Campaign
A campaign states what is being acquired and what would count as an honest finish before any
source opens. scavenger.plan_campaign@1 pins desired outcome, existing inventory, hard criteria,
weighted preferences, budget, deadline, region, evidence tier, delivery or travel semantics,
disclosure policy, and one operating mode.
Subject profiles
| Profile | Distinct judgment | Revision-one consequence |
|---|---|---|
| Bazaar | identity, condition, attributable seller claims, landed cost, and bounded bargaining for Bazoš-style listings | terms, one approved COD commitment, parcel inspection, dispute, or refusal |
| Technology | exact CPU, board, BIOS, cooling, memory, GPU, storage, case, PSU, firmware, and connector compatibility | Bazaar consequence only after compatibility and evidence are freshly revalidated |
| Property | transaction type, area, rooms, locality, route and amenity observations, listing contradictions, and professional questions | shortlist and diligence packet only; no contact, viewing, offer, reservation, deposit, signature, or payment |
A subject profile is a versioned rule set inside Scavenger, not another Composition or a source adapter. A technology profile may refine a Bazaar subject; property rules never leak into hardware or parcel judgment. Selecting a profile grants no source, credential, private anchor, money, or effect authority.
Watch observes and reports. Concierge may prepare questions and offers for review. Bounded
Autopilot may act only inside a fully pinned envelope. Changing the subject, ceiling, permitted
disclosure, payment rail, evidence requirement, or deadline creates a new campaign revision rather
than silently widening the old one.
A PreferenceRevision@1 may refine only soft ordering inside that campaign. It pins the reviewed
candidate field, admitted judgments, scope, provenance, and predecessor revision. It cannot turn a
failed hard gate into an eligible candidate, widen budget or authority, change disclosure, or
rewrite the campaign's past rankings. A campaign may make a reviewed preference revision active
for the next observation only through its declared learning mode; otherwise the revision remains a
proposal for explicit acceptance.
Continue with Market once the campaign is closed enough to measure candidates.