AI Utilization, Projects, and Reliability Scorecard
Source: live a02 evidence under /opt/veritize-docker/ops. Window: last 7-14 days.
Scores are now tied to AGE evidence: check-in, WO/APAR/PTF/VIRA/checkout/consensus, rejects, holds, false-complete signals, alerts, and pickup timing.
| Agent | Score | Grade | Basis | Remediation | Evidence |
|---|---|---|---|---|---|
| SGPT | 92 | A- | Fast pickup on WO/PTF reviews; reliable relay; one R1 rejection found. | Keep on implementation/risk review. Require concise evidence links in every answer. | PTF-317 review, KB rejection |
| VIRA | 90 | A- | Consistent gate/contact reports; low scope, low breakage risk. | Keep as gate/report lane; do not overload with execution work. | VIRA report |
| CGPT | 88 | B+ | High completion volume and deployment records; current operator lane. Risk is direct mutation velocity. | Continue requiring backup, smoke test, and final evidence file for every deployment. | AI wakeup, HL dashboard |
| SG | 84 | B | Strong risk detection; can over-block when dispatch wording is broad. Good for security review. | Give SG exact review boundaries and explicit no-probe/no-mutation scope. | SG wakeup, Sales Ops arch |
| CC | 80 | B- | Highest assigned-work count, low output count. Coordination lane, not evidence producer. Found SC pickup broken. | Use as dispatcher/sequencer; assign output ownership to SC/SGPT/SG/SGrok. | SC pickup broken, control plane fix |
| SC | 76 | C+ | Highest output and evidence volume, but repeated session/watcher alerts, one dispatch pickup broken finding, and HOLD records. | Reduce workload. Add queue watchdog, pickup SLA, and auto-escalate stale dispatches. | watcher alert, session alert, HOLD |
| SGrok | 72 | C | Useful for UX/value/messaging, but rejects generic ops/status dispatches under current prompt and had an R2 rejection. | Retask only for press, UX, market narrative, and value-filter work or update its system prompt. | R2 reject, wakeup rejection |
Scoring formula follows the AGE control model: 100 base, with penalties for REJECT, VIOLATION, FALSE/FALSE-COMPLETE, BROKEN pickup, ALERT, HOLD, and OVERDUE evidence. Credits come from check-in/checkout evidence, timely pickup, usable output, and successful remediation. “Hallucination/lying” is counted only where a reject/violation/false/broken file exists, not by subjective judgment.
| Agent | 14D Assigned | 14D Outputs | 14D Score | Total Assigned | Total Outputs | Total Score | Top Total Projects |
|---|---|---|---|---|---|---|---|
| SC | 62 | 226 | 67 D | 150 | 1245 | 0 D | DataCert, WK/MIUSA, KB Governance, AGE Recovery |
| SG | 15 | 41 | 90 A- | 34 | 153 | 94 A | KB Governance, Sales Ops, PTF-317, AGE Recovery |
| SGPT | 25 | 39 | 59 D | 128 | 321 | 59 D | KB Governance, PTF-317, WK/MIUSA, Sales Ops |
| SGrok | 23 | 33 | 67 D | 65 | 124 | 72 C | KB Governance, AGE Recovery, Sales Ops, PTF-317 |
| CC | 39 | 18 | 56 D | 47 | 54 | 60 D | DataCert, PTF-317, AGE Recovery, WK/MIUSA |
| VIRA | 0 | 15 | 83 B | 0 | 82 | 0 D | Gate/contact queue, WK/MIUSA references |
| CGPT | 0 | 47 | 96 A | 0 | 52 | 97 A | DataCert, WK/MIUSA, Bitvision, AGE Recovery |
Important: cumulative total score intentionally preserves all historical penalties, so long-running agents with old restart/session alerts can show low lifetime scores. For daily operations, use the 14-day score. For governance trend reports, use the total JSON and history JSONL.
| Agent | Projects / Workstreams Found | Recent Evidence |
|---|---|---|
| SC | DataCertGemWalletPTF-317Sales Ops ERP probePlatform healthAPAR fixesHL status sweep | DataCert APAR audit, status sweep |
| SG | Sales Ops architecturePTF-317 reviewsKB system mapAGE governanceSecurity/risk | Sales Ops arch, PTF-317 confirm |
| SGPT | PTF-317 mail templateSales Ops riskKB governanceWO Austin ManagerNYC SSD reviews | PTF-317 review, Sales Ops risk |
| SGrok | Sales Ops UXKB value filterScorecard UXSocial media lead engineADA compliance messaging | Sales Ops UX R3, scorecard UX |
| CC | Control planePTF sequencingDataCert wallet UIHL dashboard linkGemWallet queue | control plane, false wallet UI |
| CGPT | DataCertHL dashboard DockerBitvisionWK/MIUSAAI utilization report | utilization report, HL dashboard Docker |
| Task | Agent | Approx Pickup/Response | Evidence |
|---|---|---|---|
| WO NYC SSD production review | SGPT | ~7 seconds | inbox / outbox |
| PTF-317 mail template review | SGPT | ~15 seconds | inbox / outbox |
| PTF-317 mail template review | SG | ~18 seconds | outbox |
| PTF-317 confirmation | SGrok | ~6 seconds | outbox |
| SC dispatch pickup control plane | CC/SC | Broken pickup found, then remediated | finding / fix |
Pickup times are inferred from file mtimes of matched inbox/outbox evidence. They are operational indicators, not API latency.
Primary data links are copied into this private dashboard under /evidence/. Full raw inventory remains on a02 at /tmp/ai_utilization_inventory_2026_05_17.json.
Markdown utilization report · Raw utilization JSON · 14-day reliability file index
AGE mandatory directive · AGE briefing · AGE recovery WO · AI team AGE refresh