Add data verification checkpoint to article generation skill

Prevent data errors by enforcing a mandatory step to read and confirm
JSON data before writing any article content.

Changes:
- Add step 2 "READ AND CONFIRM DATA" to workflow requiring explicit
  confirmation of headline value and reference period before writing
- Add "How to Use Fetched Data" section with JSON field mappings
- Add forbidden patterns (writing from memory, estimating values)
- Document Jan 2026 failure mode where articles had plausible but
  incorrect hardcoded values (2.50% vs 2.25%, 80.8% vs 80.7%)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: dshkol

.claude/skills/the-daily-generator/SKILL.md

@@ -11,26 +11,54 @@ Generate StatCan "The Daily"-style statistical bulletins from CANSIM data tables
 
 **NEVER use synthetic, made-up, or placeholder data.** Every number must come from real Statistics Canada data fetched via the `cansim` R package. If data fetch fails, do not generate the article.
 
+## How to Use Fetched Data
+
+Every number in the article MUST come from the JSON file:
+
+| Article Element | JSON Source |
+|-----------------|-------------|
+| Headline percentage | `latest.yoy_pct_change` or `latest.mom_pct_change` |
+| Headline value | `latest.value` |
+| Chart data array | `time_series[]` - copy date and value exactly |
+| Component breakdown | `subseries[]` |
+| Provincial table | `provincial[]` |
+| Reference period | `metadata.reference_period` |
+
+**Forbidden patterns:**
+- ❌ Writing a number from memory
+- ❌ Estimating values not in JSON
+- ❌ Using "plausible" numbers to fill gaps
+- ❌ Confusing similar terms (e.g., "Bank Rate" vs "Policy Rate")
+
+**If a value isn't in the JSON, don't include it in the article.**
+
 ## Workflow
 
 ```
 1. FETCH DATA
    Rscript r-tools/fetch_cansim_enhanced.R <table-number> output
    → output/data_<table>_enhanced.json
 
-2. CREATE ENGLISH ARTICLE
+2. READ AND CONFIRM DATA (MANDATORY)
+   Before writing ANY article content:
+   - Read the JSON file completely
+   - State the headline value: "latest.yoy_pct_change is X.X%"
+   - State the reference period: "metadata.reference_period is YYYY-MM"
+   - If JSON doesn't exist or is stale, STOP - do not proceed
+
+3. CREATE ENGLISH ARTICLE
    docs/en/<slug>/index.md
 
-3. CREATE FRENCH ARTICLE
+4. CREATE FRENCH ARTICLE
    docs/fr/<slug-fr>/index.md
 
-4. UPDATE LANGUAGE MAP
+5. UPDATE LANGUAGE MAP
    Add slug pair to src/lang-map.js
 
-5. UPDATE INDEX PAGES
+6. UPDATE INDEX PAGES
    Add entry to docs/en/index.md and docs/fr/index.md
 
-6. PREVIEW AND VERIFY
+7. PREVIEW AND VERIFY
    npm run dev → http://localhost:3000
 ```
 

.claude/skills/the-daily-generator/references/data-workflow.md

@@ -224,3 +224,24 @@ If the R script cannot fetch data:
 5. Try again later
 
 Never substitute synthetic data.
+
+## Known Failure Modes
+
+### Hardcoded Plausible Values (Jan 2026)
+
+**What happened**: Articles were generated with numbers that looked reasonable but weren't from the fetched data.
+
+**Examples**:
+- Interest rates article used 2.50% (the "Bank Rate") instead of 2.25% (the "Policy Rate" from JSON)
+- Manufacturing capacity used 80.8% instead of actual 80.7% from JSON
+
+**Root causes**:
+1. JSON file wasn't read before generating article text
+2. LLM used approximate values from training data instead of exact JSON values
+3. Similar-sounding terms confused (Bank Rate ≠ Policy Rate)
+
+**Prevention**:
+1. ALWAYS read JSON file before writing ANY numbers
+2. ALWAYS state headline value out loud: "The JSON shows X.X%"
+3. Copy-paste values from JSON, don't type from memory
+4. For financial data: verify exact terminology matches the JSON field name