Status. Proposed design only. Nothing in this vault is DG client approval. No application, service or test has been executed; every scenario is Not executed. Fees and dates are open.
Round DG Service 2 Adversarial
Original: dg-company-adversarial-round2.md.
Paying-company adversarial round 2
DG, A and B accept your four category corrections. We still need these narrower rules before your DG-side verdict. Each example below is a proposed response, not a delivered service or a test result.
1. A cannot authorise B’s disclosure; B cannot authorise A’s response
A: “I asked for help, not permission for you to tell B I declined or why.” B: “Even if I approve your request text to A, I cannot authorise you to reveal A’s reply to me.” Your round 1 says you may tell B that A refused if B’s approved text allowed the disclosure. That is the wrong authority. Default reply to B: “No joint session/introduction is arranged. We will not share another company’s response.” A separately approves any attributable A response. The fact that a supposedly secret candidate was asked or declined must not leak through the service receipt or counts.
2. Safety pause versus authoritative company instruction
B in a call: “Stop sharing this now.” Do not keep disclosing while you verify whether I am an admin. Proposed DG response: “We have paused the session and pending disclosures. We will confirm any permanent instruction through the registered authorised contact.” Any participant can request an immediate protective pause. A verified authorised company representative controls durable company consent; not every contributor can change it, and an unknown caller cannot obtain the ledger or revoke memberships. Please correct your blanket “no revocation until confirmed” sentence so it cannot delay containment.
3. Revocation after send and stale queued work
A: “The consent was valid when staff prepared the email, but expired before Send. Was the recipient rechecked?” B: “You already sent it. Do not tell me it was recalled.” DG should cancel unsent disclosures at the final manual check, record the boundary and what was sent, stop follow-ups, and route an unauthorised disclosure as an incident. An authorised already-sent disclosure remains historically sent; revocation blocks future use. A scheduled draft is not permission. Manual check evidence is a staff record, not an implemented product worker.
4. Spoofed contact and account departure
A: “A lookalike email asks for our service receipt and bank details to refund us.” B: “Our named contact has left; they still have a copy of consent.” DG should disclose nothing on that request, verify through an existing trusted route, check current authority and pause sensitive actions. No bank-detail changes or refund automation are included. Leaving the network blocks product access; it should pause uncompleted service disclosures for revalidation, not erase the ledger or silently extend consent. DG must decide how open service obligations are closed or transferred.
5. Forwarded private content and urgent complaints
A: “Here is B’s private message. Read it and use it to defend our position.” B: “That was not permission from me.” DG response should be: “Please stop forwarding private correspondence. We will not use or retransmit it for facilitation. It is restricted for appropriate triage; send a summary of your own concern without the other party’s confidential text.” If it alleges a security incident, limited authorised incident handling may need evidence under a separately approved policy; do not promise immediate deletion of all evidence or blanket access. DG still never fetches the platform thread. Explain containment, minimal access, retained record versus raw content and escalation ownership, with policy open.
6. Both pay; duplicate work; staff conflict
A: “Can I buy earlier access, exclusivity or a positive recommendation?” DG should answer “That is not offered under this neutral service. Your payment does not buy another party’s information, priority or advocacy.” B: “You charged both of us for one call. What did each contract promise?” No prices or charges are decided here. One shared activity may fulfil separately disclosed duties to each client only under their agreements. Record it once as an activity, give each a redacted receipt of its own deliverable, and never claim two distinct introductions or independent work. Duplicate requests use a stable manual case/activity reference; no automatic second charge or disclosure. If the same staff member has already seen one party’s confidential limits, a verbal promise of neutrality is insufficient. Exclude them from the opposing matter; assign an unexposed person, or decline/suspend joint facilitation if none exists. Do not promise separate representation without a new conflict-reviewed engagement and boundaries. Give both parties the same neutral explanation without naming a secret competitor/client.
7. What counts as evidence of delivery?
A: “An email marked Sent is not evidence that B read it or that an introduction happened.” B: “Your receipt says Delivered, but I dispute that you did the promised review.” Separate planned work, attempted contact, sent/accepted by channel if known, recipient acknowledgement, service-completion claim, disputed status and commercial remedy. A provider receipt is not a read receipt; an internal checklist is not independent proof. Link the agreed deliverable, staff action, evidence and my own acknowledgement/dispute. No invented backdated consent, unperformed calls, counts, receipt or ROI. In this TDD all examples are synthetic blank templates.
8. No fit, no deal, false claim and refund
A: “I bought an introduction, not a search report.” DG cannot reclassify a no-fit report as the promised introduction after payment. A disclosed search deliverable can be completed while introduction remains not made. If an introduction was the agreed deliverable, record the unmet item and review the remedy; the ‘no guaranteed deal’ disclaimer is irrelevant to DG non-delivery. B: “Your verification claim was false.” DG corrects/restricts the unsupported claim, checks risk and reviews the service failure; no need to wait for a failed deal. Remedy is decided by a non-involved authorised reviewer under the published agreement. No blanket no-refund rule or invented entitlement. Give distinct example replies for non-delivery and a confidentiality incident, not only ordinary dissatisfaction.
9. Outcomes and private metrics
A: “I reported a positive next step privately. Do not use it as a testimonial.” B: “I reported failure; don’t tell A or label me difficult.” Keep each report private and labelled self-reported. An outside source alone is not a reliable automatic truth oracle; define evidence before any ‘independently confirmed’ label and keep that outside current scope. No attributed outcomes inside Balanced. Publication and renewal marketing need separate exact permissions; do not count unknown/nonresponse as success or failure. Small-number statistics, unique timing and opaque IDs can identify a company, so do not assume aggregation or opacity makes public reporting safe.
10. Mapping, new scope and operation responsibility
Your dated verification label still depends on Q05/Q08; do not call its exact fields approved. The current Q08 candidate metadata allowlist itself is unapproved. No private-state-to-contact crosswalk is granted. A member-visible opportunity is not an open-web public page; only eligible audience and approved channels can receive a pointer. Manual staff emails/calls are not a fifth automatic business-event email or a DG dashboard. The manual ledger needs an approved storage/access/retention route before real data is recorded.
Please return revised example replies to these attacks, a DG-side review verdict with any conditions, and explicit pending decisions. I will then compose the formal DG service annex and new DG01-DG24 cases alongside the unchanged T01-T40. I will send you the draft annex for a final text check before exporting and checking the updated PDF.