The AI Chronicle — founding brief

Working name; trademark, domain and publisher clearance still required. Created 15 October 2026.

The editorial promise

A general-interest newspaper about AI and its effects on work, science, public life, culture and the household. Readers should need curiosity, not technical credentials. The user's founding team is the publisher. The provisional imprint is The AI Chronicle Founding Desk; legal entity and human masthead remain to be supplied.

The first issue is a six-page, curated pilot covering 1–15 October 2026. It is a design and editorial demonstration, not an exhaustive scrape, a completed professional newsroom or a claim of independent interviews. The news sample is currently concentrated in Google and Anthropic announcements. Broader lab, independent, regulatory and India reporting must precede a public launch.

One editorial record, two modes

- Online: responsive, searchable in the future, source-linked articles organised by desk. The prototype has six working section routes and a full-text reading layout.

- E-paper: a frozen, dated issue with six page images and a selectable-text, linked PDF. PDF pagination is deliberate rather than an uncontrolled browser printout.

- Both are generated from issue.json. Correct the editorial record, rebuild and publish a correction notice. Do not quietly replace a published issue: retain the original PDF, increment its revision and link the correction.

- Prototype has no accounts, subscription billing, email capture, CMS or live feed. It does not collect reader data.

Six-page edition map

For a four-page day, merge World & Work into page 1 and Life & Learning into the back page. Preserve a complete India page. Move overflow online; do not shrink type to force it into print. A special investigation may replace a routine slot.

Topic and source type are separate fields. X and LinkedIn are discovery channels. A relevant social viewpoint may become Public Square; a verified release belongs to its subject desk. Never publish a weak story just to fill a category.

Typography and visual system

Original identity inspired by newspaper conventions; no copied masthead, proprietary newspaper fonts or reproduced article layouts.

- Print pilot: A4 portrait, 26pt margins, 17pt column gutters. A4 is easy to print at home; a larger tabloid edition can follow after review.

- Masthead: Georgia Bold, 37pt print / up to 79px desktop. Working wordmark, not a final logo.

- Headlines: Georgia Bold, approximately 29–30pt print / 39–64px web.

- Body: Georgia, 9.2pt in three columns or 10pt in two columns, 12.4–13.7pt leading; web 16px/1.7. Review body size with real readers before locking print size.

- Decks: Georgia Italic. Labels and captions: Arial. The website uses the reader’s installed Georgia/Arial with system fallbacks. No standalone font files are redistributed. The PDF embeds subsets for reading; a future brand font pair should have explicit publication rights.

- Paper #f7f3e9, ink #202b2b, oxide #9b3f2b. Hairline column rules; no decorative cards or promotional gradients.

- Original conceptual illustrations are labelled. Do not use generated pictures as evidence of events. Photography requires a rights/credit record, date, location and meaningful caption.

Collection and production cadence

The user proposed weekly collection and a daily paper. Recommended operating model: daily incremental collection, weekly source maintenance, plus a one-off 1–15 October backfill. Daily does not mean publish every story.

Proposed schedule, Asia/Kolkata: collection at 05:30; deduplication and shortlist at 06:00; human edit and verification at 06:30; edition lock at 07:15; release at 07:30. This is a proposal, not a configured automation. Publication stays manual until the selection and correction process works reliably.

A weekly digest can re-edit that week's daily records. Date filters use the original publication timestamp, not search indexing time. A source without a trustworthy date stays in the review queue. Keep publication, update, event and retrieval dates separate. Late updates belong in the next edition or a clearly dated correction.

News engine

Implemented starter: engine/collect.py, Python standard library + SQLite. RSS/Atom retrieval, URL canonicalisation, source attribution, date-window filtering, content-hash deduplication, persistent candidate storage and JSON export. It collects metadata and short feed descriptions; it does not yet write verified articles. Network and parsing failures are surfaced in a run report and an all-failed run returns a failure exit code.

Run from the chronicle directory:

python3 engine/collect.py --start 2026-10-01 --end 2026-10-15 --output candidates.json

Feed collection only returns what a feed currently exposes. It cannot reconstruct a historical fortnight that has fallen out of the feed. Backfill requires permitted archives, sitemaps, official newsroom indexes or search, followed by date verification. The pilot reporting was manually researched from source pages, not generated by this collector.

The starter deliberately does not schedule itself or auto-publish candidates. Feed URLs are configurable. X/LinkedIn/Product Hunt need permitted APIs, licensed feeds or manual submissions; no login circumvention. GitHub can use its public API within rate limits, with release dates and licence checks. Public star counts are volatile and do not constitute quality rankings.

Production architecture: adapters → raw source records → normalisation → story/event clustering → claim/evidence records → editor queue → approved articles → edition composer → web + PDF → correction log. Suggested next backend: Python jobs + PostgreSQL, object storage for rights-cleared media, and a small editorial CMS. SQLite is sufficient for this prototype.

Source registry and gaps

Priority 1: original research papers, official product releases, government/regulator documents, repository releases and creator originals. Priority 2: independent reporting to test company claims and add context. Priority 3: newsletters/social/launch boards for discovery and attributed opinion.

Initial monitoring targets: OpenAI, Google/DeepMind, Anthropic, Meta, Microsoft, NVIDIA, Hugging Face, Indian model builders; IndiaAI, MeitY, PIB and state sources; independent outlets with permitted feeds; arXiv and lab publications; selected GitHub projects, Product Hunt and creator channels. Register source URL, method, terms, access limits, geography, owner, last-success time and reliability notes.

Do not interpret silence from one source as an absence of news. Show source coverage and failed adapters in the editor dashboard. Seek regional-language sources for India. A full launch needs sources beyond company announcements and an editor who can make an independent judgment.

Selection and verification

Cluster by event, not just URL. A syndicated article and a press release about the same announcement should become one candidate event. The starter only deduplicates matching URLs/content; semantic event clustering is planned.

Proposed 100-point editorial rubric: consequence 30, new information 20, evidence 20, relevance to everyday readers 15, under-covered perspective 10, timeliness 5. Use it to support judgment, not impersonate objective truth. Reserve space across desks after scoring. Model-launch volume must not crowd out India, culture or family life.

Every reported article needs a claim list, original sources, dates, attribution, uncertainty notes and a human review state. Company performance claims should be attributed and independently tested or checked where possible. Contested or high-consequence claims require independent corroboration. Research summaries must preserve study population and limitations. Do not turn a target into an achieved result.

Opinion, analysis, sponsored content, service journalism and news need different visible labels. No invented quotes, interviews, audience metrics, tests or testimonials. A missing creative-work pick or social original should remain visibly unfilled, as in the pilot. The culture page does include a sourced music-platform news report.

Article contract

Required production fields: id, event_id, canonical_slug, headline, deck, body, desk, format, author/editor, published_at, updated_at, event_at, geography, audience, claims[], source_refs[], evidence_status, human_review, image_refs[], rights_status, corrections[], edition_ids[].

Source records: source_id, original_url, canonical_url, publisher, author, published_at, retrieved_at, excerpt, content_hash, access_method, rights_note. Store only content permitted by source terms. The website should link to original reporting and publish original summaries rather than reproduce paywalled articles.

Launch gates and next steps

1. Review this issue's identity, section balance and reading size; choose the human publisher name.

2. Add original India reporting, opposing perspectives and a verified creative work; diversify company coverage.

3. Confirm final font publication rights, image rights and working-name clearance.

4. Add editorial review states, event-level clustering, claim verification, archive backfill and run monitoring.

5. Add archive, article routes, search and correction history; test with parents, students and nontechnical readers.

6. Schedule daily collection only after source access and operating costs are agreed. Human editor approves the edition before publication.

Success measures: factual corrections per issue, source diversity, reader understanding, completion time, useful return visits, collection failures and cost per published story. Page views and raw article counts alone do not measure a useful newspaper.