CoARA Action Planner

Self-assess your institution against the ten CoARA commitments and get a prioritised, editable reform action plan — entirely in your browser.

A static React app. Answer 24 diagnostic questions, and it scores your institution's maturity across the ten commitments of the CoARA Agreement on Reforming Research Assessment, then generates a prioritised action list and a corpus-grounded, editable action-plan narrative you can export. Nothing is uploaded; everything runs client-side and persists only in your browser's localStorage.

Built for the people who actually have to write the plan — research offices, institutional leadership, reform working groups, and anyone drafting a CoARA action plan for submission.

🔗 Live: https://rijdho.github.io/coara-action-planner/

Available in English, Spanish, French and German (auto-detected, switchable).

This is the open twin of the hosted Reform Assessment toolkit. That family keeps its calibrated methodology server-side; this repository moves the same questions, actions and prioritisation algorithm into the browser, in readable form, so the assessment logic can be inspected, cited, and adapted. Inspired by the open, community-first philosophy of Metadata Game Changers.

What it does

Five steps: Start → Questionnaire → Plan → Results → Report.

flowchart LR
    accTitle: The five steps of an assessment
    accDescr: Start names the institution and sets up one perspective per respondent role. Each perspective answers the same 24-question questionnaire, and the tool consolidates the answers into one profile while surfacing where the readings diverge. Plan applies optional tuning, Results scores maturity and ranks actions, and Report drafts an editable action-plan narrative.

    start["🏛️ Start<br/>institution, perspectives"]
    questionnaire["📋 Questionnaire<br/>24 questions, 0–5 maturity"]
    consolidated["🔀 Consolidated profile<br/>divergence surfaced"]
    plan["🎚️ Plan<br/>horizon, context, targets"]
    results["📊 Results<br/>radar, gaps, ranked actions"]
    report["📝 Report<br/>editable narrative"]

    start --> questionnaire
    questionnaire -->|"once per respondent role"| consolidated
    consolidated --> plan --> results --> report

    classDef input fill:#dbeafe,stroke:#2563eb,stroke-width:2px,color:#1e3a5f
    classDef derived fill:#fef9c3,stroke:#ca8a04,stroke-width:2px,color:#713f12
    classDef output fill:#dcfce7,stroke:#16a34a,stroke-width:2px,color:#14532d

    class start,questionnaire,plan input
    class consolidated derived
    class results,report output

Loading

• Start — name the institution and set up perspectives: answer the questionnaire once per respondent role (e.g. research office, leadership, a working group), and the tool consolidates them and surfaces where their readings diverge.
• Questionnaire — 24 diagnostic questions mapped to the ten commitments, scored on a 0–5 maturity model (Unaware → Aware → Exploring → Planning → Implementing → Embedded).
• Plan — optional tuning: time horizon, institutional context, priority commitments, whether to include high-effort actions, and per-commitment target levels (your ambition).
• Results — a maturity radar, a per-commitment breakdown, and a prioritised action list (each action ranked by the size of its gap × expected impact, adjusted for effort, context and ambition). Export the profile as PNG, or the full input state as a reproducible JSON config.
• Report — a pre-structured, editable CoARA action-plan narrative drafted from your answers, with the responsible unit / timeframe / indicator fields left as blanks to fill in. Every section carries the matching CoARA guiding question, quoted from the Secretariat's Action Plan Guidelines (October 2023), so the draft can be checked against them line by line. All ten commitments get a heading whether or not the tool had actions to suggest. Copy, download (.md/.txt), or print.

The methodology is the point

The calibration lives in src/data/, in plain readable JavaScript:

File	What it holds
questions.js	24 diagnostic questions, each with 0–5 answer options mapped to a commitment
commitments.js	the ten CoARA commitments + the 6-level maturity model
actions.js	47 recommended actions (each with fromLevel/toLevel/effort/impact, a theme key, real institutional examples, and a planText — the action restated as institutional first-person prose for the generated plan) and the prioritiseActions algorithm
plan.js	the second ranking stage — the ambition gate, the high-effort filter, and every plan-setting weight (context affinity, focus, horizon, target, divergence) and the applyPlan function
evidence.js	per-theme prevalence across the 335-plan corpus (THEME_FREQUENCY + the universal / common / emerging / frontier bands shown as "N% of 335 plans" on Results) — measured in corpus/, which publishes the manifest, the keyword signatures and the pipeline
guidingQuestions.js	the CoARA Secretariat's 19 guiding questions, quoted verbatim and mapped onto the report's sections
context.js	6 institutional contexts that re-weight priorities
perspectives.js	the 10 respondent roles and their ROLE_WEIGHTS — how divergent readings are consolidated
i18n/{es,fr,de}.js	full Spanish / French / German overlays of the above

Ranking happens in two stages, and both are calibration. prioritiseActions scores an action on its own merits — gap x impact, minus an effort penalty. applyPlan then applies what the user asked for on the Plan tab: it filters (high-effort actions, and a hard ambition gate that drops anything already at or beyond a declared target) and re-ranks (context affinity, focus, horizon, target, divergence). What Results and Report display is the output of both, so both are pinned by tests.

One consequence worth stating plainly: choosing an institutional context can move a single action by at most +4 — +2 if the action carries that context tag, +2 more if its commitment is one the context prioritises. That is less than a full maturity step at high impact, so context re-orders actions within a band rather than overriding the assessment. The ceiling is a named constant and a test, not an accident.

The questions and actions were hand-calibrated by reading 15 real institutional action plans (UCM, Helmholtz, DCU, UCLouvain, AQU Catalunya, FRQ, LBG, SocRSE, Eurodoc, YUFE/UNIRI, UB, UPC, OGS, SDU, U. Pannonia) and cross-checked against a corpus of published CoARA action plans from Zenodo — 314 at the April 2026 calibration, 335 as of the August 2026 recount. It is offered as a starting point to inspect and adapt, not as an authoritative scoring — see Caveats.

Run locally

npm install
npm run dev        # http://localhost:5173

Build the static site:

npm run build      # → dist/
npm run preview    # serve the build locally

No backend, no API keys, no tracking.

Tests

The calibration and the prioritisation algorithm are covered by unit tests (Node's built-in runner, no dependencies beyond what the app already needs):

npm test          # or: node --test tests/*.test.mjs

prioritise.test.mjs asserts on exact priority scores, and those cases double as the parity contract with the server-side engine behind the hosted sibling: the same answers must produce the same ranking on both. A change that moves a number here should be mirrored there or documented as a deliberate divergence. The algorithm cases use synthetic actions on purpose, so recalibrating the real catalog cannot break tests that are about the maths.

calibration.test.mjs guards the failures that are silent rather than loud — a mistyped theme still renders, it just quietly loses its "N% of 335 plans" evidence band; an action whose fromLevel is not below its toLevel can never be recommended at all; a commitment with no entry-level action tells an institution it is weakest there and then offers nothing to do about it.

plan.test.mjs covers the second stage — the ambition gate that removes actions, and each plan weight asserted as a literal score rather than in terms of the constant it is testing (a suite written the other way passes even when the constant changes, which is how the first draft of this file failed its own mutation check).

guiding-questions.test.mjs treats the CoARA guiding questions as what they are — a quotation. It pins the published count, checks every commitment is reached, and records the one place the source text is knowingly cleaned up rather than silently corrected.

corpus-parity.test.mjs pins the two copies of the corpus evidence together — corpus/data/theme-frequency.csv, where prevalence is measured, and src/data/evidence.js, which the app reads. A recount that updates one and forgets the other would leave the tool quoting figures its own corpus no longer supports, and nothing would render differently. Keeping the corpus in this repository rather than its own is what makes that check possible.

i18n.test.mjs pins the most fragile contract in the repository: the Spanish, French and German action overlays align with ACTIONS by array index. Inserting an action mid-list without inserting one at the same position in all three overlays shifts every later translation onto the wrong action — nothing throws, the app just shows the wrong text in three languages.

Deploy

Any static host works (the build is self-contained with relative asset paths). This repo ships a GitHub Actions workflow (.github/workflows/deploy.yml) that builds and publishes to GitHub Pages on every push to main. Enable it once under Settings → Pages → Source: GitHub Actions.

The app is client-rendered, so crawlers that do not run JavaScript (LinkedIn, Slack, Mastodon) see only index.html. public/og-card.png is the link-preview image referenced by the Open Graph tags there; regenerate it with node docs/og-card.mjs whenever docs/maturity-profile.png changes, or previews will keep showing the old radar.

Caveats

• Directional, not authoritative. The maturity scores and action rankings are a structured prompt for institutional reflection, not a certification. Two institutions at the same "level" can be in very different places.
• The radar is a shape, not a score. With ten axes in a fixed order, the polygon's area and outline carry no meaning — read the per-commitment numbers, not the picture.
• Keyword-derived corpus evidence. The "N% of 335 plans" figures on Results come from keyword matching over the full text of the published action plans — read them as directional bands (near-universal / common / emerging / frontier), not exact counts. Low prevalence is not a reason to skip an action: frontier practices are an opportunity to lead. corpus/ publishes which plans were read, the exact keyword signature behind every theme, and the pipeline that reproduces the counts, so the figures can be checked rather than trusted — and recounted: the loop is incremental, and the last recount ran on 6 August 2026.
• The guiding questions are a resource, not a template. CoARA publishes no reporting or action-plan template for members, and the guidelines say the guiding questions "do not serve as a rigid template but rather as a resource and suggestion". Printing them beside each section makes a draft easier to check; it is not a compliance claim, and answering them all does not make a plan approved by anything.
• Your data stays local. Because everything is in localStorage, clearing your browser data erases your assessment. Use the Export config button to save a reproducible copy.

License

Copyright (C) 2026 Ricardo Hartley Belmar.

• Code: AGPL-3.0-or-later. Read, cite, fork and adapt freely; if you run a modified version as a network service, publish your changes under the same licence. Releases up to 1.2.0 were published under MIT and remain so.
• Calibration data (src/data/, corpus/data/): CC BY 4.0. Reuse and adapt with attribution.

Neither licence covers the action plans the corpus measures: those remain under the terms each depositing institution chose. corpus/ records where every one of them lives; it does not redistribute them, and will not.

Citation

If you use this tool or its calibration, please cite it — see CITATION.cff or the "Cite this repository" button. Archived on Zenodo: concept DOI 10.5281/zenodo.21492548 (always resolves to the latest version).