remapping-ghost-variables - SKILL.md Agent Skill

name: remapping-ghost-variables description: Fix ghost variable warnings by remapping charts from old to new variable IDs. Use when ETL grapher step shows "Variables used in charts will not be deleted automatically", when indicator short_names were renamed and charts still reference old variables, or when the user mentions ghost variables, orphaned indicators, or chart variable remapping. metadata: internal: true

Remapping Ghost Variables

Fix charts that reference old (ghost) variable IDs after indicator short_names were renamed in a garden/grapher step.

When This Happens

After running a grapher step, you see a warning like:

[warning] Variables used in charts will not be deleted automatically.
  rows=   chartId  variableId
0     1234      56789
  variables=[56780, 56781, ..., 56789]

This means someone renamed indicator shortNames in the ETL step, creating new variables while charts still reference the old ones. The ETL can't delete the old variables because charts depend on them.

Workflow

Step 1: Create a staging PR

Create a PR first — all investigation and changes happen against the staging DB.

.venv/bin/etl pr "Remap chart <chart_id> ghost variables for <dataset>" data

Wait for the staging server to come up.

Step 2: Identify the affected chart and old variables

From the warning, extract:

Chart IDs from the rows table (e.g., chart 1234)
Ghost variable IDs from the variables list

All queries below use make query which automatically targets the staging DB for the current branch.

Query the old variables to understand what was renamed:

make query SQL="SELECT id, name, shortName FROM variables WHERE id IN (<ghost_variable_ids>)"

Step 3: Find what the chart currently uses

make query SQL="SELECT variableId FROM chart_dimensions WHERE chartId = <chart_id>"

Then get details on each:

make query SQL="SELECT id, name, shortName FROM variables WHERE id IN (<chart_variable_ids>)"

Step 4: Find the new replacement variables

The new variables are in the same dataset but with different shortNames. Get the dataset ID first:

make query SQL="SELECT datasetId FROM variables WHERE id = <any_ghost_variable_id>"

Then search for the new variables. Look for the pattern difference between old and new shortNames:

make query SQL="SELECT id, shortName FROM variables WHERE datasetId = <dataset_id> AND shortName LIKE '<pattern_matching_new_name>' AND id NOT IN (<ghost_variable_ids>)"

Compare old vs new shortNames to confirm the mapping. Common rename patterns:

Encoding changes (e.g., gte_40 → gt__40)
Prefix/suffix changes
Restructured naming conventions

Step 5: Remap via the indicator upgrader (preferred)

Insert the mappings, then run the upgrader — it updates dimensions, map configs, sortBy, and narrative charts, and records the mapping in wiz__variable_mapping so chart diff picks it up:

STAGING=<branch> .venv/bin/python -c "
from apps.wizard.utils.db import WizardDB
WizardDB.add_variable_mapping(
    mapping={<old_id>: <new_id>},
    dataset_id_old=<old_ds_id>, dataset_id_new=<new_ds_id>,
    comments='<why>')"
STAGING=<branch> .venv/bin/python -m apps.indicator_upgrade.cli upgrade --dry-run
STAGING=<branch> .venv/bin/python -m apps.indicator_upgrade.cli upgrade

Check WizardDB.get_variable_mapping_raw() first — upgrade applies all stored mappings, not just yours.

Step 5 (fallback): Remap the chart via Admin API

If the indicator upgrader can't be used, update the chart config manually on staging with AdminAPI:

from apps.chart_sync.admin_api import AdminAPI
from etl.config import OWIDEnv

env = OWIDEnv.from_staging('<branch_name>')
api = AdminAPI(env)

# Get current chart config
config = api.get_chart_config(<chart_id>)

# Remap old variable IDs to new ones
REMAP = {
    <old_var_id>: <new_var_id>,
    # ... add more mappings if multiple variables need remapping
}

for dim in config.get('dimensions', []):
    if dim['variableId'] in REMAP:
        old = dim['variableId']
        dim['variableId'] = REMAP[old]
        print(f'Remapped {old} -> {REMAP[old]}')

# Update the chart
result = api.update_chart(<chart_id>, config)
assert result['success'], f"Update failed: {result}"
print('Chart updated successfully')

Guidelines

Always use staging — never update charts directly on production
Verify the data is identical — the new variable should contain the same data as the old one, just with a different shortName
Handle multiple charts — the warning may list multiple chartId rows; remap all of them
Handle multiple variables per chart — a chart may reference several ghost variables; remap all in one update
Check for narrative charts — ghost variables may also be referenced by narrative charts (check chart_dimensions isn't the only reference)