Executive Summary · Jul 20–26, 2026 → Jul 20–26, 2026
Data quality flag — this report contains a known defect
A median_salary Trend value was displayed where job_postings_count was intended (e.g. 'snowflake mentions moved +15000 this week, to 172500' — 172500 is a median salary, not a mention count). Root cause: Trend queries in ReportingAgent filtered by period but not by metric. Fixed 2026-08-15 (see docs/MAINTENANCE.md, tests/test_metric_isolation.py). The underlying job_postings_count data for W30 was never wrong -- only this report's narrative text. Left unedited rather than silently corrected; flagged here instead.
Week of Jul 20–26, 2026: software engineer mentions rose from 0 to 29. Weekly counts here swing widely, so read this as a snapshot, not a trend.
Rising: software engineer, security, data analysis
Falling: aviation safety inspection
Decision at hand: where to place the next increment of hiring and training budget.
Track record: 72 forecasts scored, mean accuracy 0.74, calibration delta -0.12 — the system has been overconfident and has lowered its own confidence accordingly.
Three lines of evidence
| Signal | Type | Job postings this week | Change | Direction |
|---|---|---|---|---|
| software engineer | role | 29 | +29 (+100%) | up |
| security | skill | 21 | +21 (+100%) | up |
| data analysis | skill | 13 | +13 (+100%) | up |
| automation | skill | 8 | +8 (+100%) | up |
| aviation safety inspection | skill | 0 | −8 (-100%) | down |
Recommended action
- executive: Fund capability in risk management (skill). Demand is rising 0% across the last 6 observed periods, and the 4w forecast holds that direction. (confidence 0.44)
- executive: Fund capability in stakeholder management (skill). Demand is rising 5% across the last 6 observed periods, and the 4w forecast holds that direction. (confidence 0.39)
Caveat we will not bury
This reads posting volume from a narrow source set. It is a leading indicator with a lag, not
a hiring plan. Confidence figures are the system's own, and they are adjusted downward
automatically when past forecasts miss.
How we know
Evidence
Confidence
Source diversity
2 sources
93% from one source — treat as narrow
Last updated
2026-07-26
Reality validation
Not yet scored against reality.
How this forecast was made
- Approach
- Linear trend fitted to recent weekly counts, with a range to account for uncertainty
- Can it be edited?
- No — forecasts are locked at publication and scored as-is
Primary sources