# Decile 10 · The road after the deal

二哥丰 / ยี่กอฮง · 13 September 2026 MYT. Experimental runway distributions for PENGUUSDT. This build is implemented and tested; predictive qualification has not been earned.

## Open the dashboards

- [Runway](https://ergefeng-omai-workspaces.aukarwai.chatgpt.site/runway/): the new projection appears first, with the underlying Console available below.
- [Console](https://ergefeng-omai-workspaces.aukarwai.chatgpt.site/console/): existing readings and a new runway section below them.
- [Intraday](https://ergefeng-omai-workspaces.aukarwai.chatgpt.site/intraday/): existing readings and the same runway section.
- [Previous Console](https://ergefeng-omai-workspaces.aukarwai.chatgpt.site/previous-ui/v001/console/) and [previous Intraday](https://ergefeng-omai-workspaces.aukarwai.chatgpt.site/previous-ui/v001/intraday/) preserve the prior UI. They retain its earlier refresh behaviour.
- [Mobile decile command](https://ergefeng-omai-workspaces.aukarwai.chatgpt.site/deciles/) remains separate and unchanged.

## What the new reading means

At a qualifying closed-minute Decile 10 observation, preserve the origin price, origin measuring rung, data time, actual receipt/issue time and model version. Further qualifying minutes within the same ten-minute cluster do not repeatedly reapply the first-deal model or shift its origin. The raw source observations remain exportable. On a cold start, “first observed” does not prove the first qualifying minute was witnessed. Historical reconstruction is labelled and waits for its scan to finish before issuing a projection; Today can paint immediately while that scan runs.

Select 26, 53, 106, 212 or 424 minutes. These are observation checkpoints, not exit instructions. Both upward and downward room are shown as the central 50% of historical excursions, a median and a 90th-percentile stretch. Price levels use PENGU's six decimals. The model supplies probabilities of reaching one origin-rung, conditional arrival timing and contrary movement before the arrival minute. Movement inside that arrival minute has unresolved ordering. Both-side boundary probabilities explicitly include neither and same-minute ambiguity.

In plain terms: a dealt hand starts a measuring tape. You can see how much road similar historical hands had above and below that starting price. The clock can span several 53-minute candles. A 50% band is designed to include about half the outcomes, not all of them; a stretch level is not a promised destination.

This build does not qualify LONG or SHORT. An out-of-range volatility state withholds a new size estimate rather than extrapolating. The existing dated estimate, if any, stays visibly anchored to its original deal. The 53-minute chart uses a provisional 08:00 MYT daily reset and a final 9-minute segment. The supplied 08:00/08:53 opens do not establish TradingView's day-boundary convention; that remaining uncertainty is displayed.

## Actual research result

4,225 first-D10 clusters were reconstructed; 4,156 had complete outcomes. Chronological development used 3,111 training events, 419 validation events and 626 later diagnostic events across 142 MYT days, with boundary-overlapping outcomes removed. Existing inputs supplied 896,145 contiguous Binance-derived OHLC minutes. No bulk market download was needed for this comparison.

- Volatility-conditioned baseline B lowered later-test excursion quantile error by **5.22% in frozen rungs** and **4.59% in price-percent units**, relative to the pooled first-deal baseline A.
- Additional 13/26/53/106/212 geometry C worsened B's rung error by **0.40%**. Its 95% MYT-day block interval for improvement was negative throughout, approximately −0.674% to −0.144%. Geometry is excluded from serving.
- Central 50% bands covered 46.6–54.5% across the two directions and five checkpoints in that diagnostic test. This is band coverage, not a directional win rate.

The historical trigger was rescored with a fixed later model whose training cutoff is not established. These are exploratory historical diagnostics, not untouched or prospective evidence. The serving table was refitted on completed development events; quoted test figures concern the earlier chronological fit. The matched non-D10 comparison did not establish a special directional or size advantage caused by Decile 10. Source-derived mathematical constants and fitted SHEFA models were not transplanted into this new forecast.

Full definitions, gates, subgroup metrics, calibration errors, control matching and censoring are in the research report and exported ledgers. No costs, P&L or financial amounts enter the new analysis.

## Refresh and the Today row

The old Today scan finished once and stopped updating. The replacement inserts the newest closed-minute D10 reading first, then reconstructs older minutes progressively. Closed Binance WebSocket updates can deliver directly; visible pages check REST every ten seconds as fallback. Price streaming and slower open-interest requests do not block that D10 update. Five initial data pages now load concurrently. A retained device cache can paint first, keeping its source time.

- While reading: automatic updates and manual refresh preserve scroll position.
- Returning from another browser tab/app: refresh immediately and scroll to the top once.
- Regaining the internet while already reading: retry without scrolling to the top.
- The original forecast clock now uses its data-origin time, so a retained forecast does not drift forward whenever the display repaints.

Missing/invalid f15 stays missing for that exact minute. A previous-minute value is never substituted. A minute cannot be treated as complete before Binance provides its completed data. HTTP 429/418 applies host-wide backoff rather than aggressive retries. See [Binance's public API guidance](https://developers.binance.com/en/docs/products/derivatives-trading-usds-futures/general-info) and [page visibility behaviour](https://developer.mozilla.org/en-US/docs/Web/API/Page_Visibility_API).

## Ledger, local files and Cloudflare

The hosted Site adds an authenticated immutable forecast ledger scoped to the signed-in user. Duplicate records retrieve the first stored original rather than causing “different content” storage failure. Returns to the page also fetch newly saved shared records. During a storage interruption, unsaved forecasts remain exportable in the open tab. Export full JSON or the measurement CSV from the projection history area. Source observations, including missing f15 and their original timestamps, accompany the full export.

The standalone Cloudflare upload contains all dashboard scripts, model tables and fonts. Extract it and upload its contents to Cloudflare Pages; its root index also links directly to the three dashboards. Local index files work without a build process. Market refresh still needs internet access to Binance. This static upload has no authenticated online ledger service: its banner says to export the tab's observations before closing. The separate server-source package includes the authenticated Sites Worker and migrations; do not expose a trusted identity-header API through an unprotected public Worker.

All local files are under `E:\ergefeng\builds\decile-runway\v001`. Final Site source is `hosting-v005`, portable files are `cloudflare-upload\v002`, numerical work is `model`, delivery changes are `refresh`, and immutable server code is `record-api`. Earlier source/build versions remain intact. The original decile utility service and its tunnel were not restarted or edited.

## Validation limits

2,076 model assertions include Python/JavaScript parity, missing-score recovery, quantile/probability coherence, future/prefix checks and 53-minute boundary handling. Delivery changes have 37 lifecycle/application checks. The record API has 23 grouped checks. The final panel has ten integrated checks using real fitted model data, including duplicate records, first-cluster preservation and foreground ledger retrieval. Independent review found and corrected record-ID/schema and provisional-cluster integration errors.

Five live Binance requests returned HTTP 200 in 0.414 seconds from Windows at approximately 16:51 MYT on 13 September 2026. That is a connection check, not an iPhone speed guarantee. OMAI fonts, semantic light/dark palettes, mobile stacking and stable numeric styles are packaged. No physical iPhone 15 Pro Max, iOS Safari or iOS Chrome test, and no new visual-fidelity certification, is claimed.