Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 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 | 5x 5x 5x 5x 5x 5x 506x 506x 506x 434x 506x 222x 58x 202x 57x 57x 32x 122x 31x 30x 30x 183x 181x 181x 181x 180x 180x 180x 167x 263x 163x 162x 1x 49x 49x 49x 49x 89x 89x 89x 89x 89x 89x 89x 89x 89x 89x 26x 26x 89x 89x 92x 92x 92x 95x 95x 95x 302x 300x 95x 302x 300x 95x 62x 62x 62x 62x 177x 177x 49x 49x 49x 177x 62x 62x 62x 26x 26x 26x 26x 62x 203x 203x 132x 132x 112x 57x 57x 57x 1x 56x 56x 2x 54x 54x 54x 54x 54x 54x 54x 54x 54x 54x 54x 54x 54x 40x 40x 38x 2x 1x 47x 47x | /**
* Lifecycle Statistics — corpus-wide dwell distributions for EP procedures.
*
* Builds an in-memory model of how long procedures historically dwell at each
* lifecycle stage, indexed by `(procedureType, stage)`. Two distributions are
* derived per key:
*
* - `dwellAtStage` — days between an event and the next event observed for the
* same procedure. Used to detect bottlenecks: a current procedure whose
* "days in current stage" exceeds the historical 95th percentile is flagged
* as a real stuck procedure rather than a heuristic guess.
* - `remainingToCompletion` — days between an event at the given stage and the
* final event of the same procedure (used as the completion timestamp).
* Used as the forecasting baseline for `estimatedCompletionDays`.
*
* The model is corpus-wide (latest N=500 procedures), cached for 30 minutes,
* and computed deterministically (median + percentile use sorted inputs with
* stable tie-breaking).
*
* **Why this exists:** `monitor_legislative_pipeline` previously approximated
* `daysInCurrentStage`, `bottleneckRisk`, and `estimatedCompletionDays` from
* procedure metadata alone. With the real `/procedures/{id}/events` timeline
* available, we can build verifiable, distribution-based metrics instead of
* heuristics.
*
* **Security:**
* - No PII is stored — only event types, dates, and procedure types.
* - Audit logging captures corpus size and computation time only (see
* `monitorLegislativePipeline.ts`).
*
* ISMS Policy: SC-002 (Input Validation), AC-003 (Least Privilege)
*
* @see https://github.com/Hack23/ISMS-PUBLIC/blob/main/Secure_Development_Policy.md
*/
import { epClient } from '../clients/europeanParliamentClient.js';
import type { EPEvent, Procedure } from '../types/europeanParliament.js';
/** Number of latest procedures to sample when building the dwell distribution. */
export const CORPUS_SIZE = 500;
/** Cache TTL for the corpus-wide lifecycle model (30 minutes). */
export const CACHE_TTL_MS = 30 * 60 * 1000;
/** Bounded concurrency for per-procedure event fetches. */
export const EVENT_FETCH_CONCURRENCY = 8;
/** Maximum events to fetch per procedure when building the corpus. */
const EVENTS_PER_PROCEDURE = 50;
/**
* Statistics for a single `(procedureType, stage)` pair.
*
* All numbers are in days. `sampleSize` exposes how many observations the
* statistics were derived from — callers should treat low-sample cells as
* less reliable and may degrade `forecastBasis` to `INSUFFICIENT_DATA`.
*/
export interface StageDwellStatistics {
/** Median number of days a procedure dwells at this stage before moving on. */
medianDwellDays: number;
/** 95th-percentile dwell — exceeding this is the bottleneck threshold. */
p95DwellDays: number;
/**
* Median number of days remaining from this stage until the final event of
* the same procedure. Used as the forecasting baseline.
*/
medianRemainingDays: number;
/** Number of observations contributing to the dwell distribution. */
sampleSize: number;
}
/**
* Full lifecycle model built from a corpus of procedures.
*/
export interface LifecycleStatisticsModel {
/** Keyed by `${procedureType}::${stage}` — deterministic ordering. */
byTypeAndStage: ReadonlyMap<string, StageDwellStatistics>;
/** Total number of procedures inspected (after filtering for ≥2 events). */
corpusSize: number;
/** Total number of dwell observations across all keys. */
totalObservations: number;
/** Wall-clock duration of the corpus build, in milliseconds. */
computationTimeMs: number;
/** Timestamp the model was built (epoch ms). */
builtAt: number;
}
interface MutableStageBuckets {
dwellSamples: number[];
remainingSamples: number[];
}
interface CacheEntry {
model: LifecycleStatisticsModel;
expiresAt: number;
}
// Cache + in-flight builds are keyed by `corpusSize` so that callers
// requesting different sample sizes do not silently observe each other's
// model. The map is bounded to one entry per distinct size used and is
// expected to stay tiny in practice (production callers use the default).
const memoCacheByCorpusSize = new Map<number, CacheEntry>();
const inFlightBuildByCorpusSize = new Map<number, Promise<LifecycleStatisticsModel>>();
/**
* Normalize an EP event type to a stable stage key.
*
* The EP API sometimes returns event types as short codes (`REFERRAL`,
* `COM_VOTE`) and sometimes as URI strings (`def/ep-activities/REFERRAL`).
* This helper strips any URI prefix and uppercases the result so both forms
* map to the same stage in the dwell distribution.
*
* @param type - Raw event type from the EP API
* @returns Normalized stage key, or empty string when unavailable
*/
export function normalizeStageKey(type: string): string {
Iif (typeof type !== 'string') return '';
const trimmed = type.trim();
if (trimmed === '') return '';
const lastSegment = trimmed.includes('/') ? trimmed.slice(trimmed.lastIndexOf('/') + 1) : trimmed;
return lastSegment.toUpperCase();
}
/**
* Build a `(procedureType, stage)` map key.
*
* @param procedureType - The EP procedure type (e.g. `COD`, `NLE`)
* @param stage - Normalized stage key
*/
export function lifecycleKey(procedureType: string, stage: string): string {
return `${procedureType}::${stage}`;
}
/**
* Compute the lower-median of a numeric array.
*
* Deterministic: input is sorted ascending with the stable JS sort.
* For even-length arrays the lower of the two middle elements is returned —
* this avoids floating-point ties that depend on iteration order.
*
* @param values - Numeric samples (will be copied, not mutated)
* @returns Lower median, or 0 for empty inputs
*/
export function median(values: readonly number[]): number {
if (values.length === 0) return 0;
const sorted = [...values].sort((a, b) => a - b);
const mid = Math.floor((sorted.length - 1) / 2);
return sorted[mid] ?? 0;
}
/**
* Compute a percentile using nearest-rank with `Math.ceil`.
*
* Deterministic and tie-stable: identical fixture sets produce identical
* percentiles regardless of insertion order. Inputs are always sorted before
* the percentile rank is read so the return value is consistent for all
* inputs in the documented range.
*
* @param values - Numeric samples
* @param percentile - Percentile in the range [0, 100]. Values ≤ 0 return
* the minimum (sorted[0]); values ≥ 100 return the maximum.
* @returns The percentile value, or 0 for empty inputs
*/
export function percentile(values: readonly number[], percentile: number): number {
if (values.length === 0) return 0;
const sorted = [...values].sort((a, b) => a - b);
if (percentile <= 0) return sorted[0] ?? 0;
const rank = Math.min(sorted.length, Math.max(1, Math.ceil((percentile / 100) * sorted.length)));
return sorted[rank - 1] ?? 0;
}
/**
* Compute integer days between two ISO date strings, clamped to ≥0.
*
* @param fromIso - Earlier ISO date string
* @param toIso - Later ISO date string
* @returns Whole-day delta, or 0 when either date is invalid
*/
export function daysBetween(fromIso: string, toIso: string): number {
if (fromIso === '' || toIso === '') return 0;
const start = new Date(fromIso);
const end = new Date(toIso);
if (isNaN(start.getTime()) || isNaN(end.getTime())) return 0;
// Use UTC day boundaries to avoid DST/time-of-day shifts
const startUtcDays = Math.floor(start.getTime() / (1000 * 60 * 60 * 24));
const endUtcDays = Math.floor(end.getTime() / (1000 * 60 * 60 * 24));
return Math.max(0, endUtcDays - startUtcDays);
}
/**
* Sort events chronologically by date, with stable tie-break on event id.
*
* @param events - Raw events from `/procedures/{id}/events`
* @returns A new array sorted in ascending date order
*/
export function sortEventsChronologically(events: readonly EPEvent[]): EPEvent[] {
return [...events]
.filter((e) => e.date !== '' && !isNaN(new Date(e.date).getTime()))
.sort((a, b) => {
if (a.date < b.date) return -1;
if (a.date > b.date) return 1;
Eif (a.id < b.id) return -1;
if (a.id > b.id) return 1;
return 0;
});
}
/**
* Extract dwell-at-stage and remaining-to-completion observations from a
* single procedure's chronologically-sorted event list.
*
* @param procedureType - The EP procedure type (e.g. `COD`)
* @param sortedEvents - Events sorted via {@link sortEventsChronologically}
* @param buckets - Mutable accumulator keyed by `${type}::${stage}`
*/
function accumulateProcedure(
procedureType: string,
sortedEvents: readonly EPEvent[],
buckets: Map<string, MutableStageBuckets>
): void {
Iif (sortedEvents.length < 2) return;
const lastEvent = sortedEvents[sortedEvents.length - 1];
Iif (lastEvent === undefined) return;
for (let i = 0; i < sortedEvents.length - 1; i++) {
const current = sortedEvents[i];
const next = sortedEvents[i + 1];
Iif (current === undefined || next === undefined) continue;
const stage = normalizeStageKey(current.type);
Iif (stage === '') continue;
const dwell = daysBetween(current.date, next.date);
const remaining = daysBetween(current.date, lastEvent.date);
const key = lifecycleKey(procedureType, stage);
let bucket = buckets.get(key);
if (bucket === undefined) {
bucket = { dwellSamples: [], remainingSamples: [] };
buckets.set(key, bucket);
}
bucket.dwellSamples.push(dwell);
bucket.remainingSamples.push(remaining);
}
}
/**
* Fetch event sequences for a list of procedures with bounded concurrency.
*
* Uses `Promise.allSettled` so a single 404 or transient error does not abort
* the whole corpus build — failures are silently skipped (their contribution
* to the distribution is simply omitted).
*
* @param procedures - Procedures to enrich with their event timeline
* @param concurrency - Maximum parallel fetches (≤8 recommended)
* @param deadline - Optional wall-clock deadline (epoch ms). If `Date.now()`
* exceeds the deadline between batches, the loop short-circuits and the
* partial result is returned. This is the corpus rebuild's cooperative
* cancellation mechanism: when the request-path budget fires, this
* function stops queueing further events fetches so it no longer competes
* with the request's own rate-limited calls.
* @returns Map of process-id → events array (failed fetches are absent;
* procedures whose batch never ran due to the deadline are also absent)
*/
export async function fetchEventsBounded(
procedures: readonly Procedure[],
concurrency: number = EVENT_FETCH_CONCURRENCY,
deadline?: number
): Promise<Map<string, EPEvent[]>> {
const results = new Map<string, EPEvent[]>();
const bound = Math.max(1, Math.min(concurrency, EVENT_FETCH_CONCURRENCY));
for (let i = 0; i < procedures.length; i += bound) {
Iif (deadline !== undefined && Date.now() >= deadline) break;
const slice = procedures.slice(i, i + bound);
const settled = await Promise.allSettled(
slice.map(async (proc) => {
const events = await epClient.getProcedureEvents(proc.id, { limit: EVENTS_PER_PROCEDURE });
return { id: proc.id, events: events.data };
})
);
for (const r of settled) {
if (r.status === 'fulfilled') {
results.set(r.value.id, r.value.events);
}
}
}
return results;
}
/**
* Build the lifecycle statistics model from a procedure corpus and their
* event timelines.
*
* Deterministic: identical input procedures and events produce identical
* statistics (the model is independent of `Map` insertion order — only the
* `byTypeAndStage` map's `Map`/`Object.entries` iteration order may differ,
* but the per-key values are stable).
*
* @param procedures - The corpus of procedures
* @param eventsByProcedureId - Event timelines fetched via {@link fetchEventsBounded}
* @returns Frozen lifecycle statistics model
*/
export function buildLifecycleStatistics(
procedures: readonly Procedure[],
eventsByProcedureId: ReadonlyMap<string, EPEvent[]>
): LifecycleStatisticsModel {
const start = Date.now();
const buckets = new Map<string, MutableStageBuckets>();
let usableProcedures = 0;
for (const proc of procedures) {
const events = eventsByProcedureId.get(proc.id);
if (events === undefined || events.length < 2) continue;
const sorted = sortEventsChronologically(events);
Iif (sorted.length < 2) continue;
accumulateProcedure(proc.type !== '' ? proc.type : 'UNKNOWN', sorted, buckets);
usableProcedures++;
}
const byTypeAndStage = new Map<string, StageDwellStatistics>();
let totalObservations = 0;
// Iterate keys in sorted order so the resulting map's insertion order is
// deterministic, regardless of how procedures were processed.
for (const key of [...buckets.keys()].sort()) {
const bucket = buckets.get(key);
Iif (bucket === undefined) continue;
totalObservations += bucket.dwellSamples.length;
byTypeAndStage.set(key, {
medianDwellDays: median(bucket.dwellSamples),
p95DwellDays: percentile(bucket.dwellSamples, 95),
medianRemainingDays: median(bucket.remainingSamples),
sampleSize: bucket.dwellSamples.length,
});
}
return {
byTypeAndStage,
corpusSize: usableProcedures,
totalObservations,
computationTimeMs: Date.now() - start,
builtAt: Date.now(),
};
}
/**
* Lookup statistics for a `(procedureType, stage)` pair.
*
* Performs an exact lookup by `lifecycleKey(procedureType, normalizedStage)`.
* Returns `undefined` when no matching cell exists or the cell has zero
* samples, so callers must handle the missing-data case explicitly.
*
* @param model - The lifecycle statistics model
* @param procedureType - The procedure type (e.g. `COD`)
* @param stage - Normalized stage key
* @returns Stage statistics, or `undefined` when no usable data exists
*/
export function lookupStageStatistics(
model: LifecycleStatisticsModel,
procedureType: string,
stage: string
): StageDwellStatistics | undefined {
const normalizedStage = normalizeStageKey(stage);
if (normalizedStage === '' || procedureType === '') return undefined;
const exact = model.byTypeAndStage.get(lifecycleKey(procedureType, normalizedStage));
if (exact !== undefined && exact.sampleSize > 0) return exact;
return undefined;
}
/**
* Build (or return cached) lifecycle statistics for the latest N procedures.
*
* The corpus is cached for {@link CACHE_TTL_MS}; subsequent callers within
* the window receive the same model instance for free. Refreshing is lazy:
* a stale cache is rebuilt on the next call rather than on a timer, so
* idle processes pay no background cost. Cache and in-flight builds are
* keyed by `corpusSize` so distinct sample sizes never share a model.
*
* @param options - Optional overrides
* @param options.corpusSize - Number of procedures to sample (default: {@link CORPUS_SIZE})
* @param options.forceRefresh - Ignore cached model and rebuild
* @param options.deadline - Optional wall-clock deadline (epoch ms) for the
* underlying {@link fetchEventsBounded} loop. When set, the rebuild stops
* queueing additional events fetches once the deadline is reached and
* builds the model from whatever was gathered (possibly empty). This is
* how the request path keeps a cold-cache rebuild from starving its own
* rate-limit budget. Concurrent callers that omit `deadline` still share
* the same in-flight build, so the budget set by the first caller wins.
* @returns The lifecycle statistics model (possibly partial when `deadline`
* fires before the full corpus has been fetched)
*
* @security The corpus contains only procedure types, event types, and dates —
* no PII. Bounded concurrency (≤8) prevents API rate-limit exhaustion.
* @since 0.8.0
*/
export async function getLifecycleStatistics(options: {
corpusSize?: number;
forceRefresh?: boolean;
deadline?: number;
} = {}): Promise<LifecycleStatisticsModel> {
const corpus = options.corpusSize ?? CORPUS_SIZE;
const cached = memoCacheByCorpusSize.get(corpus);
if (options.forceRefresh !== true && cached !== undefined && cached.expiresAt > Date.now()) {
return cached.model;
}
// Concurrency-safe rebuild: share an in-flight promise per corpusSize so
// simultaneous callers requesting the same sample size don't each trigger
// a duplicate corpus-wide fetch. Different corpusSize values get their own
// independent build so a small ad-hoc build cannot pollute the default
// production model.
const inFlight = inFlightBuildByCorpusSize.get(corpus);
if (inFlight !== undefined && options.forceRefresh !== true) {
return inFlight;
}
const deadline = options.deadline;
const buildPromise = (async (): Promise<LifecycleStatisticsModel> => {
try {
const procResp = await epClient.getProcedures({ limit: corpus });
const events = await fetchEventsBounded(procResp.data, EVENT_FETCH_CONCURRENCY, deadline);
const model = buildLifecycleStatistics(procResp.data, events);
// Only cache full builds. Partial builds (deadline-truncated) are
// returned to the caller but not memoized, so subsequent requests
// retry the rebuild instead of being stuck with a sparse model.
const isPartial = deadline !== undefined && events.size < procResp.data.length;
Eif (!isPartial) {
memoCacheByCorpusSize.set(corpus, { model, expiresAt: Date.now() + CACHE_TTL_MS });
}
return model;
} finally {
inFlightBuildByCorpusSize.delete(corpus);
}
})();
inFlightBuildByCorpusSize.set(corpus, buildPromise);
return buildPromise;
}
/**
* Returns the cached lifecycle-statistics model for the given corpus size if
* present and unexpired, otherwise `null`. Never triggers a rebuild and
* never blocks.
*
* Intended for request-path callers (e.g. `monitor_legislative_pipeline`) that
* need to avoid the cold-cache corpus rebuild competing with their own
* rate-limited `/events` fan-out. Out-of-band warmup (background jobs,
* sibling tools whose primary purpose is the corpus itself) should keep
* using {@link getLifecycleStatistics} which will rebuild on miss.
*
* @param corpusSize - Sample size to look up (default: {@link CORPUS_SIZE}).
* @returns The cached model, or `null` on cache miss / expiry.
*
* @security No network calls; safe to use inside tight request budgets.
* @since 0.8.0
*/
export function getCachedLifecycleStatistics(
corpusSize: number = CORPUS_SIZE,
): LifecycleStatisticsModel | null {
const cached = memoCacheByCorpusSize.get(corpusSize);
if (cached !== undefined && cached.expiresAt > Date.now()) {
return cached.model;
}
return null;
}
/**
* An empty lifecycle model that callers can use as a fast fallback when the
* corpus rebuild fails or exceeds its time budget. With this model every
* lookup returns `undefined` so forecasts gracefully degrade to the
* `INSUFFICIENT_DATA` heuristic basis without aborting the request.
*/
export function emptyLifecycleStatisticsModel(): LifecycleStatisticsModel {
return {
byTypeAndStage: new Map(),
corpusSize: 0,
totalObservations: 0,
computationTimeMs: 0,
builtAt: Date.now(),
};
}
/**
* Reset the in-memory cache. Intended for tests and forced refreshes.
*/
export function resetLifecycleStatisticsCache(): void {
memoCacheByCorpusSize.clear();
inFlightBuildByCorpusSize.clear();
}
|