha256Hex(Buffer.from(raw, "utf8"));
263
+ }
264
+ if (canonicalHex !== entry.expectedDigest) {
265
+ codes.push(CONF_FAIL.DIGEST_DRIFT);
266
+ issues.push(`digest drift for ${entry.path}: expected ${entry.expectedDigest} got ${canonicalHex}`);
267
+ }
268
+ }
269
+
270
+ // The canonical-key ordering invariant: a canonical valid manifest is a single
271
+ // reproducible digest. (Extra-file presence also breaks purity but is the
272
+ // extra/missing covers it.) Noncanonical keys are caught above.
273
+ if (codes.length === 0) {
274
+ return { ok: true, entryCount: manifest.fixtureEntries.length };
275
+ }
276
+ return { ok: false, codes: dedupeCodes(codes), issues };
277
+ }
278
+
279
+ /** True if the manifest+corpus is canonical (converges to one digest). */
280
+ export function canonicalManifestsConverge(fixtureRoot: string): boolean {
281
+ const v = validateCanonicalV2(fixtureRoot);
282
+ return v.ok;
283
+ }
284
+
285
+ function exists(p: string): boolean {
286
+ try {
287
+ statSync(p);
288
+ return true;
289
+ } catch {
290
+ return false;
291
+ }
292
+ }
293
+
294
+ function dedupeCodes(codes: ConfFailureCode[]): ConfFailureCode[] {
295
+ const out: ConfFailureCode[] = [];
296
+ for (const c of codes) if (!out.includes(c)) out.push(c);
297
+ return out;
298
+ }