PHFrame

PHFrame is an early-stage framework for building extensible public-health data systems. It combines generated projects, declarative health schemas, persistent storage, automatic APIs, a public-health engine, and a standards-based Web Component interface. The original CSV/Excel dashboard generator remains available as an export workflow.

Current version: 0.8.0a11 (global floating AI assistant)

Why PHFrame?

Public-health teams often begin with spreadsheets and later need validated imports, persistent storage, audit history, and APIs. PHFrame provides a progressive path from those files to a configurable application while keeping the data model explicit.

Current capabilities include:

• Declarative datasets and validation in phframe.yaml
• Generated CRUD APIs backed by SQLite or PostgreSQL
• Atomic CSV, Excel, JSON, and XML imports with reusable column mappings
• Import audit history and safe schema migration checks
• Portable HTML dashboards for offline sharing
• Declarative indicators with count, sum, average, rate, ratio, and percentage calculations
• Drag-resizable dashboards with add/remove controls and browser-saved visualization choices
• Browser-managed typed columns for worldwide, programme-specific data models
• Browser imports with downloadable examples and scheduled generic API, DHIS2, KoboToolbox, and ODK synchronization
• Editable branding, navigation, footer, themes, primary color, dashboard title, logo, and favicon
• Custom navigation pages with drag-and-drop text, table, and visualization blocks or external URL redirects
• Safe rich-text footer and page content with bold, italic, underline, lists, and hyperlinks
• Optional private mode with PBKDF2-hashed local user credentials and signed, HTTP-only login sessions
• Privacy-aware AI summaries grounded in configured aggregate evidence, with human approval and append-only audit events
• De-identification previews that remove protected/direct identifiers and generalize person-level dates and ages

[!IMPORTANT] PHFrame is alpha software. Evaluate it with non-sensitive data before considering production use. Its optional local login is a preview and is not yet a complete production deployment security model.

Create a public-health application

phframe new "Malaria Surveillance"
cd malaria-surveillance
phframe check
phframe serve

Open http://127.0.0.1:8000/app for the application interface. The generated project includes:

• phframe.yaml — project, database, dataset, field, and plugin configuration
• data/phframe.db — local SQLite database
• plugins/ — application-specific extension modules
• /api — discoverable dataset metadata
• /api/case_reports — generated collection API
• /api/case_reports/{id} — generated record API
• /app — responsive Web Component application

Web Component application

The Phase 3 interface is served directly by PHFrame and has no React or external CDN dependency. It includes:

• Hash-based application routing and a responsive application shell
• Metadata-driven record forms and protected-field-aware tables
• Saved filters and organisation-unit selection
• KPI cards, grouped SVG charts, epidemiological curves, and offline tile choropleths
• Declarative dashboard composition in phframe.yaml
• Light, dark, and high-contrast themes using CSS design tokens
• English and Bengali localization foundations with project-specific translation overrides
• Keyboard focus styling, skip navigation, reduced-motion support, live notifications, modal dialogs, and confirmations

Use Pages in the application navigation to create an internal page or an external link. Internal pages have a drag-and-drop canvas where editors can add rich text, live dataset tables, and configured metrics or charts. Saving a page adds it to the navigation automatically. External pages redirect the visitor to the supplied HTTP or HTTPS URL.

Use Settings to configure the header and dashboard titles, logo, favicon, navigation labels, access mode, theme, and primary color. The primary color is automatically adjusted for readable accents in dark mode while the selected color remains the header brand color. The footer editor supports formatted text and safe web links.

Privacy-aware AI assistance

Phase 5 adds a responsible AI workspace at /app#/ai. Its default local provider produces conservative summaries from configured aggregate indicators, dimensions, threshold states, and data-quality results. It does not send row-level records or protected fields to a model. Every factual observation links to a numbered evidence item and its PHFrame API endpoint.

The workspace is chat-first. A floating Ask PHFrame control is available on every application page and opens a responsive, closable assistant without interrupting the current task. Users can ask plain-language questions such as “Are cases increasing over time?”, “Which locations have the highest counts?”, “Is the latest value unusual?”, “What data-quality problems should I check?”, or “Why might this alert be triggered?”. PHFrame selects relevant aggregate evidence instead of repeating the whole dashboard, calculates first-to-latest change, flags latest-point anomalies using the prior-point baseline, retains evidence context for follow-up questions, and clearly separates observed signals from unproven causes. Messages are labeled Me automatically; no chat-name field is required.

Any analyst answer can become a situation-report draft. Reports enter the same human approval queue and can be downloaded as Markdown with draft status, reviewer information, evidence digest, and source endpoints embedded in the file.

Every generated summary begins in draft status. A human must provide a review note and explicitly approve or reject it. Generation and review decisions are stored as append-only audit events, while an SHA-256 digest binds the stored summary to the exact evidence snapshot used at generation time.

The workspace also provides a de-identification preview. Fields marked protected and fields typed as identifier are removed, dates are reduced to years, and ages are placed in ten-year bands with 90+ top-coding. This is a technical exposure-reduction tool—not a determination of legal compliance or a guarantee against re-identification. HHS recognizes both Safe Harbor and Expert Determination approaches and warns that de-identified data retains some identification risk: HHS de-identification guidance.

PHFrame’s human approval, privacy receipt, evidence register, limitations, and audit history are informed by WHO guidance emphasizing autonomy, safety, transparency, accountability, inclusiveness, and sustainability for health AI: WHO ethics and governance guidance. They also align with the privacy-enhanced, accountable, transparent, explainable, safe, and reliable characteristics described by the NIST AI Risk Management Framework.

Optional external provider

External AI is disabled by default. To use an OpenAI-compatible chat-completions service:

1. Open Settings → Responsible AI.
2. Select External OpenAI-compatible API.
3. Provide an HTTPS endpoint, model name, and the name of an environment variable containing the API key.
4. Explicitly enable aggregate evidence transfer, save, and restart PHFrame with that environment variable set.

For example:

export PHFRAME_AI_API_KEY="replace-with-provider-key"
phframe serve

The API key itself is never saved in phframe-settings.json or returned by PHFrame. Even for an external provider, the request contains aggregate evidence only. Administrators remain responsible for provider agreements, jurisdictional requirements, consent, data-protection impact assessment, retention settings, and organizational approval.

Phase 5 endpoints:

• POST /api/ai/deidentify/{dataset} — preview protected-field removal and generalization
• GET|POST /api/ai/chat — retrieve a conversation or ask an evidence-targeted analyst question
• POST /api/ai/chat/{id}/report — promote an analyst answer into a reviewable report draft
• GET|POST /api/ai/summaries — list or generate evidence-grounded drafts
• GET /api/ai/summaries/{id} — retrieve a draft, evidence snapshot, and privacy receipt
• GET /api/ai/summaries/{id}/export — download a governed Markdown report
• POST /api/ai/summaries/{id}/review — approve or reject once with a required note
• GET /api/ai/audit — list append-only generation and decision events

Configure the interface and dashboard:

ui:
  theme: light
  locale: en
  translations: {}

dashboards:
  main:
    label: Malaria Surveillance Dashboard
    widgets:
      - type: kpi
        title: Total cases
        indicator: total_cases
      - type: chart
        title: Cases by district
        dimension: cases_by_district
      - type: map
        title: Geographic distribution
        dimension: cases_by_district
      - type: epi_curve
        title: Cases over time
        dataset: case_reports
        date_field: report_date
        value_field: cases

Browser imports

The browser import workspace accepts .csv, .xlsx, .xlsm, .json, and .xml. It provides downloadable examples for the selected dataset before upload. JSON may be an array of objects or contain a data, records, results, or items array; XML uses a root element containing one child element per record.

Use Data builder in the application to add optional typed columns. Additions are migrated safely and persisted to phframe.yaml; existing records are retained.

Dashboard cards can be reordered and resized, removed or restored, and switched between appropriate number, gauge, bar, donut, line, column, tile, and table views. Use Add visualization to chart configured indicators, dimensions, or any typed numeric, categorical, and date field. Layout choices are stored in the browser.

Resize cards from any edge or corner. The grid span, height, chart layout, legend, and content overflow adapt together; mobile layouts continue to collapse to a single column.

Dashboards are not limited to the configured starter view. The dashboard selector can switch among any number of server-saved user dashboards. Use Create new to start from a blank canvas or one of six professional templates: Executive overview, Surveillance operations, Programme monitoring, DHIS2 aggregate, Worldwide geospatial, and Blank canvas. PHFrame recommends a template from the selected dataset's numeric, categorical, date, DHIS2, and coordinate capabilities. Use Customize this dashboard to turn a project-configured dashboard into an editable saved copy.

Each dashboard accepts live metrics, grouped charts/tables, trends, coordinate maps, and rich content blocks. Content blocks support editable headings, paragraphs, bold/italic/underline text, lists, and safe hyperlinks. Dashboard names and descriptions can be edited without changing phframe.yaml.

Publish a dashboard with Cloudflare Pages

Open Settings → Cloudflare Pages and enter the Cloudflare account ID, a default Pages project name, and the name of the environment variable that holds the API token. Keep the secret outside the project:

export CLOUDFLARE_API_TOKEN="your-cloudflare-api-token"
phframe serve

On the Dashboard screen, choose Publish dashboard. Snapshot mode embeds the current aggregate results and needs no backend. Live mode periodically requests an HTTPS aggregate API through a fixed-origin Pages Worker; its edge response is cached for the selected refresh interval. If that API requires a bearer token, configure UPSTREAM_API_TOKEN as a Cloudflare Pages secret after deployment.

Every bundle and deployment first runs a privacy review. Unsupported widgets, protected dimensions, and protected source fields are blocked. The generated site contains zero row-level records. Download bundle provides the same privacy-reviewed ZIP for manual deployment with Wrangler. One-click deployment requires Node.js/npx, a Cloudflare account ID, and a token with Pages edit permission. PHFrame stores publication URLs and audit results, never API tokens.

An externally reachable public PHFrame server can supply its reviewed aggregate feed at:

GET /api/publications/feed/{dashboard_id}

For a worldwide coordinate map, add numeric columns named latitude and longitude (also recognized: lat/lon, lat/lng, or y_coordinate/x_coordinate). PHFrame recommends the geospatial template and offers the coordinate map automatically. The map endpoint aggregates coordinates into 0.01-degree buckets before returning them to the browser and never includes protected record fields:

GET /api/geospatial/{dataset}?latitude=latitude&longitude=longitude

For administrative choropleth maps, open Settings → Geographic boundaries, search for a country by name or ISO3 code, select ADM0–ADM5, and download the layer. PHFrame stores the simplified GeoJSON inside the project and uses the newest installed layer for geographic dimension maps. Area names in the data are matched to the downloaded boundary names. Boundary files come from geoBoundaries gbOpen under CC BY 4.0 and include attribution in the map.

Branding and access settings

Open Settings to edit the product name, header title, dashboard title, navigation labels and visibility, footer, primary color, default theme, logo, and favicon. Uploaded assets are stored in the project's ignored data/branding/ directory, so each deployment can have independent branding.

Projects are public by default. To enable private mode, create a username and a password of at least 10 characters in Settings and select Private. Passwords use PBKDF2-HMAC-SHA256 with per-user salts, and browser sessions use signed, HTTP-only, same-site cookies. Use HTTPS in any networked deployment. This local alpha authentication is appropriate for evaluation and controlled deployments; SSO, roles, account recovery, audit controls, and production hardening remain future work.

Open /app#/import to import .csv, .xlsx, or .xlsm files through a guided workflow:

1. Select the target dataset and file.
2. Preview up to ten rows.
3. Map source columns to typed dataset fields.
4. Save the mapping as a reusable server-side template.
5. Validate without writing, or import all rows atomically.
6. Inspect row-level failures through the import run error API.

Browser uploads are limited to 25 MB. GET /api/import-mappings lists reusable mappings, and GET /api/imports/{run_id}/errors provides structured error reports.

DHIS2, KoboToolbox, and ODK connectors

The Connectors screen provides separate guided setup choices for generic JSON REST APIs, DHIS2 data value sets, KoboToolbox assets, and ODK Central project/form OData feeds. It can create, test, schedule, synchronize, and remove each connector. Generic APIs accept a JSON array or common record containers such as data, records, results, items, and value. Source paths map into typed dataset fields; synchronized records then become available to dashboard visualizations. Credentials are referenced through environment-variable names and are never written into project configuration.

Connectors pull JSON records, apply nested source-to-dataset mappings, validate every record, and write atomically. Credentials are read only from environment variables and are never returned through metadata APIs.

Example KoboToolbox connector:

connectors:
  kobo_cases:
    type: kobo
    dataset: case_reports
    base_url: https://kf.kobotoolbox.org
    resource: your_asset_uid
    schedule_minutes: 60
    auth:
      token_env: KOBO_TOKEN
    mapping:
      case_id: case_id
      disease: disease
      status: status
      report_date: report_date
      district: district
      cases: cases

DHIS2 uses resource as the data-set UID and reads dataValues; ODK uses PROJECT_ID/FORM_ID and reads the OData Submissions feed. Nested source fields use dot notation, such as __system.submissionDate.

connectors:
  dhis2_values:
    type: dhis2
    dataset: aggregate_values
    base_url: https://play.dhis2.org/example
    resource: DATA_SET_UID
    auth:
      token_env: DHIS2_TOKEN
    parameters:
      period: 202608
      orgUnit: ORG_UNIT_UID
    mapping:
      dataElement: data_element
      period: period
      orgUnit: organisation_unit
      value: value

  odk_cases:
    type: odk
    dataset: case_reports
    base_url: https://central.example.org
    resource: 7/malaria_case
    auth:
      token_env: ODK_SESSION_TOKEN
    parameters:
      $top: 500
    mapping:
      meta.instanceID: case_id
      disease: disease
      status: status
      __system.submissionDate: report_date
      district: district
      cases: cases

Run connectors manually, validate them, or invoke due schedules from cron/systemd:

phframe sync kobo_cases
phframe sync kobo_cases --dry-run
phframe sync --all
phframe sync --all --due
phframe syncs --limit 50

schedule_minutes determines whether --due selects a connector; PHFrame records every completed, validated, or failed synchronization. The browser console at /app#/connectors exposes the same status and history. See the official DHIS2 API authentication, KoboToolbox API v2 migration, and ODK Central OData documentation for provider-side setup.

Example request:

curl -X POST http://127.0.0.1:8000/api/case_reports \
  -H 'content-type: application/json' \
  -d '{
    "case_id": "MAL-001",
    "disease": "Malaria",
    "status": "confirmed",
    "report_date": "2026-07-21",
    "district": "Bandarban",
    "cases": 1
  }'

The schema supports generic string, integer, number, boolean, date, datetime, and location fields, plus required and protected metadata.

Reusable public-health types add domain-aware storage and validation:

• identifier, disease_code, organisation_unit, and facility require non-empty text.
• age accepts whole years from 0 through 130.
• sex accepts female, male, intersex, or unknown.
• case_classification accepts suspected, probable, confirmed, or discarded.
• epi_week accepts ISO week values such as 2026-W33.
• reporting_period accepts ISO weeks, calendar months, or quarters.

These types use portable text or integer database columns, so applications retain SQLite and PostgreSQL compatibility.

Organisation-unit hierarchy

Declare reporting structures with stable codes and parent relationships:

organisation_units:
  bangladesh:
    name: Bangladesh
    level: country
  chattogram_division:
    name: Chattogram Division
    level: division
    parent: bangladesh
  bandarban:
    name: Bandarban
    level: district
    parent: chattogram_division

GET /api/organisation-units lists the hierarchy and root codes. GET /api/organisation-units/{code} returns the unit with its children and ordered ancestors. Values written to organisation_unit fields must reference a configured code. PHFrame rejects missing parents and hierarchy cycles during configuration loading.

Indicators

Define deterministic indicators alongside datasets in phframe.yaml:

indicators:
  total_cases:
    dataset: case_reports
    operation: sum
    field: cases
    date_field: report_date
  incidence_per_100000:
    dataset: case_reports
    operation: rate
    numerator: cases
    denominator: population
    multiplier: 100000
    date_field: report_date

Retrieve results through the generated API:

curl 'http://127.0.0.1:8000/api/indicators/total_cases'
curl 'http://127.0.0.1:8000/api/indicators/total_cases?district=Bandarban'
curl 'http://127.0.0.1:8000/api/indicators/total_cases?start=2026-07-01&end=2026-07-31'
curl 'http://127.0.0.1:8000/api/indicators/total_cases?period=2026-W30'

The supported operations are count, sum, average, rate, ratio, and percentage. Rate, ratio, and percentage indicators return null when the summed denominator is zero.

Named periods accept ISO epidemiological weeks (2026-W30), calendar months (2026-07), and quarters (2026-Q3).

Data-quality rules

Configure completeness, numeric range, and allowed-value checks without changing imported records:

data_quality:
  cases_nonnegative:
    dataset: case_reports
    field: cases
    check: range
    min: 0
  valid_case_status:
    dataset: case_reports
    field: status
    check: allowed
    values: [suspected, probable, confirmed]

GET /api/data-quality evaluates every rule. GET /api/data-quality/{rule} returns its record count, valid count, violation count, and percentage score.

Saved filters and dimensions

Reusable filters keep common cohorts consistent across indicators and grouped summaries:

filters:
  confirmed_cases:
    dataset: case_reports
    values:
      status: confirmed

dimensions:
  confirmed_cases_by_district:
    dataset: case_reports
    field: district
    filter: confirmed_cases

Apply a saved filter with GET /api/indicators/total_cases?filter=confirmed_cases. Dimension results are available from GET /api/dimensions/{name} and return each distinct value with its record count. Request fields can override saved filter values when an ad hoc refinement is needed.

Surveillance thresholds

Attach alert levels to deterministic indicators:

thresholds:
  high_weekly_case_count:
    indicator: total_cases
    operator: gte
    value: 10
    severity: warning
    message: Weekly case count has reached the surveillance alert level.

Evaluate rules with GET /api/thresholds or GET /api/thresholds/{name}. The endpoints accept the same period, filter, start, end, and field-filter parameters as indicators and return normal, triggered, or no_data. Supported operators are gt, gte, lt, lte, and eq.

Schema migrations

PHFrame tracks the configured dataset schemas in the project database. Preview safe changes with:

phframe migrate --check

Apply them with:

phframe migrate

Adding optional fields is automatic. PHFrame refuses destructive field removal, incompatible type changes, and new required fields that would invalidate existing records.

Persistent data imports

Validate an import without writing records:

phframe import case_reports monthly-cases.xlsx --dry-run

If spreadsheet headings already match dataset fields, import directly:

phframe import case_reports monthly-cases.xlsx

Map different source headings and save the mapping for future reporting periods:

phframe import case_reports monthly-cases.xlsx \
  --map 'Case Number=case_id' \
  --map 'Disease Name=disease' \
  --map 'Classification=status' \
  --map 'Reported=report_date' \
  --map 'Area=district' \
  --map 'Case Count=cases' \
  --save-mapping mappings/case-reports.yaml

Reuse it later:

phframe import case_reports next-month.xlsx --mapping mappings/case-reports.yaml
phframe imports

Imports are atomic: PHFrame validates every row before inserting anything. Each validation or import attempt is recorded in the internal audit history and exposed at GET /api/imports.

Development and production configuration

Generated projects use SQLite for local development. Paths in SQLite URLs are resolved relative to phframe.yaml:

project:
  name: Malaria Surveillance
  database: sqlite:///data/phframe.db
  environment: development

server:
  host: 127.0.0.1
  port: 8000

Start the development server with automatic reload:

phframe serve --reload

For PostgreSQL, install the optional driver:

pip install 'public-health-framework[postgres]'

Keep production credentials outside source control:

export PHFRAME_ENV=production
export PHFRAME_DATABASE_URL='postgresql+psycopg://user:password@localhost/phframe'
export PHFRAME_HOST=0.0.0.0
export PHFRAME_PORT=8000

phframe check
phframe migrate
phframe serve

Alternatively, phframe.yaml can refer to an environment variable:

project:
  database: ${DATABASE_URL}

PHFrame redacts database credentials in system-check output. SQLite and PostgreSQL use the same dataset, CRUD, import-audit, and migration APIs. Reload mode is intentionally disabled when PHFRAME_ENV=production.

Dashboard export

• Reads .csv, .xlsx, and .xlsm files
• Supports an interactive column-selection wizard
• Recognizes location, date, measured value, population, and category roles
• Calculates record count, completeness, totals, and rates per 100,000
• Generates grouped summaries, monthly trends, a data-quality table, and a data preview
• Writes one responsive HTML file that can be emailed or opened without a server

Install for development

Python 3.10 or newer is required.

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[dev]'

Quick start

Use the guided workflow:

phframe analyze examples/malaria_surveillance.csv --interactive --open

Or provide the column roles directly, which is useful for automation:

phframe analyze examples/malaria_surveillance.csv \
  --location district \
  --date report_date \
  --value cases \
  --population population \
  --category facility_type \
  --title "Malaria Surveillance Dashboard" \
  --output malaria-dashboard.html

For an Excel workbook, use --sheet 0 (the default), --sheet 1, or a sheet name:

phframe analyze monthly-report.xlsx --sheet Surveillance --interactive

Run phframe analyze --help for all options.

Development

pytest

Contributions are welcome. See CONTRIBUTING.md for the development workflow and SECURITY.md for responsible vulnerability reporting.

Project status

PHFrame follows semantic versioning after the 0.2.0a1 preview. See CHANGELOG.md for release notes and PLAN.md for the full architecture and phased roadmap.

Roadmap

See PLAN.md for the full architecture and phased roadmap.

License

MIT