SOURCE SNAPSHOT
Start with provenance and source scope.
INPUT AND EVIDENCE LEDGER
Classify every field before using it.
| Field | Value | Origin |
|---|---|---|
| Reported prices | ₩780M, ₩800M, ₩820M, ₩830M, ₩840M, ₩1,200M | Controlled test fixture |
| Cancelled record | ₩1,200M | Official 해제 status applied |
| Comparable filter | Same complex and area bucket | APT rule under test |
| Statistic | Median | Robust baseline choice |
REPRODUCIBLE WORK
Run the quality-control chain.
| Check | Formula or rule | Result |
|---|---|---|
| Unfiltered median | Average of third and fourth values | ₩825M |
| Apply status | Exclude the record marked 해제 | Five active records remain |
| Active median | Middle of five active values | ₩820M |
| Audit finding | ₩825M − ₩820M | ₩5M upward bias in this fixture |
DECISION READING
The audit finding.
The bias is small in this fixture but can be material in thin markets. The status filter must happen before summaries, highs, recovery ratios, or rankings are calculated.
HUMAN ACCOUNTABILITY + AI USE RECORD
The owner is accountable; the technical checks were AI-assisted.
The TOOOLBOX site owner authorized the publication scope and remains the correction contact. AI assistance was used for drafting, implementation, arithmetic recomputation, and source-link checks. This record does not claim line-by-line human fact-checking or licensed professional review.
- Record ID
- TB-APT-13-20260902
- Owner role
- TOOOLBOX site owner and developer
- Technical check date
- 2026-09-02
- Human fact-check
- No separate human fact-check recorded
- AI use
- Drafting, code, arithmetic, and link checks
- Correction channel
- Public contact page
- Confirmed the disclosure system's cancellation legend.
- Ran the fixture with and without the cancelled record.
- Checked that no fixture price is described as a live market transaction.
BOUNDARIES
What passed data checks still cannot prove.
- The six prices are a transparent test fixture, not claims about a named apartment.
- Cancellation status does not explain why the transaction was cancelled.
- Other revisions and matching errors require separate checks.
SENSITIVITY
Failure modes to retest.
- The effect grows when the sample is smaller or the cancelled price is more extreme.
- Mean prices are usually more sensitive to an outlier than medians.
- Late cancellation reports require the stored baseline to be recomputed.
MORE APT DECISION NOTES
Test the result yourself.
This note preserves a dated calculation. Change the inputs in the live workspace, or compare another evidence pattern below. Live source availability and results can differ from the dated example.