Mario Marcolongo — Portfolio & Application CV System

Data Quality & Information Retrieval | AI Evaluation & Adversarial Testing | Scientific Fact-Checking & Evidence Synthesis
Information retrieval · evidence synthesis · AI evaluation · source verification

Live website: mariomarcolongo.com

Purpose

This repository generates Mario Marcolongo's public portfolio, evidence record, work-sample pages, machine-readable dossiers and role-specific application CVs.

The public positioning is intentionally ambitious and source-backed. It presents current strengths in exploratory AI evaluation, paid scientific verification, research and product delivery, and source-quality investigation without claiming independent software-development, penetration-testing or senior red-team-engineering experience that has not yet been demonstrated.

Career hierarchy

The homepage prioritizes the evidence most relevant to higher-upside roles and to realistic paths toward them:

1. AI evaluation and model behavior — exploratory adversarial testing, evaluation support, test planning and evidence reporting.
2. Scientific AI quality and research data — primary-source verification, provenance, metadata, domain-expert review and research support.
3. Trust, safety and source quality — source provenance, public-record investigation, structured-data review and public-source research support.
4. Research, editorial and community coordination — a credible bridge path for evidence synthesis, editorial production and participant-facing roles.

Longer-term targets include automated evaluation, AI-safety programs, evaluation infrastructure and higher-responsibility trust and safety work. Engineering-heavy roles remain a development target requiring independently understood Python, software-testing and security experience.

Public routes

• Portfolio (/) — readable metric-led evidence cards, three principal cases, supporting products and curated data artifacts, an explicit high-autonomy working profile, concise experience and targeted CV selection.
• Notandia / MDPI Filter continuity (/mdpi-filter.html) — stable canonical record connecting the former MDPI Filter branding and repositories to the current Notandia browser-extension and Zotero projects.
• AI Evaluation Record (/security.html) — scope, methodology, platform-reported Gray Swan activity, the live participant profile as the primary destination, dated evidence and explicit limitations.
• Source Verification Work Samples (/integrity.html) — provenance, health-information monitoring, entity reconciliation and structured-data cases.
• Research Operations Record (/research-operations.html) — source-backed public case study of accessibility-aware qualitative research support and facilitation.
• AI Evaluation & Model Behavior CV (/cv-resume.html) — two-page application document.
• Scientific AI Quality & Research Data CV (/cv-research.html) — two-page application document.
• Trust, Safety & Source Quality CV (/cv-integrity.html) — two-page application document.
• Research, Editorial & Community Coordinator CV (/cv-editorial.html) — two-page bridge document.
• Master CV (/cv.html) — comprehensive evidence archive; not the default application attachment.

All four role-based application CVs are verified as exactly two A4 pages. The master CV is intentionally comprehensive and may be longer. Former company-specific CV URLs redirect to the closest durable role-based document rather than remaining independently maintained.

Notandia / MDPI Filter continuity

Notandia is the public-facing rebrand and continuation of MDPI Filter. The canonical recruiter-facing evidence URL is /notandia.html; the legacy /mdpi-filter.html route redirects there so prior links continue to work.

The current canonical repositories are:

• notandia/browser-extension for Chrome, Microsoft Edge, Firefox and Safari;
• notandia/zotero-plugin for Zotero.

The build rewrites the retired github.com/orgs/mdpi-filter/repositories URL out of clickable links and CI fails if it reappears as an active hyperlink. Historical source repositories remain linked from the continuity page for provenance.

Focus-group research attribution

Marta Panzeri gave permission to identify her and the institutional context of the collaboration. The public record therefore names:

• Marta Panzeri;
• the Department of Developmental Psychology and Socialisation (DPSS);
• the University of Padua;
• the related public thesis context.

The description covers Mario's own contribution to remote focus-group facilitation with autistic participants, including structured prompts, participant-sensitive pacing, two-person handoffs, recruitment, literature research, technical preparation and privacy boundaries. It does not disclose participant identities, recordings or confidential session content, and it does not imply institutional endorsement.

The collaboration is retained as named experience and in the relevant CV; it is not given the same homepage prominence as the three principal high-upside evidence cases.

Evidence architecture

• data/source.js — canonical factual dossier and public evidence boundaries.
• data/application-profiles.js — role-specific CV selection and wording.
• data/portfolio-human.js — homepage hierarchy, evidence cards, explicit link destinations, working-profile language and application routes.
• data/public-evidence.js — dated, immutable public metrics shared across presentation layers.
• data/homepage-positioning.js — fresh homepage presentation model, isolated from role-specific CV overlays.
• src/pages/ — Astro source pages.
• src/styles/portfolio-v6-overrides.css — readable evidence-card system and corrected contact contrast layered over the established portfolio layout.
• scripts/generate-llm-dossiers.js — canonical machine-readable dossiers.
• scripts/postbuild.js — generated root mirrors and artifact-presence checks; source pages own their copy and presentation.
• scripts/verify-dist.js — generated-output, metadata and evidence assertions.
• scripts/verify-notandia-transition.js — continuity-page rendering checks and retired-link rejection.
• scripts/verify-rendering.js — desktop, tablet, mobile, light/dark and no-JavaScript rendering checks.
• scripts/verify-live-release.js — post-deployment checks for the public homepage, evaluation record and current dated evidence route.
• scripts/lighthouse-static-server.js — production-like compressed static server used by the Lighthouse gate.

Generated outputs include:

• llms.txt
• llms-full.txt
• cv-llm.txt
• static HTML mirrors for public routes
• four role-based two-page PDFs
• a comprehensive master-CV PDF
• responsive WebP portfolio previews
• page-specific Open Graph images and favicon assets

Gray Swan evidence

The live Gray Swan participant profile is the primary external destination. The current dated evidence page preserves the original 29 July 2026 screenshot showing Proving Ground rank #74, top 6%, and 113 platform-displayed total breaks, alongside the separate Arena metrics (#365 rank, 28 global unique breaks, 1,120 points and 255 submissions). The portfolio uses responsive WebP previews derived from that original PNG rather than the now-outdated synthetic SVG.

A locally retained same-day WACZ capture records the immediately preceding state at 17:09 UTC: the same #75/top-6% position with 109 total breaks, Arena rank #372, 26 unique breaks, 1,080 points and 245 submissions. Its hash and scope are documented without publishing potentially sensitive archive-request data. The 25 July account-holder snapshot and independently preserved 24 July Perma.cc record remain available as historical evidence. All counts are presented as time-stamped platform records, not independent vulnerability reproduction or certification.

Private phone number in local PDFs

The public repository and website do not contain a personal phone number. Locally generated PDFs read the ignored data/private.local.js contact config when it defines window.MARIO_PRIVATE.phone. To override it for one command, set CV_PHONE in the local environment:

CV_PHONE="+39 ..." npm run pdf

The generator injects the value at render time without writing it into tracked source files or public HTML. CV_PHONE takes precedence over the ignored local config.

After building, npm run preview serves a loopback-only local preview that injects the phone into the five public CV responses without modifying dist/. Use npm run preview:public when you need to inspect the exact public output without private contact data.

Verification and release gates

npm ci
npm run build
npm run verify:render
node scripts/verify-notandia-transition.js
npm run pdf
npm run verify:live

The build fails closed on stale positioning, unsupported claims, missing generated files, invalid structured data and inconsistent dossier mirrors. npm run verify:live separately checks the deployed homepage, AI-evaluation record and current dated evidence page, including rejection of the superseded Gray Swan caption.

Production normally deploys from pushes or merged pull requests. A deliberately named deploy/production-* pull request can safely refresh production from the trusted current main commit; the workflow never checks out code from that deployment-trigger branch while using production secrets.

The permanent Lighthouse 100 audit workflow is read-only. It builds the repository, runs the complete rendering and PDF suites, serves the production output with compression and cache behavior, and requires exact category scores of 100 for mobile and desktop in:

• Performance
• Accessibility
• Best Practices
• SEO

The workflow archives the complete Lighthouse JSON reports and diagnostic opportunities. Category scores are lab measurements rather than a guarantee that every real-world visit will always reproduce the same timings.

Public records

• ORCID 0000-0003-2846-7115
• Notandia / MDPI Filter continuity record
• Gray Swan Proving Ground profile
• Gray Swan evidence — 29 July 2026
• Historical Gray Swan evidence — 25 July 2026
• Historical Gray Swan capture — 24 July 2026
• Wikimedia CentralAuth record
• ENA accession PRJEB109744
• Yourself to Science

Licensing

Licensing differs by component and is documented at the relevant project, dataset or record level. Portfolio source code, datasets, images and editorial content must not be treated as sharing one blanket licence.