Frozen, anchored, and scored — nothing else
방향 없음·권유 아님 — 모델 예측과 채점 기록만 싣는다No direction, no recommendation — only model predictions and their scoring record.
Freeze
After each close, the 80% and 95% intervals for the next session's index change are computed and frozen. The freeze time and the prediction hash (prediction_sha256) stay in the file.
Anchor
The manifest hash-chain row that first carries the sha256 of the frozen bytes becomes the anchor; the web-repository commit time is the second anchor. An anchor after the target session opened is flagged.
Score
Once the target session closes, the realised change is scored: inside/outside the band, the percentile on the predicted distribution (PIT) and the Winkler score (width plus breach penalty). Prediction fields never change; only outcome is added.
What is absent
No direction — bands are symmetric intervals only. No levels — no prices or index points. No symbols. No returns or hit rates. Not a recommendation.
The prediction hash is the sha256 of a key-sorted JSON (separators , and :, UTF-8) of six keys: market, session_date, target, estimator, prediction, spec_sha256. anchor, outcome, frozen_at and notice are not hashed. Hover a number to see its facts source and as_of. Every number sits next to its freeze time and hash.
Index volatility bands — publication withheld (data terms of use under review)
Internal freezing and scoring continue; the record is published as is once the terms of use are confirmed.
Index volatility bands — publication withheld (data terms of use under review)
Internal freezing and scoring continue; the record is published as is once the terms of use are confirmed.
Index volatility bands — publication withheld (data terms of use under review)
Internal freezing and scoring continue; the record is published as is once the terms of use are confirmed.
Interval coverage by model × horizon — against the nominal 80
Intraday volatility models stamp 80% intervals at 15, 30 and 60 minute horizons on the same symbols; the share of realised changes inside the interval (cover) and the Winkler score (width plus breach penalty, in log-change units; lower is better) are counted. Model names are public textbook methods or internal selector names; gbm is a fixed normal baseline. Only symbol counts are published.
Real-time volatility race — publication withheld (data terms of use under review)
Internal freezing and scoring continue; the record is published as is once the terms of use are confirmed.
Real-time volatility race — publication withheld (data terms of use under review)
Internal freezing and scoring continue; the record is published as is once the terms of use are confirmed.
Only model-performance statistics are shown; no quotation values are published.
Base rates, transitions, durations — realised regime classification (not a forecast)
Regimes are realised categories from a historical all-symbol-day aggregate, written as neutral codes (R1–R6) only. They are not a forward direction and are shown in neutral colour only. In the transition test, p<0.05 means 'there is structure', not 'it can be predicted'. The regime ML score is a shadow (ablated) result and records only whether deployment is possible.
Regimes — publication withheld (data terms of use under review)
Internal freezing and scoring continue; the record is published as is once the terms of use are confirmed.
Pre-registered, scored, rejected, confirmed — counts and labels only
Counts from the experiment ledger (rows, duplicates included, timestamped rows only), model-registry status, the ML conclusion board, the accuracy axis and recent rejected labels are copied as they are. One classifier from the delusion-check layer does the classification; return-related items are never published even when the source has them. Verify a record →
Delusion check as_of 2026-09-29 16:43:06 +09:00 · ML board as_of 2026-09-21 16:46:38 +09:00
- rows
- 1
- unique
- 1
- registered
- 0
- scored
- 1
- rejected
- 0
- passed some legs
- 0
- rows
- 1,516
- registered
- 695
- scored
- 492
- rejected
- 270
- passed some legs
- 59
- registered (unique)
- 436
- rejected (unique)
- 75
Rejected unique total 75 · rejected labels withheld total 1
- models
- 110
- promoted
- 11
- observing
- 17
- descriptive
- 69
- rejected
- 13
- targets
- 9
- confirmed
- 1
- insufficient
- 8
- budget exhausted
- 4
Withheld (data terms of use under review)
| Category | n |
|---|---|
| value | 3 |
| venture | 3 |
| rotation / ETF | 2 |
| trading rules | 1 |
| volatility | 1 |
- Source: Korea Exchange statistical information (attribution under Article 10(3) of the Korea Exchange terms of use)
- Only model-performance statistics are shown; no quotation values are published.
- Index names such as KOSPI, S&P 500 and NASDAQ are trademarks of their respective owners; there is no affiliation with or endorsement of this site.
- Data sources in summary: public data (filings and statistics), exchange statistical information, and the program's own prediction and scoring record. No price or index levels and no security identifiers are published.
방향 없음·권유 아님 — 모델 예측과 채점 기록만 싣는다No direction, no recommendation — only model predictions and their scoring record.
Descriptions of the research system on this site introduce an internal research program of Root Inside Co., Ltd. and are provided for research and educational purposes only. They are not a solicitation to buy or sell any financial instrument and do not substitute for your own judgment. All investment decisions, and their outcomes, are the sole responsibility of the user.