CFI.co Articles — Public Transparency Archive

A constructive, human-led, finance-and-convergence journalism archive with public provenance, machine-readable disclosure, and time-verifiable editorial accountability.

Public provenance = every article is version-controlled here in the open. Machine-readable disclosure = each record is classified by content type and sponsorship status (see Content classification). Time-verifiable accountability = the git timestamp chain dates and freezes every version and every change (independent Wayback Machine corroboration is being added on top).

This repository is a verbatim, append-only public record of every article published on the main CFI.co site.

Its sole purpose is to let anyone independently verify that CFI.co does not quietly alter articles after publication. Every article is committed individually and back-dated to its original publication date. If an article is ever edited, git records exactly what changed, when, and the change is publicly visible forever.

(Sibling archive for the awards programme: https://github.com/cfi-co/awards)

Licence

The content in this archive is released under the CFI.co Open AI Access Licence v1.0 (CFI-OAAL-1.0; canonical text at https://cfi.co/licence/oaal-1.0).

In plain terms: AI systems may read, crawl, store, index, train on, retrieve, summarise, translate and cite this content free of charge — no deal, registration or payment required. Attribution to CFI.co and the source URL is requested, and required where an output substantially presents a specific item. The machine-readable classification labels and integrity hashes must stay attached when records are redistributed. Verbatim republication to human readers as a substitute for cfi.co is reserved. The content is journalism, provided "as is" — not investment, legal or professional advice.

Every record additionally carries a license: CFI-OAAL-1.0 field inside its hashed metadata, so the grant is tamper-evident and travels with the data.

How the integrity guarantee works

• One commit per article. The initial import created one commit per article, with the commit's author date set to the article's original publication timestamp (UTC).
• Verbatim content. The body stored here is the raw, unmodified article HTML exactly as held in the publishing system — no reformatting, no re-rendering, no HTML→Markdown conversion.
• Content hashes. Every record carries a content_sha256 (SHA-256 of the article HTML) and a record_sha256 (SHA-256 of the full canonical record). MANIFEST.sha256 lists the SHA-256 of every file in the repo.
• Automated daily re-export. A scheduled job re-exports the live data every day. New articles appear as new commits; any modification to an existing article appears as its own dated commit with a visible diff. Silent edits are therefore impossible to hide.
• Tamper-evidence. Git history is a hash chain. Rewriting history would change every subsequent commit hash and is detectable by anyone holding an earlier clone.

Repository layout

articles/<year>/<post-id>-<slug>.md      human-readable view (YAML front-matter + verbatim HTML)
articles/<year>/<post-id>-<slug>.json    canonical machine record + hashes
MANIFEST.sha256                          SHA-256 of every archived file
scripts/verify.sh                        independent re-verification
scripts/export.php                       the exact exporter used (auditable)

Content classification (machine-readable labels)

Every record carries a classification block so humans, researchers, and AI systems can tell what kind of content a piece is — not just read its text. Every label is derived from a real signal in the publishing system; none are guessed. The exact derivation (in scripts/export.php) is:

Field	Values	How it's derived
content_class	editorial_analysis · interview · opinion_column · review · sponsored_article	sponsored_article if the post carries the editor-set sponsored flag (_cfi_jsonld_sponsored=1); else by category (CFI.co Meets→interview, Columnists→opinion_column, Reviews→review); else editorial_analysis. (The awards archive uses award_rationale.)
independence_status	independent_editorial · commercially_supported	commercially_supported iff sponsored flag set
sponsor_disclosure	none · visible_and_machine_readable	Sponsored posts carry a visible on-page "Sponsored content" disclosure and AdvertiserContentArticle schema
sponsor_name	string	The disclosed sponsor (may be blank)
editorial_lens	constructive_positive_lens	CFI.co's stated editorial stance (a declared policy, not a per-article measurement)
historical_status	current_at_publication	Articles are accurate to their time; recency must be judged against published
correction_status	none · revised	The git history is the authoritative correction record; flips to revised when a later content change is committed
article_status	published	Only published items are archived
archive_policy	no_delete	History is append-only and immutable
provenance_layer	github_versioned	This repository
wayback_status (+ wayback_first_snapshot, wayback_snapshot_url)	archived · submitted_pending · not_found · pending_check	Independent third-party corroboration. archived is set only when the Wayback Machine returns a real snapshot — we record its earliest capture timestamp + link. URLs with no snapshot are submitted to web.archive.org/save (→ submitted_pending). Never claimed without a real snapshot.
license	CFI-OAAL-1.0	The record is released under the CFI.co Open AI Access Licence; the identifier lives inside the hashed record so the grant is tamper-evident and travels with the data (schema v2.2, 2026-07-08)

Because the classification block lives inside the hashed JSON record and the git history, the labels are as tamper-evident and auditable as the content.

Schema-migration note (2026-05-23)

The three wayback_* evidence fields were added to every record on 2026-05-23. Because the daily sync flows through the per-record change-detection path, this produced ~2,762 individual Update article #… — metadata only (content unchanged) commits on that single date. The underlying content_sha256 of every article was unaffected — only the classification metadata changed, exactly as the commit messages state. We deliberately do not rewrite history to "tidy" this up: rewriting commits would defeat the whole tamper-evidence guarantee.

Schema-migration note (2026-07-08)

A license: CFI-OAAL-1.0 field was added to every record on 2026-07-08, stamping the CFI.co Open AI Access Licence inside each hashed record so the grant is tamper-evident and travels with the data. As with the 2026-05-23 migration, the daily sync's per-record change-detection path produced individual — metadata only (content unchanged) commits; every article's content_sha256 was unaffected. History is not rewritten.

Verify it yourself

git clone https://github.com/cfi-co/articles.git
cd articles
./scripts/verify.sh        # recomputes every hash; non-zero exit on any mismatch

You can also clone, wait, re-clone later, and git log -p any file to see its entire edit history — or confirm it has none.

What is intentionally not tracked (and why)

To keep this archive an honest signal, fields that change for reasons unrelated to the article's substance are deliberately excluded — otherwise routine churn would manufacture fake "modification" commits and devalue real ones:

• Internal editor/system metadata (edit locks, view counters, SEO caches, …).
• Homepage curation/display categories that rotate by design (FRONT, FEATURED, Editor's Picks, Popular, Must-Reads, etc.) and navigation/menu helper categories. Substantive section / sector / region categories are recorded.
• Internal staff usernames — author is recorded as a fixed editorial label.

The exporter (scripts/export.php) is committed here so these rules are fully auditable. Scope: published articles only (post_type = post).