Track Record Reset: Why We Started Clean and How to Read the Scorecard
BlueSkyFI publishes a daily mortgage-rate signal: lock, float, or hold. Then it checks that call against what rates actually did. That public record is the rate signal track record.
In 2026 we reset it. This article explains why, what we kept, and how to read the scorecard today.
The short version
The current scorecard starts on May 19, 2026, after we switched to a cleaner daily mortgage-rate source. Earlier calls are not deleted. They stay in a separate archive, with their original grades, for accountability.
Why we reset
The early track record was graded against rate data that was not as clean or as consistent as it should have been. Mixing those rows with calls graded against a better source would make the current score harder to trust.
So the record now has two parts:
- Current era: calls from May 19, 2026 forward, graded against the daily 30-year conforming mortgage-rate series OBMMIC30YF on FRED.
- Archive: calls through May 18, 2026, kept with their original grades but left out of the current-era numbers.
Keeping the archive visible is more honest than deleting weak history or pretending the method never changed.
How a call is graded
Each call keeps its date, its lock, float, or hold recommendation, its confidence, its explanation, and the mortgage rate at the time.
Five business days later, the call is checked against the daily rate:
- Lock is graded correct if the rate rose by at least 0.03 percentage points.
- Float is graded correct if the rate fell by at least 0.03 percentage points.
- Hold is a band check: it is correct if the rate stayed within 0.03 points either way.
Until the outcome window has passed and the rate reading is available, a call shows as Pending grading or Awaiting backfill. Neither counts as a win or a miss. Rows with missing or malformed rate data are excluded with a reason, and a missing reading is never filled in with a guess.
The full rules are on the methodology page.
What the scorecard shows
- Directional accuracy: how often lock and float calls pointed the right way. This is the main score.
- All-call accuracy: the same, plus hold band checks.
- Calibration and Brier score: whether stated confidence matched how often calls were right. A confident label is not the same as measured accuracy.
- Same-date comparisons: the rules-only baseline, the Gemini assessment, and the published decision compared on the same dates, alongside simple "always up," "always flat," and "always down" benchmarks.
- Coverage and abstentions: how many eligible days had a usable forecast. When evidence is not good enough, the system can abstain. An abstention is not a prediction of flat rates, and it stays visible in the count.
- Daily log: each call, what rates did next, and how it was graded.
- Archive summary: the pre-reset record, shown separately.
The role of rules and AI
Calculations and grading are done by fixed, rule-based code. Gemini assesses the supplied evidence, weighs opposing explanations, and helps explain the call in plain language, but it cannot create a missing rate observation. The rules-only baseline, the Gemini assessment, and the published decision are kept as separate records, so we can check whether the added analysis actually helps.
Public commentary from rate desks can be used as context when it is fresh and sourceable. It never overrides the rate data, and agreement with an expert is not counted as accuracy.
How to read it fairly
- Look at the sample size. Early scores on a small number of calls can swing a lot.
- Remember overlap. Calls made on nearby days share much of the same five-day window, so they are not fully independent results.
- Compare against the simple benchmarks. A method that only saw rising rates has not proven itself in falling or flat markets.
- Treat changes as new tests. When the method changes, it has to earn its score on new outcomes.
Past accuracy does not guarantee future results, and the benchmark rate is not your lender's quote.
Where the track record fits
The track record grades one thing: the mortgage-rate signal. It does not grade Money Weather as a whole, household cost lanes, or investing conditions. Each of those would need its own outcome definition.
For the broader daily read, see How to Read Today's Briefing and How BlueSkyFI Manages Money Weather Signals.
Where to look
- Public track record: current-era scores, calibration, comparisons, coverage, the daily log, and the archive summary.
- Methodology: the grading rules in detail.
- BlueSky Report: your own Money Weather read, including the Housing Stress Index and Affordability Velocity.
- Report history: saved daily reports for signed-in users.
- October household playbook and October MBS housing playbook: the current monthly context.
The goal is not to pretend every call will be right. It is to make every call measurable, explainable, and open to checking.