Pbi drafter

PBI Drafter

Overview

Execute the Dashboard Drafter pipeline from a /*FACTORY ... */ Config Block and produce a ready-to-open PBIP dashboard. This skill only generates — it does not draft configs. Use pbi-drafter-configurator first if you don't have a config block yet.

Trigger phrases: generate dashboard, run dashboard drafter, build dashboard from config, run factory, create dashboard, PBI dashboard generator, ダッシュボード生成, or a /*FACTORY ... */ block with generation intent.

---

Step 1 — Validate Input

• Confirm the input contains a /*FACTORY ... */ block. If not, tell the user to run pbi-drafter-configurator first.

• Check that the SOURCE: file exists at the specified path. If not, report the missing path and stop.

• Verify the config uses actual CSV column names before AS (e.g. amount AS "Amount", not SUM_Measure_1 AS "Amount"). If slot placeholder names are used as source columns, warn the user:

  ⚠️ The config uses template slot names (e.g. SUM_Measure_1 AS "Amount") as source columns. This will cause Power BI to error with "column not found" on refresh. Use pbi-drafter-configurator to regenerate the config with actual CSV column names, or manually replace the slot names with the real CSV column headers.

---

Step 2 — Pre-flight

• Remind the user: "Close PBI Desktop before generation to avoid file-lock errors."
• Verify the template folder exists at pbi-drafter/template/ relative to the repo root.
  • If missing, stop and report the path.

---

Step 3 — Generate

Working directory: <repo-root>/pbi-drafter/

Before running, sanitise the TITLE from the config block:

• Replace & with and
• Remove any character in < > : " / \ | ? *

These characters are invalid in Windows folder names and will cause run_factory to fail silently or produce a broken output path. If sanitisation changes the title, inform the user: "TITLE sanitised: 'Profit & Loss' → 'Profit and Loss'".

Run the following Python (inline, no separate script file needed):

import sys, shutil, re
from pathlib import Path

repo_root = Path(r"<repo-root>")
drafter = repo_root / "pbi-drafter"
sys.path.insert(0, str(drafter))

from src.config_parser import parse_config
from src.factory import run_factory

config_text = r'''<paste the full /*FACTORY ... */ block here>'''

cfg = parse_config(config_text)

raw_title = cfg['title']
safe_title = raw_title.replace('&', 'and')
safe_title = re.sub(r'[<>:"/\\|?*]', '', safe_title).strip()
if safe_title != raw_title:
    cfg['title'] = safe_title
    print(f'TITLE sanitised: {raw_title!r} -> {safe_title!r}')

title = cfg['title'].replace(' ', '_')
out = drafter / 'output' / title
if out.exists():
    shutil.rmtree(out)
result = run_factory(
    drafter / 'template',
    drafter / 'output',
    cfg,
)
print('Generated:', result)

If generation fails, show the full error traceback — do not suppress or summarise it.

---

Step 4 — Verify Output

After successful generation:

1. Output first 20 lines of Fact.tmdl from pbi-drafter/output/<Title>/<Title>.SemanticModel/definition/tables/Fact.tmdl
2. Check no template slot names remain in visible sourceColumn: entries (lines without isHidden). Template slot names are: SUM_Measure_1–10, CNT_Measure_1–5, AVG_Measure_1–5, Key_Dim_1–10, Other_Field_1–10, DateKey.
3. Verify format strings — SUM columns with $ format → formatString: $#,0.00; # → integer; % → percentage; #.0 → one decimal.
4. Report hidden field count: count lines containing isHidden in Fact.tmdl.

---

Step 5 — Report to User

✅ Dashboard generated at:
   pbi-drafter/output/<Title>/<Title>.pbip

Open in PBI Desktop and click "Refresh" to load data.

[first 20 lines of Fact.tmdl]
[any verification warnings]

---

Critical Rules

Rule	Detail
Working directory	<repo-root>/pbi-drafter/
Never modify template/	The template is the golden master. All output goes to output/.
Delete before regenerate	Always shutil.rmtree(out) before run_factory to avoid stale artefacts.
TITLE must be Windows filename-safe	Forbidden characters in TITLE: `& < > : " / \
Actual CSV column names	Source token before AS must be the real CSV column name. Slot placeholder names cause "column not found" in Power BI.
CNT columns → type text	CNT source columns (used for DISTINCTCOUNT) load as type text. ID fields like E00001 or INV-12345 are strings — this is intentional and correct.
DAX_ measures excluded from M Query	The DAX_ prefix measures are TMDL artefacts only — they are never part of the M Query field_map.
en-US.tmdl untouched	Never read or write en-US.tmdl — PBI Desktop auto-regenerates it.
sourceColumn = M Query output name	After the rename step, sourceColumn in TMDL must match the business display name the M Query outputs.
Date categoricals excluded from KEY	Month, Quarter, Year, FY, MonthName, Period, Week columns should NOT be in KEY slots.
Full traceback on failure	If run_factory or parse_config raises, show the complete traceback — never summarise.
350 TDD tests	All pipeline steps are covered. Run python -m pytest tests/ -q in pbi-drafter/ if something seems broken.