📖About

DISTINCT FROM CAROL-INI-3465, which covered SCHEDULED processes being charged under the wrong worker's name and is already delivered. This is about ACTIONS TAKEN FROM AN AGENT'S CHAT, a surface that did not exist when 3465 was written.

Ninad's rule for the new action surface is that chat pays only for the asking and the service that performs the action pays for the doing. That is not what happens. Every one of the 32 catalogued actions bills the Chat track: Elrond stopping the pipeline, Radagast restarting a service, Argus running a test suite and Midas lifting a budget all charge Chat. The executor does open its own work ticket, but the ticket has no field that can name a paying track, and the track resolver sends any caller whose id ends in -chat to the Chat track ahead of everything else, so no ticket change alone can reach it. The catalogue rows name a service but nothing reads them, and one of those names is a placeholder that resolves to nothing at all.

Give the work ticket an explicit paying-track field, honour it ahead of every other rule in BOTH the spend gate and the spend report so the limit and the report can never disagree, and have each action name the track of the service that performs it, resolving the target's own service for actions carried out on somebody else's droid or app.

Also fix a live mis-attribution found in the same area: the chat grounding query builder hardcodes the Policy Gleaner's identity as the caller for every agent's chat, so other agents' chat spend is stamped with his name, and before the chat work ticket went live this charged the Constitution service's Policy track for other people's conversations. Ninad's ruling of 2026-07-31 is that the Policy track stays on Constitution; only the borrowed identity is wrong.

⚖️Decisions

• Auto-detected remediation target INI-999902337 from title/description scan (matched CAROL-INI-3465 -> row id 999902337 (CAROL-INI-3465-00: Money is charged to the process that actually did the work)); override by setting remediates_initiative_id explicitly at bypass_start. (system-auto-detect)
• Elrond's bypass methodology checklist (a reminder, not a gate -- you've got this): 0. File it requested_mode='bypass' (planner-vs-bypass is a deliberate choice). bypass_start REFUSES a non-bypass initiative (CAROL-INI-1846), and the dispatcher only skips the bypass lane when the mode says bypass -- a 'planner' mistag lets Merlin's pipeline grab the placeholder step and block your finished work. 1. Filed as planned status -- let the bypass claim/activate it; never file active. 2. Open the bypass (bypass_start) with your droid id + the remediation answer (remediates_initiative_id=NNN, or remediates_nothing=True). 3. Work the blocks for your work-type: template -> design -> code -> test -> review. Do the real work; record decisions on the initiative as you make them. 4. Reality is recorded for you at close -- code (files changed), each decision, and the twin-review verdict become real activities tied to this initiative and show in the Activity Tracker like a planner run (CAROL-INI-1840). No dummy rows. 5. Keep the initiative status moving; it parks in 'reviewing' and is tagged uat-pending for you at close (CAROL-INI-1836), so the stuck-watchdog leaves it alone until UAT. 6. Close runs the gates (design/architecture compliance + caller-audit). If a gate flags something pre-existing or unrelated to your change, waive it with a clear written rationale -- audit, don't skip. 7. Bypass skips the planner's auto-orchestration, NOT the standards. Same template checklist, same review, same observability as a planner run. (elrond)
• [status-router] planned -> executing | event=bypass_executing | bypass transition (or-bx-01)
• The Policy track stays on the Constitution service; only the borrowed identity was wrong. — Ninad's ruling, CLI-187. The premise that policy work needed a Governance track did not hold: a dedicated Policy track already existed under Constitution, and the Policy Gleaner's own spend was already booked to it correctly. The actual defect was that the chat grounding query builder hardcoded the Gleaner's worker id as the caller for EVERY agent's chat, so other agents' chat spend was recorded under his name and, before the chat work ticket went live, charged his track for other people's conversations. Constitution keeps the law and the records including policy formation; moving it would have reversed the earlier deliberate ruling that gave that service to Orion. (Ninad)
• The work names its payer; the caller's identity cannot. — Every rule that decided who pays read the identity the call ran under. An agent-chat action runs under the chat droid's id, so every rule said 'chat' and Chat paid for work it did not do — and no change to the declared kind could reach it, because the -chat identity rule outranks the kind. The work ticket therefore carries an explicit paying track, honoured ahead of every other rule in BOTH the spend gate and the spend report so the limit and the report can never disagree about the same money. (orion)
• An ambiguous payer is named as no payer, never as a guess. — Two shapes of ambiguity exist and both are refused rather than resolved: an agent whose service runs several lanes (Elrond's initiatives service has five and no core lane) and an agent that owns several services (Hagrid owns two). In both cases the resolver returns nothing and the ordinary rules decide, which is exactly the behaviour that existed before. Putting money on a lane nobody chose is worse than leaving it where it already was. Both cases are asserted in the checks so the line is deliberate rather than accidental. (orion)
• An unknown track name is discarded, so a typo can never move money. — The named payer outranks every other rule, which makes it the one field where a typo would silently move spend to a track that governs nothing. It is validated against the real track list before it is honoured, in the gate and in the report alike; an unrecognised name is dropped and the ordinary rules decide. Fail open, never invent a payer. (orion)
• [status-router] executing -> reviewing | event=dispatcher_transition | dispatcher state change (ds-s1)
• DEFECT FOUND BY CLI-188: the grounded-query caller was composed, not resolved, and every agent was refused — Raised for this initiative UAT, fixed forward rather than reverted because the fix serves its own intent. The helper this initiative added builds the billing id by appending -chat to the AGENT id, giving agt_044-chat. Chat droids are named after the agent (scriber-chat) and never after its id: zero of the 42 chat droids match the composed form, so the money gate refused every grounded query as an unregistered caller. Grounding returned empty for all 39 agents and each fell back to inventing answers - the symptom Ninad reported was Scriber unable to say how many videos he had published. The caller is now RESOLVED from the registry (owner + id LIKE %-chat), falling back to the agent id, so the spend still carries the name of the conversation that caused it as this initiative intended. Fixed under CAROL-INI-3526; regression test_ini3526 asserts every agent bills a registered caller and is proven red against the pre-fix code (39 of 39 unregistered). (orion)
• [status-router] reviewing -> closed | event=operator_signoff | Auto-accepted (CAROL-INI-1859): Orion-initiated, >2 days in reviewing with no objection. (el-srac-01)

✅Success criteria

• When an agent carries out an action from its chat, the work is charged to the service that performs it, and chat is charged only for the asking. (must_have)
• An action carried out on another service's droid or app is charged to that service, not to the one that triggered it. (must_have)
• The spending report and the spending limit name the same payer for the same action. (must_have)
• An action that would take a service past its daily limit is refused before any money is spent, and the refusal names that service's own limit. (must_have)
• No agent's chat spending is ever recorded under another agent's name. (must_have)