What this panel is
Every finding is a gap in our data, not a judgment about a person.
The honest answer to most questions this platform is asked today is “we cannot tell, and here is exactly why”. Each finding below names the decision it blocks and the change that would unblock it. Almost all of them are process changes someone can make this week — not features still to be built.
Each finding is labelled this period or all data. Some questions — has anyone reviewed anything lately — only make sense over the selected range; others — who has no account linked — are facts about the whole dataset and do not move when you change the range.
all data
No work-definition authority
No task tracker is connected, so all 503 work units are derived from pull requests. A derived unit proves code exists; it cannot prove the code was what anyone asked for. Evidence confidence is therefore capped below the scoring floor, so no overall score is produced.
- Blocks
- Composite score · all alerts · estimate accuracy · commitment reliability · lead time
- Fix
- Adopt a task tracker and record planned work in it. This is the single highest-leverage change available.
evidence: Work units, and where their definition came from
all data
6
Unmapped identities
6 of 21 people who track time have no GitHub account linked. The platform shows "—" for their code activity, never 0 — asserting zero output for an unmapped identity would be the single most damaging thing this system could do.
- Blocks
- Any statement at all about these people's delivery
- Fix
- Collect the missing code-repository usernames and link them to the right people.
evidence: People who track time, matched against linked code accounts
all data
92d
No CloudTrail trail — history is expiring daily
CloudTrail keeps 90 days of management events without a trail writing to S3, and no trail is configured. The oldest event that still exists is 2026-06-26. Every day that passes deletes a day permanently — this is the only finding here that gets worse while nobody acts on it, and no later fix can recover what has already gone.
- Blocks
- Any infrastructure trend longer than 90 days · year-on-year comparison · incident history
- Fix
- Create a trail to an S3 bucket (CloudTrail → Trails → Create). A few dollars a month, and it only protects history recorded after it exists.
evidence: Oldest retained infra_events row: 2026-06-26
all data
No AI usage data
Claude Code analytics are not connected, so AI cost and tokens-per-point cannot be reported. A cost diagnostic only — by §8 it carries zero positive weight in any score, so nothing about a person's standing depends on it.
- Blocks
- Tokens per accepted point · the AI term in cost per point
- Fix
- Connect AI usage reporting with an administrator key.
evidence: AI usage records
all data
0/24
No contracted hours on record
Capacity utilisation needs contracted hours as its denominator. Using tracked time instead would divide tracked time by tracked time and always return 100% — a number that looks like a measurement and is not one.
- Blocks
- Capacity utilisation · overtime and CLT compliance checks
- Fix
- Add contracted weekly hours for each person from HR records.
evidence: Contracted hours held against each active person
all data
0.3%
Commits excluded from output
9 of 3062 ingested commits are excluded from team output — upstream history that predates a fork, and automated committers. Reporting the raw count would have overstated team output by roughly 3.5×. Every exclusion stores its reason on the row and is visible on the person's page; none is applied as a hidden filter.
- Blocks
- Nothing — this is a disclosure, so the headline count can be trusted
- Fix
- No action. Review the reasons if a number looks wrong.
evidence: Excluded commits and the reason recorded against each