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 | /**
* @module ai-analysis/types
* @description TypeScript interfaces for the AI-first analysis pipeline.
*
* The AI analysis pipeline produces structured intelligence assessments from
* parliamentary documents through multiple refinement iterations:
* 1. Initial analysis — document classification, entity extraction, domain detection
* 2. Refinement — SWOT enrichment using full-text content from enriched documents
* 3. Validation — stakeholder completeness and confidence scoring (deep mode only)
*
* Templates in `generators.ts` consume `AnalysisResult` to render HTML sections;
* analytical text originates from the analysis pipeline with placeholder fallbacks
* when document evidence is missing for a stakeholder × quadrant combination.
*
* @author Hack23 AB
* @license Apache-2.0
*/
import type { Language } from '../types/language.js';
import type { RawDocument } from '../data-transformers/types.js';
// ---------------------------------------------------------------------------
// Analysis depth
// ---------------------------------------------------------------------------
/**
* Controls how many AI analysis iterations run.
*
* - `quick` — 1 pass: initial document analysis only (fast, for breaking/time-sensitive)
* - `standard` — 2 passes: initial analysis + SWOT refinement (default for most article types)
* - `deep` — 3 passes: initial + refinement + stakeholder validation (deep-inspection)
*/
export type AnalysisDepth = 'quick' | 'standard' | 'deep';
// ---------------------------------------------------------------------------
// Core analysis data structures
// ---------------------------------------------------------------------------
/**
* A single SWOT entry backed by evidence extracted from actual document content.
* Impact and confidence are derived analytically, not from hardcoded defaults.
*/
export interface AnalysisSwotEntry {
/** The analytical claim derived from document content. */
text: string;
/** Evidence rating for this entry. */
impact: 'high' | 'medium' | 'low';
/**
* Source document identifiers that back this entry.
* Empty when the entry is a structural placeholder.
*/
sourceDocIds: string[];
/**
* Confidence rating for this entry: proportion of supporting evidence
* relative to total documents in the set.
*/
confidence: 'HIGH' | 'MEDIUM' | 'LOW';
}
/** The three well-known stakeholder roles used by the analysis pipeline. */
export type KnownStakeholderRole = 'government' | 'parliament' | 'private-sector';
/** The four SWOT quadrant keys. */
export type SwotQuadrant = 'strengths' | 'weaknesses' | 'opportunities' | 'threats';
/**
* A SWOT matrix for a single stakeholder perspective.
* All text fields are AI-derived from document content analysis.
*/
export interface AnalysisStakeholderSwot {
/** Localised stakeholder name (Government, Parliament, Private Sector, etc.) */
name: string;
/** Internal role key for cross-language consistency */
role: KnownStakeholderRole | (string & {});
swot: {
strengths: AnalysisSwotEntry[];
weaknesses: AnalysisSwotEntry[];
opportunities: AnalysisSwotEntry[];
threats: AnalysisSwotEntry[];
};
}
/**
* A single watch point derived from document analysis.
* Replaces hardcoded watch-point templates.
*
* All string fields (`title`, `description`) contain **plain text** — they
* must be escaped via `escapeHtml()` at the call site before passing to
* `generateWatchSection()`, which renders pre-escaped HTML (it does not
* escape itself). The escaping call site is in `generators.ts` inside the
* deep-inspection path. This is consistent with the escaping contract for
* all other pipeline outputs (SWOT entries, mindmap items, dashboard labels)
* where renderers receive plain text and escape at interpolation time.
*/
export interface AnalysisWatchPoint {
/** Plain-text watch point heading (must be escaped before passing to generateWatchSection) */
title: string;
/** Plain-text watch point body (must be escaped before passing to generateWatchSection) */
description: string;
/** Urgency level for display sorting */
urgency: 'critical' | 'high' | 'medium' | 'low';
/** Document IDs that triggered this watch point */
sourceDocIds: string[];
}
/**
* A branch in the conceptual mindmap derived from document analysis.
*/
export interface AnalysisMindmapBranch {
label: string;
color: 'cyan' | 'green' | 'yellow' | 'purple' | 'orange' | 'magenta';
icon: string;
items: string[];
}
/**
* Policy domain assessment derived from document analysis.
*/
export interface PolicyAssessment {
/** Detected policy domains (localised) */
domains: string[];
/** Primary domain focus */
primaryDomain: string | null;
/**
* Brief narrative assessment of the policy landscape.
* Derived from document content, not templates.
*/
narrative: string;
/** Overall confidence in the assessment */
confidence: 'HIGH' | 'MEDIUM' | 'LOW';
}
/**
* Structured dashboard data derived from document analysis.
*/
export interface DashboardData {
/** Title for the dashboard section */
title: string;
/** Summary text */
summary: string;
/** Chart label-value pairs for document type distribution */
typeDistribution: Array<{ label: string; value: number; color: string }>;
}
// ---------------------------------------------------------------------------
// Analysis result
// ---------------------------------------------------------------------------
/**
* The complete output of the AI analysis pipeline for a given document set.
* This is the single source of truth for all analytical content in the article.
*/
export interface AnalysisResult {
/** Stakeholder SWOT analyses — all text derived from document content */
stakeholderSwot: AnalysisStakeholderSwot[];
/** Policy domain assessment */
policyAssessment: PolicyAssessment;
/** Mindmap branches derived from analysis */
mindmapBranches: AnalysisMindmapBranch[];
/** Dashboard data for visualisation */
dashboardData: DashboardData;
/** Watch points for the article sidebar */
watchPoints: AnalysisWatchPoint[];
/**
* Overall confidence score (0-100) for the analysis.
* Higher values indicate more enriched documents and stronger evidence.
*/
confidenceScore: number;
/** Number of analysis iterations completed */
iterationsCompleted: number;
/** ISO timestamp of when analysis was completed */
completedAt: string;
/** Language of the analysis */
lang: Language;
/** Number of documents analysed */
documentCount: number;
/** Number of metadata-enriched documents (contentFetched = true).
* This counts documents whose metadata was fetched via enrichDocumentsWithContent(),
* regardless of whether full text is available. */
enrichedCount: number;
/** Focus topic (null if multi-document analysis without topic) */
focusTopic: string | null;
}
// ---------------------------------------------------------------------------
// Validation
// ---------------------------------------------------------------------------
/**
* Result of validating an analysis for completeness and depth.
*/
export interface ValidationResult {
/** Whether the analysis meets the minimum quality threshold */
passed: boolean;
/** Quality score (0-100) */
score: number;
/** Specific issues found during validation */
issues: string[];
/** Suggested improvements (informational, not blocking) */
suggestions: string[];
}
// ---------------------------------------------------------------------------
// Pipeline interface
// ---------------------------------------------------------------------------
/**
* Options controlling the analysis pipeline execution.
*/
export interface AnalysisPipelineOptions {
/** How many AI iterations to run (default: 'standard') */
depth: AnalysisDepth;
/** Language for localised analysis output */
lang: Language;
/** Optional focus topic — analysis is exclusively scoped to this topic when provided */
focusTopic: string | null;
}
/**
* The AI analysis pipeline contract.
*
* Lifecycle for a `deep` analysis:
* 1. `analyzeDocuments` — iteration 1: initial extraction from document metadata + content
* 2. `refineAnalysis` — iteration 2: enrich SWOT entries using full document text
* 3. `validateCompleteness` — iteration 3: check stakeholder coverage and confidence
*
* For `standard` depth, only iterations 1-2 run.
* For `quick` depth, only iteration 1 runs.
*/
export interface AnalysisPipeline {
/**
* Iteration 1: Analyse documents and produce an initial AnalysisResult.
* Extracts document types, detected policy domains, entities, and
* assembles initial SWOT entries from document titles and available metadata.
*/
analyzeDocuments(
docs: RawDocument[],
options: AnalysisPipelineOptions,
): Promise<AnalysisResult>;
/**
* Iteration 2: Refine the initial analysis using full document text.
* Replaces generic SWOT entries with content-derived insights when
* enriched document text is available. Adds additional watch points
* and strengthens confidence scores.
*
* When no documents are enriched, the analysis content (SWOT, watch
* points, etc.) remains unchanged but metadata fields
* (`iterationsCompleted`, `completedAt`) are always updated.
*/
refineAnalysis(
initial: AnalysisResult,
docs: RawDocument[],
options: AnalysisPipelineOptions,
): Promise<AnalysisResult>;
/**
* Iteration 3: Validate completeness of the analysis.
* Checks that all three stakeholder perspectives have substantive entries,
* that policy domains are correctly identified, and that confidence
* thresholds are met.
*
* Returns a ValidationResult with a score and improvement suggestions.
* Does not modify the AnalysisResult — call `refineAnalysis` again if
* the score is below threshold.
*/
validateCompleteness(
analysis: AnalysisResult,
docs: RawDocument[],
): Promise<ValidationResult>;
}
// ---------------------------------------------------------------------------
// Iteration metadata (written to news/metadata/ for audit trail)
// ---------------------------------------------------------------------------
/**
* Metadata record written after each analysis run for audit and debugging.
* Stored as `news/metadata/ai-analysis-<slug>-<lang>.json`.
*/
export interface AnalysisIterationMetadata {
/** Article slug this analysis was produced for */
articleSlug: string;
/** Language of the analysis */
lang: Language;
/** Analysis depth used */
depth: AnalysisDepth;
/** Number of iterations completed */
iterationsCompleted: number;
/** Per-iteration duration in milliseconds */
iterationDurationsMs: number[];
/** Final confidence score */
confidenceScore: number;
/** Validation result (null when depth < 'deep') */
validationResult: ValidationResult | null;
/** Number of documents analysed */
documentCount: number;
/** Number of metadata-enriched documents (contentFetched = true) */
enrichedCount: number;
/** Focus topic (null if none) */
focusTopic: string | null;
/** ISO timestamp when analysis completed */
completedAt: string;
}
// ---------------------------------------------------------------------------
// Coalition tension detection
// ---------------------------------------------------------------------------
/** Stress level derived from document-type patterns within the analysed set. */
export type CoalitionStressLevel = 'low' | 'medium' | 'high';
/**
* Result of coalition tension detection from a document set.
* Summarises coalition pressure based on challenge-document ratios and
* the balance between government- and opposition-aligned documents.
*
* The current detector infers potential coalition tension by looking at the balance
* between government-initiated documents (e.g. propositions, SFS, government
* communications) and challenge-type documents (e.g. motions, interpellations).
* Committee reports are treated as neutral and excluded from both counts.
* A higher share of challenge documents is treated as a proxy for increased
* political stress.
*
* Note: this analysis does NOT use party metadata or model detailed convergence /
* divergence patterns between individual parties; it is purely document-type-based.
*/
export interface CoalitionTensionResult {
/** Overall coalition stress level derived from the document-type mix. */
stressLevel: CoalitionStressLevel;
/** Human-readable narrative explaining the detected tension heuristic. */
narrative: string;
/**
* Number of government-initiated documents in the set
* (e.g. propositions, SFS, government communications, ds, sou, dir).
* Committee reports (bet) are excluded as neutral.
*/
governmentDocCount: number;
/**
* Number of challenge-type documents in the set
* (e.g. motions, interpellations), used as a proxy for opposition pressure.
*/
oppositionDocCount: number;
/**
* Ratio of challenge-type documents to all classified documents
* (government + opposition; committee reports excluded from denominator).
*/
challengeRatio: number;
/** Document IDs supporting the tension assessment. */
sourceDocIds: string[];
}
// ---------------------------------------------------------------------------
// LLM-ready analysis interfaces
// ---------------------------------------------------------------------------
/**
* Interface for SWOT entry generation — heuristic or LLM-backed.
* Swap the implementation to integrate an LLM without changing callers.
*/
export interface SwotEntryGenerator {
/** Build an evidence-backed SWOT entry from a single document. */
buildEntry(
doc: RawDocument,
topic: string | null,
lang: Language,
passageMaxChars: number,
): AnalysisSwotEntry;
/** Build a structural placeholder when no documents exist for a quadrant. */
buildPlaceholder(
role: KnownStakeholderRole,
quadrant: SwotQuadrant,
topic: string | null,
lang: Language,
domains: string[],
): AnalysisSwotEntry;
}
/**
* Interface for confidence scoring — heuristic or LLM-calibrated.
*/
export interface ConfidenceScorer {
/** Calculate overall confidence from document evidence and SWOT quality. */
calculateScore(
docs: RawDocument[],
stakeholderSwot: AnalysisStakeholderSwot[],
): number;
}
/**
* Interface for coalition tension detection — heuristic or LLM-backed.
*/
export interface CoalitionTensionDetector {
/** Detect coalition tension from a document set. */
detect(
docs: RawDocument[],
lang: Language,
): CoalitionTensionResult;
}
/**
* Interface for watch-point urgency classification — heuristic or LLM-backed.
*/
export interface UrgencyClassifier {
/** Classify urgency from document type and context. */
classify(
docType: string,
docCount: number,
context: { hasFullText: boolean; hasPendingVote: boolean },
): 'critical' | 'high' | 'medium' | 'low';
}
|