Google’s Open Knowledge Format Adds Five Trust Signals

Jul 29, 2026 06:00 PM - 1 month ago 833

Google has announced an update to the Open Knowledge Format (OKF), type 0.2. The caller type adds 5 spot related features that a “consumer” of the OKF tin usage to verify 5 aspects astir the OKF bundles.

The 5 spot signals and their related OKF fields and conception type are:

  1. Provenance (field: sources)
  2. Trust (fields: generated, verified)
  3. Freshness (field: stale_after)
  4. Lifecyle: (field: status)
  5. Attestation (new conception type: Attested Computation)

The announcement positions these 5 signals arsenic answering 5 questions astir the OKF bundle successful bid to found trust.

The 5 questions:

  1. “What was this created from? (provenance)
  2. How overmuch should I spot it? (trust)
  3. Is it still true? (freshness)
  4. Is it the existent version? (lifecycle)
  5. Was this number produced the measurement we said it must be? (attestation)”

Provenance

Google introduces a caller “sources” section that records wherever the accusation successful a conception came from. This enables consumers to place the original sources utilized to create it. This provides root accusation that consumers tin usage to measure a concept’s trustworthiness, answering t he question, “What was this created from?”.

This is Google’s illustration of the sources field:

sources: - id: warehouse-schema resource: https://wiki.acme.internal/data/warehouse/schemas/sales title: Acme Retail storage schema — income dataset author: team:data-platform usage_count: 1240 last_modified: 2026-06-15 - id: revenue-policy resource: policies/revenue-recognition.md title: Revenue Recognition Policy (FY2026) author: human:jsmith@acme last_modified: 2026-06-15

Google explains what it each means:

“The caller sources section records the materials a conception derives from: an outer doc, a bundle-relative path, aliases moreover a scope descriptor for illustration “all queries successful task X.” At the aforesaid time, an introduction tin transportation nonsubjective credibility signals: author, usage_count, last_modified.

The deliberate prime present is what we didn’t add. OKF records the signals, not a credibility score. A people is subjective, doesn’t larboard crossed consumers, and goes old the infinitesimal it’s written.

Instead, credibility is inferred from the signals by whoever is consuming (and tin beryllium dynamically scored by the consumer, if desired), the aforesaid measurement you’d spot a heavy used, precocious updated, authoritatively authored root much than an anonymous one. And erstwhile the assemblage cites a circumstantial source, it does truthful pinch an mean markdown footnote keyed to the root id ([^export-schema]), truthful attribution is per-claim alternatively of a dangling database astatine the bottom.”

Trust: Generated And Verified Fields

The adjacent portion of the spot signals are the Generated and Verified fields.

The “generated” section records who created a concept, while the “verified” section records who independently confirmed it. A user (such arsenic an AI agent, an LLM, aliases an application) tin usage the verification accusation to select concepts based connected whether they are unverified, machine-confirmed, aliases human-reviewed.

This is the illustration of the Generated and Verified fields successful use:

type: Metric title: Revenue generated: { by: reference_agent/gemini-2.5-pro, at: 2026-06-30T14:00:00Z } verified: - { by: human:jsmith@acme, at: 2026-07-01T09:00:00Z }

Freshness and Lifecycle: position and stale_after Fields

The position section communicates what portion of the lifecycle a conception is successful truthful that a user tin place whether it’s successful draft, current, aliases it’s outdated (deprecated). It signals whether a conception is simply a draft, stable, aliases is deprecated.

The stale_after section specifies the day aft which the conception should beryllium re-verified earlier being used. Consumers tin usage these fields to place concepts that request to beryllium re-verified aliases to exclude outdated concepts from caller activity while preserving them for humanities reference.

Here’s the illustration of this successful use:

type: Metric title: Gross Margin (legacy, pre-FY2026) status: deprecated

Attested Computation: For Verifying Calculations

Attested Computation is not a field, it’s a caller Type. Attested Computation defines the approved measurement to cipher a worth and provides a measurement to verify that the calculation was performed the measurement it was expected to beryllium calculated.

The charismatic explanation explains:

“Provenance answers wherever a declare came from. Attestation answers a harder mobility that matters the infinitesimal an supplier reports a dollar figure: was this number produced the measurement we said it must be, aliases did the supplier improvise its ain SQL?

OKF v0.2 introduces a caller conception type, Attested Computation. It carries not conscionable what a worth intends but a sanctioned measurement to compute it, and the intends to cheque that the sanctioned point really ran.”

Here’s the example:

--- type: Attested Computation title: Revenue for a fiscal year runtime: bigquery parameters: - { name: year, type: integer, required: existent } executor: resource: skills/run-on-bq.md receipt: [job_id, executed_sql, result] attester: resource: attesters/sql_equality.py generated: { by: reference_agent/gemini-2.5-pro, at: 2026-06-30T14:00:00Z } verified: - { by: human:jsmith@acme, at: 2026-07-01T09:00:00Z } status: stable stale_after: 2026-12-31 sources: - id: revenue-policy resource: policies/revenue-recognition.md title: Revenue Recognition Policy (FY2026) author: human:jsmith@acme last_modified: 2026-06-15 --- # Computation SELECT SUM( CASE WHEN o.currency = 'USD' THEN o.net_amount ELSE o.net_amount * fx.rate_to_usd END ) AS revenue_usd FROM `acme.sales.orders` AS o LEFT JOIN `acme.finance.fx_daily_rates` AS fx ON fx.currency = o.currency AND fx.rate_date = DATE(o.order_ts) WHERE o.order_status = 'delivered' AND DATE_DIFF(CURRENT_DATE(), DATE(o.order_ts), DAY) >= 30 AND EXTRACT(YEAR FROM o.order_ts) = @year"

Updated Open Knowledge Format Documentation

Google has updated the GitHub Repository to bespeak this update and has published an announcement that serves arsenic an explainer.

Featured Image by Shutterstock/Jack_the_sparow

Category News SEO Generative AI
Follow Us On Google
More