US Brand Evidence Snapshot Diff

Compare two small US-brand evidence snapshots and get one machine-readable diff. This is an inline-only Actor for agents and analysts who already have two snapshots and need a stable answer about what was added, removed, changed, or unchanged.

What you provide

Set primarySide to before or after, then put 1-50 rows for that side in the required primary array. Put 1-50 rows for the opposite side in the optional counterpart array, or omit counterpart to represent an empty opposite snapshot. Each row has an explicit bounded ASCII dedupeKey, a brand name, country: "US", and optional closed evidence fields. The same key may occur at most once per side. Raw text limits are checked before NFC normalization. Optional fields are omitted rather than set to null or an empty string. The Actor does not download URLs and does not read a Dataset.

What you receive

You receive exactly one deterministic report with sorted result arrays and counts. Added and removed entries preserve the normalized row. Changed entries include both canonical rows, their digests, and field-level differences. Unchanged entries include the canonical row and matching digests. Snapshot rows and explicit set fields are sorted; order inside arbitrary evidence arrays is preserved and can produce a change. Counts and Actor-owned digests are recomputed before delivery.

Example task

Use this inline input to compare the two snapshots and return the four categories plus the field-level price evidence change:

{
  "schemaVersion": "1.0",
  "primarySide": "before",
  "primary": [
    {"dedupeKey": "acme|offer|001", "brandName": "Acme Northstar Labs", "country": "US", "eventType": "offer", "evidence": {"facts": {"priceUsd": "24.00"}}},
    {"dedupeKey": "acme|award|001", "brandName": "Acme Northstar Labs", "country": "US", "eventType": "award"},
    {"dedupeKey": "unchanged|news|001", "brandName": "Example Federal Tools", "country": "US", "eventType": "news"}
  ],
  "counterpart": [
    {"dedupeKey": "acme|offer|001", "brandName": "Acme Northstar Labs", "country": "US", "eventType": "offer", "evidence": {"facts": {"priceUsd": "19.00"}}},
    {"dedupeKey": "unchanged|news|001", "brandName": "Example Federal Tools", "country": "US", "eventType": "news"},
    {"dedupeKey": "new|launch|001", "brandName": "New Example Brand", "country": "US", "eventType": "news"}
  ]
}

The expected counts are before: 3, after: 3, added: 1, removed: 1, changed: 1, and unchanged: 1. The changed key is acme|offer|001, with evidence.facts.priceUsd changing from $24.00 to $19.00.

Machine-to-machine use

The executable input and output contracts are closed, bounded, permutation-invariant, and stable for agent pipelines. Nested finite numbers are accepted, while integer-valued numbers outside JavaScript's safe range are rejected before digesting. Set-valued fields are NFC-normalized, deduplicated, and sorted. The Actor has no network modules, no BYOK, no proxy, no Dataset input, no wall-clock identity, and no replay or exactly-once claim. Every run is a separate purchase.

Pricing and limits

PAY_PER_EVENT uses exactly two events: Actor start and one result-found report. FREE: $0.005 / $0.010; BRONZE: $0.00475 / $0.0095; SILVER: $0.0045 / $0.009; GOLD: $0.00425 / $0.0085; PLATINUM: $0.0041 / $0.0082; DIAMOND: $0.004 / $0.008. The discount tiers lower both event prices. There is no default Dataset-item event. Runtime: 512 MB, 90 seconds; input: 0-50 rows per side, 100 total, 256 KiB maximum.

Related Actors

These neighboring Actors were confirmed public and active by the fleet's durable anonymous Store read-back on 2026-08-05:

Actor	Use it with this diff
Company Profile Lookup	Build company context before comparing evidence snapshots.
Shopify Store Intelligence	Produce commerce evidence that can be compared between runs.
Funding Round Tracker	Add funding signals to a wider brand evidence workflow.
Patent Filing Monitor	Add patent events before snapshot comparison.
Intent Signal Aggregator	Feed changed evidence into a broader intent view.

FAQ

Does it verify evidence? No. It compares buyer-supplied rows.

Can identity be inferred? No. Matching uses only the caller's explicit dedupeKey.

What happens on a billing or output failure? The run fails closed and does not push an error as a paid result. Its error receipt preserves whether delivery was not attempted, unconfirmed, confirmed paid, or unknown, plus observed receipt and counter facts. If the post-charge OUTPUT write fails, the returned terminal receipt says output.state: failed instead of returning the successful diff.

The published Store contract is the inline-only input/output contract described above: schemaVersion and primarySide are required, primary has 1-50 rows, counterpart is optional with 1-50 rows, and each row requires a unique ASCII dedupeKey, brandName, and country: "US". Results are deterministic diff reports with sorted added, removed, changed, and unchanged arrays. The Actor uses exactly two paid events, Actor start and one result-found report, with no default Dataset-item event.