Open source stability statistics

ICH Q1E inspired

Python owned calculations

Public face for a real analysis engine

OpenPharmaStability

An open source, ICH Q1E inspired toolkit for reproducible pharmaceutical stability analysis.

Upload stability data, run transparent shelf life calculations, and generate audit friendly HTML, JSON, and plot artifacts from a Python statistical engine.

Hiring signal

Shows statistical modeling, pharma context, reproducible artifacts, and regulatory aware boundaries in one coherent product surface.

127.0.0.1:8765 local workspace

Setup

CSV or XLSX

assay_3batch.csv

CONDITION

25C / 60RH

ATTRIBUTE MODE

Single

List

All

POLICY

Q1E default
BQL as declared
Drug product

Supported

run 0142

17 months supported shelf life

Statistical crossing 17.95 months. Limiting attribute assay.

MODEL

common slope

POOLABILITY

partial

Artifacts

●report.html

●record.json

●plot.png

sha256
4f9c9a8b21e7

AUDIT NOTE

Extrapolation flag retained in report record.

Python engine

CSV and XLSX

HTML and JSON

Q1E inspired

Many attributes

Hashed outputs

Method ledger

Not a calculator. A documented statistical run.

The page foregrounds the work hiring reviewers should notice: regression choices, poolability logic, confidence bounds, data warnings, and reproducible artifacts.

Load stability data

CSV or XLSX input, condition selection, attribute mode, product type, metadata, and BQL policy.

Run the engine

Batch regression, ANCOVA poolability, one sided confidence bounds, extrapolation checks, and diagnostics.

Export the record

HTML report, JSON decision record, confidence plot, optional PDF, input hash, library versions, and warnings.

Capability index

A statistical toolkit with visible assumptions.

Shelf life and retest period

Lower confidence bound crossing against specification limits.

Multi batch regression

Pooled, common slope, and batch specific model paths.

Poolability testing

Slope and intercept checks with transparent p values.

Attribute limiting

Many attributes can be screened and ranked by the limiting result.

BQL policy handling

Below quantitation choices stay explicit in the decision record.

Exploratory modules

Arrhenius, MKT, and sensitivity analysis are presented as exploratory.

Report specimen

A result sheet that can survive technical review.

The report preview does not bury the model. It shows the supported decision, statistical crossing, poolability status, warnings, and the confidence plot together. The linked artifacts are generated by the current Python engine.

Assay at 25C / 60RH

3 batches, 0 to 18 mo

Lower one sided 95% confidence bound against a 95.0% lower specification limit.

Common slope fit

95% bound

Batch data

Supported decision

17months supported shelf life

Statistical crossing

17.95 mo

Poolability

partial

Model

common slope + batch intercepts

Guidance profile

ICH Q1E (default)

Warnings 5

Step one slope poolability p value 0.9056. Step two intercept p value 1.658e minus 16.

Influence check flags four observations for review. No point is auto excluded.

Open HTML reportopen Download JSON record.json View confidence plot.png

Technical case study

The statistical core is the portfolio story.

This specimen comes from the bundled assay data and current engine. It demonstrates the part hiring reviewers should care about: model selection under Q1E style poolability logic, confidence bound crossing, and transparent warnings.

Input

42 assay observations

Three batches, condition 25C / 60RH, time coverage from 0 to 18 months.

Poolability

Partial pooling selected

Slope interaction passes with p value 0.9056. Batch intercepts differ with p value 1.658e minus 16.

Bound

One sided 95% lower bound

The decision uses the lower confidence bound crossing against a 95.0% lower specification limit.

Decision

17 months supported

Statistical crossing is 17.954842 months. The reported shelf life is rounded down to 17 months.

Architecture

The UI is thin. The Python engine is authoritative.

OpenPharmaStability separates presentation from calculation. The browser workspace submits data to a thin local service, while the Python engine owns the statistical model, report generation, JSON decision record, and artifact hashing.

No result is computed in the browser. The interface only renders what the engine has already decided, so a report opened in the UI and one produced from the CLI match the same decision record.

in

CSV / XLSX dataset

core

Python statistical engine

Stability result + decision record

out

HTMLJSONplot

ui

Local UI preview (render only)

Positioning

Educational, exploratory, reproducible decision support

OpenPharmaStability is intended for educational, exploratory, and reproducible decision support use. It is not a validated GxP system and does not provide 21 CFR Part 11 audit trails, electronic signatures, or data integrity controls.

What it is

Reproducible Transparent Audit friendly records Decision support

What it is not

Validated GxP system 21 CFR Part 11 Electronic signatures Data integrity controls

Run locally

Local by default. Your data stays on the machine.

Install & launch the workspace

bash

$ pip install openpharmastability
$ openpharmastability-ui --host 127.0.0.1 --port 8765

Open http://127.0.0.1:8765 and upload a CSV/XLSX stability dataset.

Or run a single analysis from the CLI

bash

$ openpharmastability analyze examples/assay_3batch.csv \
    --condition "25C/60RH" \
    --attribute assay \
    --output report.html

Writes report.html plus a matching JSON record and confidence plot.

Local workspace

Stability analysis for assay_3batch.csv

engine connected at 127.0.0.1:8765

openpharmastability workspace

Setup

Dataset

↑

assay_3batch.csv

18 rows, 3 batches, drag to replace

Condition

Attribute mode

Single

List

All

Product type

Drug product

Drug substance

Guidance profile

BQL policy

{{ advChevron }} Advanced options

Force common slope test Allow extrapolation past coverage Hash artifacts with SHA 256

Complete ran in 0.42s

Supported decision

17mo

Limiting attribute

assay

Guidance profile

ICH Q1E

Model

common slope + intercepts

◳report.html preview

Stability report

Assay supported shelf life

Condition 25C/60RH, 3 batches, ICH Q1E inspired evaluation

Supported shelf life	17 months
Statistical crossing	17.95 months
Specification (lower)	95.0 %
Model	common slope + batch intercepts
Poolability	partial (slopes independent)

The lower one sided 95% confidence bound of the common slope regression intersects the 95.0% specification limit at 17.95 months; the supported shelf life is therefore reported as 17 months. Slope poolability was rejected at alpha 0.25, so batch specific slopes were retained for the limiting estimate.

Poolability

Slope poolability rejected at alpha 0.25 with p = 0.18. Independent slopes retained; the most conservative batch governs the supported shelf life.

Extrapolation

Reported shelf life at 17 mo extends beyond the 18 month observed coverage only marginally; treat the estimate as exploratory per ICH Q1E extrapolation guidance.

Data quality pass

No below quantitation values detected. All six time points present across three batches.

record.json

{
  "attribute": "assay",
  "condition": "25C/60RH",
  "supported_shelf_life_months": 17,
  "statistical_crossing_months": 17.95,
  "spec_limit_lower": 95.0,
  "model": "common_slope_batch_intercepts",
  "poolability": "partial",
  "guidance_profile": "ich_q1e_default",
  "warnings": 2,
  "artifact_sha256": "4f9c...a1e7"
}

Artifacts

●report.html

get

●record.json

get

●plot.png

get

●report.pdf

get

Audit facts

Input SHA 256

c1a7 9e4f 22b0 8d63 ... 4f9c a1e7

Engine version

openpharmastability 0.4.0

Run timestamp

2026-06-22T14:08:21Z

Reproducible

inputs + config hashed

Not a validated GxP system. No 21 CFR Part 11 controls.

DESIGN.md

A design system for a scientific instrument

Warm, editorial, and precise. The system favors restraint, subtle borders over heavy shadows, generous whitespace over decoration, and functional color over ornament. Built to read as a serious analytical tool, not a marketing page.

Color

Surfaces & text

Parchment

#f4f0e7 page

Warm panel

#fdfbf6 panels

Charcoal

#211f1b text

Stone

#6f685d secondary

Accent & functional data colors

Terracotta

#b35a3a accent

Sage

#5e7a56 pass and batch

Ink blue

#3f5d78 data and batch

Amber

#b07a32 warning

Accent is reserved for primary actions, links, and the regression line. Functional colors carry meaning (sage = pass / supported, amber = caution, ink = data series) and are never used decoratively.

Typography

Newsreader

Display and headings serif

IBM Plex Sans

UI and body sans serif

Aa

IBM Plex Mono

Code, data & hashes

DisplayNewsreader 500, 46 to 58px

Section headingNewsreader 500, 30 to 36px

Subhead or card titlePlex Sans 600, 15 to 18px

Body copy, readable and compactPlex Sans 400, 14 to 16px

Eyebrow or labelPlex Mono, 12px, 0.16em

Spacing, radius & elevation

Spacing scale, 8px base

8

16

24

40

56

Radius, maximum 12px

4 control

8 card

12 panel

Elevation, borders first

Components

Buttons

Status badges

SupportedPartial

Segmented control

List

Single

All

Code / mono

$ openpharmastability analyze

Warning panel

Poolability

Slope poolability rejected at alpha 0.25. Independent slopes retained.

Usage rules

Lead with the decision. The supported number is the hero.

Use subtle borders to separate; reserve shadow for the one floating surface.

Keep numbers, hashes, and code in IBM Plex Mono.

State limitations plainly and visibly.

Don't

No gradients, glassmorphism, or decorative blobs.

No nested cards or heavy drop shadows.

No claims of validation, GxP, or 21 CFR Part 11 compliance.

No accent color used purely for decoration.

note Full token reference exported as DESIGN.md alongside this component.