diff --git a/.github/workflows/publish.yml b/.github/workflows/publish.yml
index d29226e..b4be8dd 100644
--- a/.github/workflows/publish.yml
+++ b/.github/workflows/publish.yml
@@ -15,7 +15,12 @@ jobs:
id-token: write
steps:
- uses: actions/checkout@v5
- - uses: actions/setup-python@v5
+
+ - name: Engagement guard
+ run: python3 scripts/engagement_guard.py
+
+ - name: Set up Python
+ uses: actions/setup-python@v6
with:
python-version: "3.12"
- run: python -m pip install build
diff --git a/.gitignore b/.gitignore
index e6b130a..74e4b08 100644
--- a/.gitignore
+++ b/.gitignore
@@ -50,3 +50,11 @@ secrets.*
# UI-review tool output (regenerated each run)
screenshots/
+
+# --- lailara engagement scaffold ---
+# Client engagement data is runtime-only: never commit it, never deploy it.
+client-data/
+client-output/
+/engagement.yml
+/engagement.yaml
+# (engagement.demo.yml and engagement.example.yml stay committable)
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 947fb3b..c4aabc2 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,23 @@
All notable changes to datascope are documented here.
+## [Unreleased]
+
+### Fixed
+- HTML report generator timestamp now honors `SOURCE_DATE_EPOCH` (the reproducible-builds standard), so regenerated sample reports are byte-identical run-to-run. A bare `datetime.now()` in the report footer otherwise changed every run and defeated any byte-lock on the output.
+
+### Changed
+- Regenerated `samples/output/` from current source with `SOURCE_DATE_EPOCH` pinned, so the shipped showcase artifacts reflect the 2.4.0 tool and are reproducible. See `scripts/regenerate_samples.sh`. (The annotated `.xlsx` content is reproducible but its openpyxl envelope carries wall-clock member mtimes, so it is not raw-byte identical — the `.html`/`.pdf` are.)
+
+## [2.4.0] — 2026-08-05
+
+### Fixed
+- **Mixed date formats are now detected in CSV files.** The CSV loader coerced date-like strings to `datetime` on load, erasing the raw format before the mixed-date analyzer could see it — so a column mixing `2026-01-01` and `01/02/2026` was silently reconciled and never flagged, the exact silent coercion datascope exists to surface. Date-like CSV cells are now kept as strings (a CSV has no type metadata; a date is text), so `analyze_mixed_dates` sees the raw formats and reports the inconsistency. Excel date cells, which arrive already typed from openpyxl, are unaffected.
+
+### Docs
+- README missing-value row now names both thresholds it depends on — flagged at ≥10% blank (`_DEFAULT_THRESHOLD_PCT`) and Warning at ≥50% / Info below (`findings/severity.py`). Correction: an earlier draft of this entry read "threshold now reads 10%", which put the flag floor into a row labeled *Warning* — the two are different decisions and the row now states both.
+- Removed the CSV mixed-date caveat from the README. The loader fix above now surfaces mixed dates in CSVs, so the caveat added in 2.3.4 ("supply as `.xlsx`") no longer holds.
+
## [2.3.4] — 2026-07-31
### Fixed
diff --git a/README.md b/README.md
index 92953df..d22a0f4 100644
--- a/README.md
+++ b/README.md
@@ -19,7 +19,7 @@ datascope finds these problems, explains what's wrong in plain English, and tell
|---|---|---|
| **Mixed types** | 485 numbers + 15 strings in a "numeric" column | Critical |
| **Sentinel values** | "N/A", "TBD", "pending" hiding in numeric data | Critical |
-| **Missing values** | 50%+ of a column is blank — aggregations silently exclude those rows (below 50% is flagged Info) | Warning |
+| **Missing values** | Flagged at ≥10% blank; Warning at ≥50%, Info below — aggregations silently exclude those rows | Warning / Info |
| **Leading-zero inconsistency** | "00123" alongside "456" — keys that won't match | Warning |
| **Mixed date formats** | "01/15/2026" and "2026-01-15" in the same column | Warning |
| **Suspected duplicate IDs** | 98% unique in an ID column — the other 2% will fan out joins | Warning |
@@ -27,8 +27,6 @@ datascope finds these problems, explains what's wrong in plain English, and tell
Each finding is expressed as **assumption vs. reality**: what the data *appears* to be vs. what it *actually contains*. Every finding includes a downstream impact explanation, a fix recommendation, and a prevention rule.
-> **Note on CSV input:** date strings written in formats the CSV loader recognizes (e.g. `2026-01-15` and `01/15/2026`) are parsed to real dates at load time, so a CSV column mixing those two formats is normalized before analysis and does **not** raise a *Mixed date formats* finding. To surface mixed date formats, supply the column as text — for example in an `.xlsx` file with text-formatted cells, where the values stay strings and the check fires as expected.
-
---
## Installation
diff --git a/datascope/__init__.py b/datascope/__init__.py
index 15ecef3..78f7ade 100644
--- a/datascope/__init__.py
+++ b/datascope/__init__.py
@@ -2,4 +2,4 @@
from __future__ import annotations
-__version__ = "2.3.4"
+__version__ = "2.4.0"
diff --git a/datascope/loaders/csv_loader.py b/datascope/loaders/csv_loader.py
index a021394..d6db8d0 100644
--- a/datascope/loaders/csv_loader.py
+++ b/datascope/loaders/csv_loader.py
@@ -12,12 +12,10 @@
from __future__ import annotations
import csv
-from datetime import datetime
from pathlib import Path
import pandas as pd
-from datascope.analyzers.format_check import DATE_LIKE_RE
from datascope.loaders.base import dedupe_headers
from datascope.models import LoaderResult
@@ -30,17 +28,6 @@
# it must stay a string rather than becoming ``inf``/``NaN``.
_NON_FINITE_STRS = frozenset({"inf", "infinity", "nan"})
-# Date/time formats tried in order (most specific first).
-_DATETIME_FMTS = (
- "%Y-%m-%dT%H:%M:%S", # ISO 8601
- "%Y-%m-%d %H:%M:%S", # space-separated
- "%Y-%m-%d", # date only
- "%m/%d/%Y %H:%M:%S",
- "%m/%d/%Y",
- "%d/%m/%Y",
- "%Y/%m/%d",
-)
-
def _infer_cell(raw: str) -> object:
"""Infer a single cell's Python value from its raw CSV string.
@@ -50,8 +37,14 @@ def _infer_cell(raw: str) -> object:
2. Integer
3. Float
4. Boolean (true/false/yes/no, case-insensitive)
- 5. Datetime (common formats)
- 6. String fallback
+ 5. String fallback
+
+ Date-like cells are DELIBERATELY left as strings. A CSV has no type
+ metadata, so a date is text; coercing it to ``datetime`` here would erase the
+ very format evidence the mixed-date analyzer needs, silently hiding a
+ mixed-format column ("2026-01-01" vs "01/02/2026") — exactly the silent
+ coercion datascope exists to surface. (Excel dates arrive already typed from
+ openpyxl, so the Excel loader keeps its datetime cells.)
"""
stripped = raw.strip()
if not stripped:
@@ -92,15 +85,10 @@ def _infer_cell(raw: str) -> object:
if lower in _BOOL_FALSE:
return False
- # --- datetime -----------------------------------------------------
- if DATE_LIKE_RE.match(stripped):
- for fmt in _DATETIME_FMTS:
- try:
- return datetime.strptime(stripped, fmt)
- except ValueError:
- continue
-
# --- string fallback ----------------------------------------------
+ # Date-like strings intentionally fall through to here (see docstring):
+ # kept as text so analyze_mixed_dates can see the raw format and flag a
+ # mixed-format column instead of it being silently coerced away.
return stripped
diff --git a/datascope/reports/html.py b/datascope/reports/html.py
index db5ba3b..c40da98 100644
--- a/datascope/reports/html.py
+++ b/datascope/reports/html.py
@@ -10,6 +10,7 @@
import base64
import datetime
import html
+import os
from collections import defaultdict
from pathlib import Path
from typing import Any
@@ -107,7 +108,16 @@ def write_html(
filename = source_metadata.get("filename", "unknown")
rows = source_metadata.get("row_count", "?")
cols = source_metadata.get("column_count", "?")
- now = datetime.datetime.now().strftime("%Y-%m-%d %H:%M")
+ # Honor SOURCE_DATE_EPOCH (reproducible-builds standard) so regenerated
+ # sample reports are byte-identical; a bare datetime.now() here makes the
+ # generator tag change every run and defeats any byte-lock on the output.
+ _sde = os.environ.get("SOURCE_DATE_EPOCH")
+ _dt = (
+ datetime.datetime.fromtimestamp(int(_sde), tz=datetime.timezone.utc)
+ if _sde
+ else datetime.datetime.now()
+ )
+ now = _dt.strftime("%Y-%m-%d %H:%M")
counts = severity_counts(findings)
total = sum(counts.values())
diff --git a/pyproject.toml b/pyproject.toml
index b6d26e9..866f450 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
[project]
name = "datascope-dq"
-version = "2.3.4"
+version = "2.4.0"
description = "Data quality diagnostics for tabular datasets — surfaces hidden problems in plain English"
readme = "README.md"
requires-python = ">=3.10"
diff --git a/samples/output/sample_mixed_types_annotated.xlsx b/samples/output/sample_mixed_types_annotated.xlsx
index 05b021a..8d489b5 100644
Binary files a/samples/output/sample_mixed_types_annotated.xlsx and b/samples/output/sample_mixed_types_annotated.xlsx differ
diff --git a/samples/output/sample_mixed_types_diagnostic.html b/samples/output/sample_mixed_types_diagnostic.html
index b606fb6..a21545f 100644
--- a/samples/output/sample_mixed_types_diagnostic.html
+++ b/samples/output/sample_mixed_types_diagnostic.html
@@ -3,8 +3,8 @@
-
-
+
+
datascope diagnostic — sample_mixed_types.xlsx