Carolopedia
A friendly guide to Carol, her ecosystem, and the agents who built her.
📖About & Usage
About
Tracker turns completed planner executions into clear task records that Elrond can review. It counts each agent’s passed, failed, and pending checks, then writes the results to Carol's Org.
On demand, it extracts execution data, resolves agent IDs, creates per-agent records, and rolls up multi-level results. It avoids duplicate tasks, preserves the planner’s assignments and statuses, and skips unresolved agents. Tracker’s exact name is unique, but it belongs to the reusable intelligence family, which has 25 droids run across several agents. It is not pure software: it calls Claude using the opus model to turn execution results into agent task summaries.
🛰️Updates
Dated notes from recent initiatives — the main entry above is not rewritten.
Token Cost Tracker's by-service chart now has darker borderless bars, a card strip showing each service's daily spend, and the ability to highlight a service's bar segments by selecting a card or bar.
The Activity Tracker now derives planner activity from the execution book when no activity stamp exists, so planner-mode initiatives like autonomous pipeline deliveries appear correctly in done-per-day charts. This also honors the RSI exemption.
2025-04-15: Tracker now uses a shared day-count module to ensure its initiative counts match the Activity Tracker.
The Tracker Tracker now supports family-based counting and includes a new 'initiative_nnnn' column, stamped automatically by a database trigger and backfilled for all historical initiatives.
The Tracker tracker's displayed metrics are now defined in the data dictionary, and a dedicated Data Dictionary app under Governance (owner Clara) is being built. Any displayed metric without a dictionary row is now considered a defect.
The tracker now counts only COMPLETED initiatives, not worked attempts, per Ninad's ruling. This revises the same-number rule for Tracker.
The Token Cost Tracker chart now reads per-service budgets from the new `budget_allocations` table, so budgets like Scribe's Creative track appear automatically instead of being sliced from the fleet cap.
2025-03-06: The Tracker now reads the effective lane budget from the registry-first source instead of the stale daily_budget.json mirror; also adds support for day-scoped exception bumps. Tracker
Gemini is now a first-class provider in Tracker with its own label, and a corresponding prepay card has been added to the Model Subscriptions app.
The Activities card has been removed from the Activity Tracker Tracker. Now, when an initiative is selected in the Initiatives card Carol Initiatives, its activities render directly below the initiative list using the existing activity-listing table layout.
Activity Tracker now normalizes agent keys and correctly logs Carol costs, eliminating rows showing 0 tokens. This resolves a join mismatch where token-bearing droid runs used lowercase names or blanks (e.g., 'orion', 'elrond') instead of agent IDs.
The Azure consumption dashboard (Tracker) was found to be broken since its original port from the BB-blueprint, with a 404 on GET /dev/costs/ and missing routes plus an IMDS token source; these defects have now been wired to restore functionality.
The Tracker's Activity Tracker under-reports Orion bypass costs because expensive laptop-side Claude Fable reasoning is not captured. The Ninad ruling now requires all costs, including those on the laptop, to be recorded in the initiative cost ledger using the shared rate card Cost Center.
👤Owner
Elrond · Head of Engineering📚Recent initiatives
Initiatives that touched this droid — a short summary each; open one for the full story.