Skip to content

About

A web tool that guides startup founders through a structured environmental impact assessment in five steps, no consultant needed. Two diagnostic questions classify how your environmental benefit arises, which drives the pathway you map, the indicators suggested, and how claims are evidenced. Exports a Word report and a save file to resume later.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Live app (GitHub Pages): https://apiholec.github.io/env_impact_tool/

Startup Environmental Impact Assessment Tool (v3.3)

v3.3: Czech version (ENG / CZ switch)

The whole web tool is available in English and Czech. An ENG / CZ switch sits at the top right of the homepage, in the sidebar and in the mobile top bar.

  • What is translated: the homepage, all five steps, the pathway diagram, the indicator bank, the checks and recommendations (with their "based on" traces), and the Word, PDF, HTML and Markdown reports.
  • Default: Czech for browsers set to Czech or Slovak, English otherwise. The choice is remembered in the browser (eia-lang).
  • Save files do not depend on the language. Answers are always stored under their English values and only displayed translated, so a file saved in Czech opens correctly in English and the other way round. Your own text (descriptions, claims, notes) is kept as you typed it.
  • Claim checks understand both languages. Hedged wording ("by mohla", "odhadujeme", "až o"), comparisons ("sníží", "o 18 %") and disclosed self-reports ("podle farmářů", "dotazník") are recognised in Czech as well.
  • The demo loads a Czech FieldSense example (docs/samples/FieldSense.cs.json) in Czech mode. It has the same answers as the English one.
  • Code: docs/js/i18n.js provides L(en, cs) for UI strings and tr() for stored values (stages, evidence levels, indicator names, units…); docs/js/page-cs.js holds the Czech homepage text. Literature references and the thesis title stay in English.

v3.1: customer profile and next steps (feedback from SIC, Oct 2026)

Changes to the web version, from the expert consultation with the Central Bohemian Innovation Centre (SIC):

  • Customer profile in step 1. Customer segment (B2B, B2C, B2G or a mix) and, separately, the revenue model (subscription, device + subscription, one-off sale, licensing, pay per use, fees, commission). Three roles are asked apart: the buyer, the day-to-day user, and whoever must change their behaviour for the environmental benefit to happen (a municipality may buy software that only saves fuel if its employees use it). Plus why customers buy and how that links to the environmental benefit. All optional, so existing save files still load.
  • Next steps and points to consider. Step 5 and every report format now end with concrete next steps drawn from the gaps in the assessment (weakest links, missing baselines, unmeasured core claims, aspirational indicators, a missing customer or commercial reason to adopt) and with "Have you considered…?" prompts that depend on the customer type, impact mechanism and stage. The rules are in SCORING.nextSteps in docs/js/framework.js.
  • Cause-and-effect diagram in step 2. The impact pathway is drawn in three lanes: what the startup controls, what customers and users must do, and the environmental effect. Each pathway step has a new question, "Does this step depend on customers?" (no / adoption / behaviour change), pre-filled from the template. Customer steps are marked in orange, and the arrow where the technology hands over to customers is highlighted: a good technological solution is not yet a viable product. The diagram is also in the Word and HTML/PDF reports (docs/js/diagram.js). If no customer step has solid evidence, step 2 warns and the report adds a next step.
  • Indicators that reflect real impact (step 3). Every indicator is tagged by what it measures: real-world outcome, adoption & use, technology / product only, own operations, or governance. Outcome and adoption indicators are highlighted. New bank entries: real-world outcome per participating customer, active use rate, customer retention, share of products actually returned or recycled, and (as a contrast) units produced or installed. These additions carry no thesis citation. Custom indicators must say what they measure. A new integrity check warns when no core indicator measures adoption or a real-world outcome, and the Word report's indicator tables gain a "Measures" column.
  • Claims & evidence (step 4). Measured / modelled / projected stays. Evidence quality (low / medium / high) is a separate rating: a self-reported survey is measured data of low quality. Each claim now has a baseline ("compared with what?") and its evidence sources, and shows the assumption of the pathway step it relies on (from step 2). A claim–evidence check flags gaps: projected claims worded as fact, comparisons without a baseline, measured claims of low quality, claims without sources, and indicators that contradict a weak step-2 rating. A "Claims and evidence at a glance" table sits at the top of step 4 and in section 4 of the report; gaps also feed the integrity checks and next steps. Word tables now keep their column widths.
  • Impact–business alignment (step 5, report section 6). Sets the customer value next to the environmental value, says whether the customer's reason to buy is tied to the environmental benefit, and lists only the commercial assumptions the benefit depends on (customers find it worth paying for, use turns into behaviour change, the saving and the impact come from the same change, rebound, retention, the adoption a projection assumes). The team marks each as not checked / some evidence / validated. Not a viability score and not business due diligence.
  • Structured recommendations (step 5, report section 7). Each recommendation states the evidence gap, what it means, the recommended action, and why it matters at the current development stage; they are prioritised (high / medium / low) by stage. Unvalidated alignment assumptions become recommendations too. "Points to consider" stay. Report sections are now: 1 profile, 2 pathway, 3 indicators, 4 claims & evidence, 5 integrity checks, 6 impact–business alignment, 7 recommendations, 8 review plan.
  • Traceable decision rules. Recommendations and points to consider come from a fixed rule set (SCORING.RULES in docs/js/framework.js: R01–R16 recommendations, S01 strength, C01–C17 points to consider), not from a language model. Each recommendation shows its rule ID, the detected condition, the recommended action, why it matters, why now, and "based on": the exact assessment answers that triggered it. A strength rule (S01) fires when a claim is measured, high quality, sourced and documented, suggesting an evidence summary. Step 5 has a "How recommendations are made" table of all rules and which ones apply, and the report ends with the same table as an appendix.
  • Positioning. The homepage now leads with "Understand your impact. Strengthen your claims. Know what comes next." and addresses founders first, with a "Who it is for" section (founders as primary users; accelerators and incubators; impact-oriented investors; conventional investors where environmental impact is relevant), instead of framing the report mainly as something to show investors.
  • Ready-to-explore demo. "Explore a finished example" on the homepage opens the fictional FieldSense assessment (docs/samples/FieldSense.json) in one click, with a banner and a "Start your own assessment" button. It asks before replacing a real assessment in progress.
  • Calculation boundaries. Each claim in step 4 records what the figure includes and leaves out (period, sites or customers, life-cycle stages); measured or modelled claims without one get a note.
  • Proportional reassurance (review round 2). The alignment headline says "Potential alignment: n of m assumptions verified" until every critical assumption is validated; "saves them money" is split into savings on the same resource (counts towards alignment) and savings in labour or time (does not). The sidebar shows "Form completion", explicitly not evidence; step 5 adds "Evidence readiness" counts (measured outcomes, high-quality claims, assumptions verified, issues), never a score. "No gaps" became "No issues detected by these automated checks".
  • Stricter checks. Blank claims are "not checked" rather than clean; the evidence-summary strength (S01) needs a written claim, boundary and a valid pathway link; the wording prompt checks each comparative sentence, and "can"/"will" no longer count as hedges; missing quality or boundary is a warning; pathway links to deleted steps are reported.
  • Action plan. Step 5 leads with the three highest-priority actions, each with owner, deadline and "done when / evidence"; the reasoning is expandable and the rest sit under "More recommendations". Duplicate weak-link recommendations are merged. The report opens section 7 with the action plan table.
  • Classification examples. The orientation question has contrasting examples and a note for products where both answers seem true.
  • Save files gain optional keys: customer_types, revenue_models, customer_profile (the buyer), customer_user, behaviour_actor, value_drivers, customer_value, custom_kinds, change on each pathway stage, kind on each selected indicator, and quality, baseline, sources, boundary on each claim, alignment and plan. Older files load without them.

v3.0: static web version on GitHub Pages

The tool now runs as a static website in docs/, served by GitHub Pages. There is no server: everything runs in the browser, which fixes the Streamlit version's reload lag, dark-mode rendering glitches and cold starts.

  • Same framework, same rules. docs/js/framework.js is a direct port of framework/reference.py and framework/scoring.py; docs/js/report.js ports framework/report.py (Markdown, HTML, Word and JSON).
  • Save files: the web version reads Streamlit save files (same startup-env-impact-assessment format). The reverse is no longer true: the Streamlit version does not know the fields added in v3.1 (customer, revenue model, alignment, action plan, evidence quality, boundaries…) and drops them on export. The Streamlit version is retired; use the web version.
  • Progress saves automatically in the browser (localStorage), so a reload or closed tab loses nothing. The JSON save file is still the way to move an assessment between devices or keep it long-term.
  • Word export is generated in the browser with the bundled docx library (docs/vendor/), with headings, a table of contents, indicator tables, colour-coded confidence badges and page numbers. PDF export uses the browser's print dialog.
  • Works on phones: the step list becomes a slide-out drawer.
  • Not carried over: the optional AI assistant panel. It needs a secret API key, which a public static site cannot keep private.

To edit content (wording, indicators, rules), change docs/js/framework.js. To preview locally:

python -m http.server 8000 --directory docs

then open http://localhost:8000. Pushing to main redeploys the site.

Tests for the browser rules: node --test tests/web/*.test.mjs (no dependencies; also run by GitHub Actions on every push). The Python tests in tests/ cover only the retired Streamlit implementation.

The original Streamlit app (below) is retired: unchanged, not maintained, and it does not have any of the v3.1 features.


Streamlit version (v2.1.1)

Fixed bugs (v2.1.1)

  • Unclear/black input fields — inputs, dropdowns, and labels now forced light with white backgrounds, visible borders, and dark semibold labels regardless of the browser's dark-mode setting.
  • Development stage not updating dynamically — the sidebar stage card now updates in the same interaction, and the stage guidance also appears directly under the selector on step
  • Diagnostic classification too subtle — each diagnostic question now has its own bordered panel with a bold question and high-contrast answer options.
  • Diagnostic classification skippable — no more preselected answers; "Save & continue" stays disabled and steps 2–5 are blocked until the name and both questions are answered.
  • Progress counted by navigation, not completion — a step is only marked ✓ when its content is actually filled in (e.g. at least one pathway stage described); merely visiting a step does nothing.
  • Navigation looks inactive/faded — every step in the rail now shows a completion-percentage chip, full-contrast text, hover states, and a checkmark only when genuinely complete
  • Assessment progress positioning — the overall progress bar moved to the left sidebar, directly under the step list.
  • Export not accessible earlier — the sidebar now has an Export section available from any step (once step 1 is minimally complete).
  • Export flow too complicated — the complete Word report is a single click, both from the sidebar and on step 5; other formats are secondary.
  • Technical terminology too difficult — a "Key terms, in plain language" card on step 1 explains mechanism, orientation, impact pathway, and hybrid in non-expert wording.

Startup Environmental Impact Assessment Tool (v2.1)

v2.1 adds a polished visual design: branded sidebar with a numbered workflow rail, a contextual right-hand panel (live classification preview, progress, guidance cards), hero headers, and a warm paper/forest-green theme. All functionality is unchanged from v2; all widget keys are stable, so saved JSON assessments from v2 load without modification.

An interactive implementation of the four-module framework from the thesis "A Proposed Framework for Environmental Impact Assessment in Early-Stage Startups." Screening-level by design: it does not replace full LCA or GHG Protocol reporting.

Run it

pip install -r requirements.txt
python -m streamlit run app.py

Project structure

docs/                      — the web app (GitHub Pages)
  index.html, css/         — homepage and styles
  js/i18n.js               — language switch, L(en, cs), tr() vocabulary
  js/framework.js          — framework content and rules (bilingual)
  js/diagram.js            — pathway diagram
  js/report.js             — Markdown / HTML / Word / JSON reports
  js/page-cs.js            — Czech homepage text
  js/app.js                — the 5-step wizard
  samples/                 — FieldSense demo (EN and CS)
tests/web/                 — Node tests for the web rules, also in Czech
--- retired Streamlit version ---
app.py                     — Streamlit UI (the 5-step wizard)
ui.py                      — visual layer: design tokens, CSS, card
                             components, progress heuristic
.streamlit/config.toml     — theme (palette, fonts)
framework/
  reference.py             — all framework content: typology, diagnostic
                             questions, pathway templates, indicator bank
                             (with citations), stage guidance, emission
                             factors, references
  scoring.py               — feasibility–relevance bucketing + automated
                             integrity (anti-greenwashing) checks
  report.py                — Markdown / HTML / **Word (.docx)** / JSON
                             report builders and save/load persistence
tests/test_framework.py    — unit tests for scoring, persistence, and the
                             Word export (including edge cases with no
                             data filled in)

Content and logic are deliberately separated from the UI: to refine wording, add indicators, or change a rule, edit framework/ — the app picks it up automatically.

What changed vs. v1 (and why)

Change Grounding in the framework
Diagnostic classification questions instead of abstract self-labels ("if nobody changes behavior, does the benefit still occur?") Reduces misclassification; Module 1 drives everything downstream
Development-stage selector with adaptive guidance (priority modules starred per stage) Figure 11: module depth varies across ideation → growth
Dual-track pathway template for general-purpose startups (operational footprint + product externalities, incl. rebound effects) Figure 8 lists three pathway structures; v1 only had two
Per-stage assumptions + evidence strength in the pathway; weak links surface automatically Theory-of-change logic: assumptions must be explicit; weakest links are where contribution analysis is needed
Indicator details: unit, current value, target, frequency, data source, and a link to the pathway stage it evidences Materiality: indicators must trace to the impact pathway, not float free
Visual Feasibility–Relevance matrix (Figure 9 rendered live) Makes the scoring outcome inspectable at a glance
Quick Scope 1+2 estimator with generic factors, output explicitly labelled Modelled Demonstrates the measured/modelled distinction in practice; screening-level GHG accounting
Automated integrity checks: vanity metrics (high feasibility / low relevance), core set >5, projected-claims-as-evidence, unlinked indicators, undocumented assumptions Directly operationalizes the empirically observed greenwashing drivers
Save / load + versioning + next-review milestone Principle 4: designed for updating, not completion — assessments are revisited at milestones, each reload starts a new review cycle
Professional report: executive summary, integrity-check results, methodology note, reference list; exports as Word (.docx), HTML (printable to PDF), Markdown, or JSON Credible communication to investors/incubators; academic traceability
Package structure + unit tests Auditable, extensible codebase

The Word (.docx) report

The recommended export format. Built with python-docx at download time — no separate conversion step, no LibreOffice/Word installation needed on the user's machine. It includes:

  • Real Word heading styles (Title, Heading 1–3), so the document has a proper outline in Word's Navigation Pane
  • A table of contents field — on first opening in Word, right-click it and choose "Update Field" to populate it (standard behaviour for any TOC generated outside Word itself)
  • Styled tables for each indicator bucket (zebra-striped, with feasibility/ relevance/unit/target/data-source/pathway-link columns)
  • Colour-coded confidence badges — green [MEASURED], amber [MODELLED], purple [PROJECTED] — and colour-coded integrity-check results, matching the in-app HTML report's visual language
  • Page numbers in the footer

If to_docx() ever raises an exception, the app catches it and shows the error inline on the Review page rather than crashing — the other three export formats stay available regardless.

To verify template changes by eye rather than just by tests:

python -c "
from framework import report
data = report.assemble(some_state_dict)
open('out.docx', 'wb').write(report.to_docx(data))
"
soffice --headless --convert-to pdf out.docx
pdftoppm -jpeg -r 100 out.pdf page
# then open page-1.jpg, page-2.jpg, ...

Notes & limitations

  • Emission factors in the quick estimator are generic, indicative values for screening only. Replace with national, year-specific published factors (e.g. your national inventory, DEFRA, EEA) before any external reporting. All estimator outputs are modelled claims.
  • State is in-memory per browser session. Persistence works via the JSON export/import (which also implements the milestone review cycle). For a public multi-user tool, add a database (SQLite → Postgres) and simple auth.
  • Reference list: several thesis citations are included with author/year only where the full bibliographic details weren't in the extracted text — complete them from your thesis bibliography in framework/reference.py → REFERENCES.

About

A web tool that guides startup founders through a structured environmental impact assessment in five steps, no consultant needed. Two diagnostic questions classify how your environmental benefit arises, which drives the pathway you map, the indicators suggested, and how claims are evidenced. Exports a Word report and a save file to resume later.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages