Three names, three jobs

Volare — the system and the ledger. The permanent record of every event observed, the evidence behind it, the judgment applied to it, and what happened next. This is the asset.

Gladys — the routing classifier. Reads a raw event and returns which desks should carry it and how important it is on each, as one strict JSON object. This is the engine.

Dispatch — the published product. Trade titles feeding a national front page, and the reason there is a newsroom generating judgment at all. This is the proving ground.

The titles

Ten publications, one front page. Every vertical owns one corner of the content industry and files into the flagship. Six sit in the classifier’s routing vocabulary today, which is why a single event can carry a different importance on each of them at the same moment.

Dispatch Daily — the national front page for the business of content. Elevates the most consequential story from anywhere in the network. Flagship.

Content Dispatch — corporate newsrooms, owned media, newsroom operations and the AI systems running them. Live.

Capital Dispatch — funding, M&A, private equity, earnings and the economics of the media business. Live.

Creator Dispatch — the creator economy: representation, deals, tooling and audience economics. Live.

Channel Dispatch — platforms, algorithms and distribution. YouTube, TikTok, streaming, newsletters, search. Live.

Copyright Dispatch — IP, licensing, litigation and AI training rights. Live.

Capitol Dispatch — political content, campaign spending, advocacy and AI-generated politics. Live.

Career Dispatch — jobs, appointments, salaries and hiring trends. A publication and a labor dataset at the same time. Named, not yet routed.

Communications Dispatch — corporate communications, PR, agencies, crisis and executive positioning. Named, not yet routed.

Creative Dispatch — creative leadership, production, design, advertising. The work itself. Named, not yet routed.

The pipeline

Every event runs the same route. Phase 3 is where it divides: the reader gets a story, and the ledger keeps the event, the evidence and the judgment applied to it.

Phase 1 — Ingestion and entity resolution

1. API pull feed (raw JSON payload) — filings, releases, dockets, notices, postings. Each source arrives with its own identifiers and its own clock.
2. De-duplication (hash cross-check) — one real event usually arrives four ways. They collapse to a single event carrying multiple observations, each keeping its own source and timestamp.
3. Entity mapping (ticker proxy resolution) — every name resolves to a canonical key. Without it, three spellings are three companies and every series built on top is wrong.

Phase 2 — Text processing and taxonomy

4. NLP text parser (boilerplate removal) — strip the legal furniture and the standing language so what remains is the part that changed.
5. Taxonomy tagging (topic classification) — closed vocabularies matched byte-exact. Anything outside them quarantines the record rather than publishing a junk tag that would appear nowhere and report success.
6. Entity extraction (NER validation) — who is actually in this event, and in what role. Counterparties matter as much as the filer.

Phase 3 — The forked output: media vs exhaust

7. Media generation (summary synthesis) — what the reader gets. Wire line, brief, story, and the promotion decisions made on top of them.
8. Exhaust extraction (structured signal) — the values, entities, dates and judgments the story used but never printed. This is the part that becomes data.
9. Ledger packaging (cryptographic proof) — the record is sealed with its evidence and its timestamps so the observation can be proven later rather than asserted.

Phase 4 — Marketplace delivery and provenance

10. Sovereign ledger (provenance log) — append-only and point-in-time. Nothing is edited in place; corrections are new rows pointing at what they supersede.
11. Secure API (marketplace gateway) — daily snapshot plus intraday delta, stable event IDs across restatements, point-in-time files.

The standing rule — verify first, publish second. Anything above a wire line is checked against how Bloomberg, Reuters and AP framed it. Late on our version is acceptable. Wrong in the permanent record is not.

Two outputs, one record

Media — what the reader gets

• Wire alert at machine speed
• Headline and summary
• Story or analysis, verified before it runs
• Promotion to the front of a title
• Related context across the network

Exhaust — what the ledger keeps

• Base event record: identity, provenance, values, entities
• Editorial intelligence: importance by desk, escalation, placement
• Enrichment: sourced, dated, usually unpublished
• Human decisions: proposal, override, reason code
• Historical comparison: is this accelerating or spreading

How a story gets ranked

Every event is scored independently on each desk, because the same news is routinely a passing note on one beat and the lead on another. A filing enters at the bottom of the ladder and climbs only when something in the world pushes it up. Most items never move.

The ladder, bottom to top

• Index — recorded for the archive. Deliberately not displayed.
• Watch — not a story yet. Monitored because it may become one.
• Brief — notable. Worth logging, not worth stopping for.
• Major — significant. A reader on that beat has to know it.
• Lead — the most consequential item of the day for that desk.

What the event itself carries

• Magnitude (heavy) — dollars, jobs, audience reach, or the size of the change against what came before.
• Novelty (heavy) — first of its kind versus the ninth version of a routine disclosure.
• Proximity (medium) — how close the event sits to that desk’s core beat.
• Velocity (medium) — whether related events are stacking up around the same companies.
• Certainty (gate) — doesn’t raise the rank. Caps it. Nothing unverified leads.

What other newsrooms tell us

• Byline composition (heavy) — not the count. Two reporters from different beats means a rival newsroom decided this crosses sectors.
• Second-day follow-up (heavy) — almost nothing gets one by accident. The strongest confirmation available.
• Push alert (heavy) — the strongest escalation a newsroom has, and precisely timestamped.
• Dwell (medium) — hours held in the top slot. Six hours was believed. Twenty minutes wasn’t.
• Concurrence (medium) — how many reference outlets carried it high at the same hour.
• Time to match (medium) — how fast rivals chase it, and whether they credit the original.
• Art investment (light) — original photography means budget was committed before publication.
• Absence (light) — a big-looking item running unbylined means that outlet doesn’t believe it.

Worked example — a publisher discloses AI licensing revenue

One event, one day, three desks.

• 06:01 — investor release crosses. Revenue figure disclosed with a year-over-year change. Nothing else known yet. → Capital: brief
• 06:14 — the figure is material against prior guidance. Magnitude and novelty both register; this line item didn’t exist two years ago. → Capital: major
• 08:20 — two reference outlets carry it above the fold. Concurrence confirms it isn’t a one-desk read. → Content: brief
• 09:45 — a rival runs it with two bylines: the media reporter and the IP reporter. The story has crossed a sector line. → Copyright: major
• 11:30 — push alert goes out. The outlet woke people up for it. → Capital: lead
• Next morning — two outlets run second-day follow-ups on licensing terms rather than the number. The precedent, not the revenue, is now the story. → Copyright: lead

Every step above is kept with its timestamp and its evidence, including what we thought at 06:01, before any of it was known.

The editorial signal layer

Newsrooms broadcast their own judgment constantly, in ways only people who have run one know to read. Volare captures that as structured observation rather than treating it as the end of the process.

Bylines

• Baseline-relative. Three bylines is routine on one desk and an outlier on another. The signal is the departure.
• Beat composition. An M&A reporter plus a media reporter means the newsroom decided the story crosses sectors.
• Escalation. One byline at 9am, two by noon. The trajectory, not the endpoint.
• Contribution tags. Reporting mass without a byline, and often the moment people were pulled off other work.
• Franchise names. An outlet’s scoop reporter attaching outweighs two junior bylines.

Placement

• Dwell. Hours held in the top slot, taken from repeated snapshots rather than one look.
• Concurrence. How many reference outlets held it above the fold at the same hour.
• Push alerts. The strongest escalation available, and precisely timestamped.
• Print A1 versus digital hero. Print is a slower, committee judgment. Divergence means something either way.
• Format escalation. Brief to story to analysis to package with sidebars.

Behavior

• Second-day follow-up. Nothing gets one by accident.
• Time to match and credit lines. Yields a scoop-attribution dataset nobody maintains for this industry.
• Time of day. Six a.m. is coordinated. Four-thirty Friday is a burial.
• Headline revision. Sharpened as confidence rises, softened under legal pressure.
• Beat staffing. Three outlets hiring the same new beat is a forward indicator of where news volume is heading.

What the ledger guarantees

Integrity

• Append-only. Corrections are new rows. “We saw it first” only survives if nothing was erased.
• Point-in-time. Every record answers what was known as of a timestamp. No look-ahead.
• The negative record. Dropped and quarantined items are kept with their reason codes, so the archive isn’t only what got published.

Time

• Four clocks: event, source publication, ingest, our publication.
• Latency is a dataset. The gaps between those clocks are themselves the product.
• Two speeds. The observation is written instantly; only publication waits for verification.

Labels

• Versioned scoring. Rubric and model stamped on every scored row, so a change in scoring is distinguishable from a change in the world.
• Human overrides as rows. Machine said brief, editor said lead, reason: cross-sector. Expert labels applied before outcomes were known.

The business model

Volare sits on top. The system and the ledger. Everything below is a different way of selling the same work.

Dispatch — the media network. Trade titles feeding a national front page.

• Subscribers — paid tier built on the compiled Dispatch packets
• Sponsors — newsletter, podcast and jobs inventory against a decision-maker readership
• Events — highest margin, and the reason the titles need authority

Editorial platform — the pipeline itself, licensed to publishers who want it and can’t build it. Licensing and implementation.

Volare data products — the exhaust, packaged for buyers who never read the story. Feeds, API, custom research.

Status

Coded and running. The classifier across six closed-vocabulary desks and a five-value importance ladder. Strict whitelist validation with quarantine on any vocabulary miss. Upstream-identifier enforcement and slug-level idempotency. Three outcomes. Provenance tagging. Live publication pages filtering on their own tags.

Operating rule, human-executed. Verify before publish. The wire framing check above the wire line. Broadcast-standard error tolerance at the wire tier itself. The byline heuristics. Placement reading. Promotion triggering deeper enrichment. Real, and not yet automated.

Next to build, in order. The four timestamp fields. A ledger behind the CMS that keeps drops and quarantines. Entity resolution to canonical keys. Human decision capture. Homepage snapshotting, which unlocks dwell, escalation and headline diffing in a single build. Then versioned scoring and point-in-time export.

Nothing here is dropped from the description for not being automated yet. It is labeled instead.

Limits, stated first

• No history. The ledger starts when it starts. Filings and dockets backfill deeply and cheaply; editorial signals cannot be reconstructed retroactively, because nobody archived placement at this granularity.
• One labeler. Importance judgments come from a single editor, which means one set of instincts and one set of blind spots. Reason codes mitigate it. A second labeler fixes it.
• Source terms. Filings and dockets are clean. Publisher text is not redistributable. Metadata about a publication’s own choices is a stronger position than its content, and still needs counsel before it becomes a commercial feed.
• Small N early. Signals will be thin per entity at first. The per-desk vector is built so thinness shows honestly rather than being smoothed away by averaging.

Methodology

100% compliant, white-hat first-party opt-in data collection via free B2B career utility gates.

Volare Media Intelligence — architecture. Working document. Not for distribution.