Primary Data Source
EODHD supplies the primary end-of-day US and Hong Kong market data. PatternLedger builds weekly bars from stored daily bars.
Public Method
Data, Updates & Research Limits
PatternLedger uses end-of-day market data. It is not an intraday feed, a broker, or an automatic trading service.
Daily Updates
Each market's Daily boards update after that market closes and the required price data passes coverage checks.
Weekly Updates
Weekly boards update after the final market session of the week. This includes holiday-shortened weeks.
Delays and Limits
Exchange holidays, source delays, failed coverage checks, or maintenance can delay publication. The board date shows the latest included market session. All results are research evidence, not financial advice or a promise of future performance.
Platform Status
Data Freshness & Scan Audit
These checks apply before scanner-backed current views are trusted. They are data-readiness checks, not ranking rules.
Data Readiness Gates
1. Market coverage gate
Current scanner-backed views wait until daily bar coverage reaches at least 95.0% for the required market date.
2. Latest-bar availability
A ticker is considered current only if its stored OHLC history includes the exact required daily bar for its market. A missed-date recovery first checks the normal coverage gate and spends no provider quota when that gate is already met. Below the gate, it compares only active scanner-eligible tickers still missing that exact date with the provider's advertised exchange-bulk cost. EODHD's entire-exchange price request costs 100 API-call units, so 100 or more missing tickers use that one exact-date bulk request; a smaller recovery uses per-symbol requests in the configured symbol-batch size. Recovery validates every returned symbol, market, Daily timeframe, and date before storage. One invalid response stores nothing from that batch on the per-symbol path, while earlier valid batches remain durable; a retry recalculates the missing set and resumes without refetching completed batches. A bulk response that remains below the coverage gate fails closed instead of cascading into a higher-cost per-symbol fallback. Recovery never substitutes a dominant or newer date, and truncates both pattern detection and liquidity history at the requested candle. Exact-date detector work is staged locally in deterministic symbol batches keyed to the market, candle date, current detector revision, active roster and metadata, and configured history-batch size. Scheduled market-close scans use their required board date on this same resumable path, so a stopped scan reuses every compatible completed detector batch. A transient database failure therefore resumes only the missing compatible batches of detector work; stale, corrupt, or differently partitioned checkpoints are rejected. The prior board remains visible until every symbol batch is complete, then the replacement pattern rows and completed scan-run status commit in one database transaction. Latest-board and publication-snapshot selection always prefers the newest completed candle date; only reruns of that same date are resolved by their newest completion or start time. A later historical replay therefore cannot demote a newer market session. Normal daily refreshes also request a bounded recent overlap, 21 calendar days by default, so provider corrections replace the matching stored dates without requesting full history.
2a. Daily provider-quota reserve
Every charged EODHD request reserves its documented request-unit cost in a credential-free daily ledger before it is sent. The ledger is locked across concurrent workers and survives process restarts; the guard uses the higher of its durable estimate and the periodically observed account-wide counter. With the standard 30,000-unit daily limit, recovery, history backfill, corporate-action history rebase and confirmation, and on-demand ticker-history repair stop before the next request would raise usage above 22,000, preserving 8,000 units for current scheduled market-close data. Current symbol-universe reconciliation, latest exchange-bulk prices, and live corporate-action discovery have priority and may use the reserve, but no work may knowingly exceed the full daily limit. Current exchange-bulk bars are stored before reserve-bounded history work begins, and a symbol batch that does not fit is rejected before any of its concurrent requests are issued.
3. Corporate-action adjustment integrity
Before a live provider-backed scan, the exchange-wide EODHD bulk dividend and split feeds are reconciled against a durable action ledger. Reconciliation is serialized per database and market across local processes; an overlapping scheduler or manual run exits before provider calls or history replacement and retries later instead of racing the atomic ledger update. A new ledger first checks the prior 120 calendar days to cover pre-ledger adjustment seams; a run that advances the certified cutoff replays a rolling seven-day overlap for delayed or revised actions, while an immediate retry of the same already-certified cutoff reuses its checkpoint without spending quota on the overlap again. Discovery is fetched and durably checkpointed in five-session chunks, so a provider failure preserves every completed chunk and the retry resumes from the rolling overlap instead of repeating the full initial window. Temporary network failures and HTTP 408, 425, 429, 500, 502, 503, or 504 responses get at most two retries after one second and then two seconds. Every retry reserves the full provider quota cost before it is sent, and other HTTP failures stop immediately. An EODHD HTTP 402 sets a credential-safe provider cooldown until the next UTC reset and writes that cooldown durably, so even a replacement scheduler process blocks before sending another request and then automatically resumes the checkpointed retry after that reset. The provider's complete symbol list is reconciled atomically before pending rebases: renamed or delisted aliases that disappeared from that list are deactivated, while an empty, duplicate, cross-market, or suspiciously truncated list is rejected and rolled back. If the provider retains both sides of a ticker change, the old code is retired only through an audited primary-source registry containing the exact old code, new code, effective date, expected identity metadata, and regulator or issuer URL. Stored chronology must also show that the old series predates the effective date, the successor begins no later than it, and the successor ends later. Fuzzy same-name or price-similarity guesses are never sufficient. Before discovery, the raw ledger is checked for verified old-code events even when their symbol row is already inactive; either such an event or an effective old code still marked active forces a complete universe reconciliation instead of trusting the normal seven-day universe cache. Once the effective date passes, any provider snapshot that would reactivate the verified old code without its active successor, or that no longer passes the registry metadata and chronology checks, is rejected and rolled back. In the same universe transaction, every pending event under the verified old code is retargeted to the successor; the same protected transfer occurs if that verified old code has disappeared from the complete provider list, including repair of a pre-existing hidden pending row under an already-inactive old code. Later discoveries are canonicalized inside their storage transaction only while the old code is inactive and the successor is active, with just the registry symbol rows locked against a concurrent universe refresh. Canonicalization uses the overall scan cutoff rather than an earlier five-session chunk end, so overlap discovery cannot recreate a hidden old-code row. The provider action ID and raw payload preserve original-ticker provenance, and uncertified old bars are never stitched onto the successor. A verified inactive alias therefore cannot hide a pending action or poison an otherwise valid batch. A fully zero OHLC and adjusted-close row with zero or missing volume is an EODHD non-trading/listing placeholder and is excluded; any partial, traded, or otherwise non-positive OHLC row still fails validation. Date-specific recovery certifies and applies only actions on or before that board date; later pending actions stay pending. No affected history rebase begins until discovery has reached the requested cutoff. Every affected active ticker is then refetched across its complete stored Daily history through that cutoff, batch-validated, and its adjusted OHLC series is replaced atomically. The rebase must reach the ticker's latest stored session at or before that cutoff; it never fabricates or requires an action-date candle when the ticker did not trade that session, but it still rejects a provider response that truncates a session already held in storage. Its validation baseline is clipped to the same board-date cutoff, so a later stored bar cannot masquerade as a material edge revision and trigger a duplicate confirmation request. An exact-date retry still constructs the provider needed to resume pending rebases without repeating already-completed targeted bar-fill batches. The unpaginated current full provider series is authoritative, so obsolete interior dates, stored edge dates, and old row counts are not preserved rigidly. Any first-date or last-date change, a revision larger than the greater of five rows or 0.5% of stored rows, or one that removes more than three consecutive stored rows must reproduce exactly in one independent refetch using a SHA-256 fingerprint of every date and OHLCV row before it can replace storage. Even a confirmed material revision must average at least one row per 31 calendar days across its full span. A too-sparse response also fails. OHLC ordering permits only a scale-aware floating-point tolerance of the larger of 0.000000000001 price units or one trillionth of the candle's price scale; substantive range inconsistencies still fail. A market-wide discovery, fetch, or atomic replacement failure remains fail-closed. An isolated missing, invalid, or non-deterministic full history instead leaves that ticker's pending action and stored bars untouched, quarantines the ticker from Daily and Weekly detection for that run, and retries it automatically on later runs; every unaffected ticker continues, so mixed-vintage prices are never scanned and one provider defect cannot freeze the market board.
4. Weekly bar finality
Weekly signals use the completed final trading session of each market week. Holiday-shortened weeks close on the last open session, and startup or local checkpoint-triggered recovery repairs a missed weekly run after daily market-close data is current, while an unfinished week is not treated as final.
5. Business-description completeness
Company and receipt descriptions are displayed only when they are source-backed or curated. Name-only, generic publicly traded company, pending-enrichment, and truncated summaries are rejected from display, but an unavailable description is omitted rather than blocking otherwise complete time-sensitive scanner results; enrichment can finish independently on a later run.
6. Durable market-close completion
A market-close update is complete only after its scan, required incremental backtests, market report or combined Top Trades and AI work, snapshot checks, and publication have succeeded. Each stage and each bounded detector or backtest batch has a durable local checkpoint. Local SQLite is authoritative for market symbols, prices, saved signals, backtests, composite reports, scheduled AI reviews, and historical boards. Market calculations and local status checks do not query Neon. Neon stores account data only. A restart resumes from the first unfinished stage without repeating completed work or unneeded market-data requests.
7. Publication snapshot fidelity
Publication builds a staged, bounded SQLite web snapshot from one read-only view of the authoritative local database. It carries the visible scanner boards, chart history, exact aggregate current-rule backtest evidence, compact recent trade examples, every archived composite report, required AI and fundamental reviews, and every AI trade-journal entry. Private My Tracked Trades entries stay in authenticated account storage and are not embedded in the website snapshot. The database snapshot keeps the latest Daily and Weekly boards. A small Top Signal Archive file stores the final Top 10 lists for the newest Daily and Weekly dates. A separate compact protected archive inside the website deployment keeps the two prior Daily board dates. Together, they cover the latest 3 Daily board dates that are still too recent for the delayed public R2 archive. Older eligible Daily and Weekly boards come from R2. For each retained date, the database snapshot keeps at most 10 website-eligible raw hits in each market, timeframe, pattern, quality, and date slot. This raw-hit cap does not remove or shorten a stored Top Trades report or saved Top Signal Archive list. Ordinary charts keep the configured recent-bar baseline. Current-rule trendlines keep every bar from the oldest validated anchor plus 199 earlier frame bars for correct MA100 and MA200 values. Source fingerprints and required artifacts must pass before an atomic swap. A failed build or check leaves the previous complete snapshot unchanged. Snapshot construction calls neither Neon nor the market-data provider. Vercel serves only these validated saved files.
7a. Delayed public board archive
A separate public archive keeps historical Signal Desk, Top Trades, and complete AI-ranked boards outside Neon. Normal publication is built from the authoritative local SQLite database and saved composite reports. When automatic R2 publication is enabled, each website-snapshot publication refreshes the delayed archive before the new snapshot can be pushed. An R2 failure stops that snapshot publication, so a date cannot fall out of the protected recent set before its delayed copy is available. A one-time recovery can add older verified boards from the former database. Once a delayed board is published, later builds keep and reuse it when its stored file is valid, so a partial source or database move cannot remove that date or cause another database read for the same board. A board is released only after every included market source date is at least 3 real exchange sessions behind that market's latest completed board. Each board file has a content hash in its name, so a cached historical file cannot silently change. A small date index points to the exact files, and publication uploads only files named by that index. Vercel reads these public files from Cloudflare R2 when a historical board is not in the bounded website snapshot. Historical board views do not query Neon or the market-data provider. Current paid boards never enter the public archive before the delay ends.
8. Relative-shock continuity gate
A structurally valid candle can still be excluded when its within-session high-low span, measured against the previous close, is both at least 15% and at least 20 times the ticker's own trailing 20-session median intraday span, with a 0.25% baseline floor. An overnight earnings or news gap is not enough by itself: an orderly one-session repricing with a normal intraday span remains eligible. This high-confidence relative test isolates exceptional provider contamination without treating routinely volatile instruments as bad merely because their normal moves are large. A flagged Daily candle creates a per-symbol continuity break: stored provider prices are never deleted, replaced, interpolated, or manually fabricated; Daily detection and liquidity restart with the clean suffix, and Weekly detection excludes the entire affected calendar week before aggregation. The affected ticker is omitted when the required board candle is flagged, while the rest of the market scan continues. Scanner geometry, backtest signals and entries, horizon outcomes, and AI outcome calibration may not cross a continuity break. Saved scanner hits are also rejected from backtests when their OHLC no longer matches the currently stored signal candle after an adjusted-history revision.
Universe Data Coverage
Coverage shows whether the required US/HK daily bars have arrived for the current publishing cycle.
US Universe
Required date 2026-08-27 |
latest stored 2026-08-27 |
threshold 95.0%
Coverage
100.0%
With Latest Bar
521
Active Universe
521
Missing Latest
0
| Date | Symbols | Coverage |
|---|---|---|
| 2026-08-27 | 521 / 521 | 100.0% |
| 2026-08-26 | 521 / 521 | 100.0% |
| 2026-08-25 | 521 / 521 | 100.0% |
| 2026-08-24 | 521 / 521 | 100.0% |
| 2026-08-21 | 521 / 521 | 100.0% |
| 2026-08-20 | 520 / 521 | 99.8% |
| 2026-08-19 | 521 / 521 | 100.0% |
| 2026-08-18 | 520 / 521 | 99.8% |
HK Universe
Required date 2026-08-28 |
latest stored 2026-08-28 |
threshold 95.0%
Coverage
100.0%
With Latest Bar
279
Active Universe
279
Missing Latest
0
| Date | Symbols | Coverage |
|---|---|---|
| 2026-08-28 | 279 / 279 | 100.0% |
| 2026-08-27 | 279 / 279 | 100.0% |
| 2026-08-26 | 279 / 279 | 100.0% |
| 2026-08-25 | 279 / 279 | 100.0% |
| 2026-08-24 | 279 / 279 | 100.0% |
| 2026-08-21 | 279 / 279 | 100.0% |
| 2026-08-20 | 279 / 279 | 100.0% |
| 2026-08-19 | 279 / 279 | 100.0% |
Scan Audit
The latest scanner jobs by market and timeframe. If a board looks stale, check these run dates first.
HK weekly
2026-08-28 11:37
Candle
2026-08-28
Symbols
3750
Hits
573
Kind
market_close
HK daily
2026-08-28 11:34
Candle
2026-08-28
Symbols
3713
Hits
656
Kind
market_close
US daily
2026-08-27 22:58
Candle
2026-08-27
Symbols
11344
Hits
4218
Kind
scheduled
HK daily
2026-08-27 12:15
Candle
2026-08-27
Symbols
3732
Hits
718
Kind
market_close
US daily
2026-08-26 22:57
Candle
2026-08-26
Symbols
11326
Hits
4216
Kind
scheduled
HK daily
2026-08-26 12:26
Candle
2026-08-26
Symbols
3736
Hits
685
Kind
market_close
US daily
2026-08-25 23:19
Candle
2026-08-25
Symbols
11362
Hits
3965
Kind
scheduled
HK daily
2026-08-25 14:33
Candle
2026-08-25
Symbols
3733
Hits
765
Kind
market_close
US daily
2026-08-24 23:37
Candle
2026-08-24
Symbols
11352
Hits
4055
Kind
scheduled
HK daily
2026-08-24 11:54
Candle
2026-08-24
Symbols
3714
Hits
696
Kind
market_close
US weekly
2026-08-21 23:34
Candle
2026-08-21
Symbols
12126
Hits
3595
Kind
scheduled
US daily
2026-08-21 23:30
Candle
2026-08-21
Symbols
11361
Hits
4088
Kind
scheduled
HK weekly
2026-08-21 11:20
Candle
2026-08-21
Symbols
3751
Hits
479
Kind
market_close
HK weekly
2026-08-18 04:04
Candle
2026-07-31
Symbols
400
Hits
566
Kind
manual
US weekly
2026-08-18 04:04
Candle
2026-07-31
Symbols
3432
Hits
4783
Kind
manual
HK weekly
2026-08-18 04:04
Candle
2026-07-03
Symbols
480
Hits
685
Kind
manual
US weekly
2026-08-14 23:24
Candle
2026-08-14
Symbols
12252
Hits
3234
Kind
scheduled
HK weekly
2026-08-14 12:20
Candle
2026-08-14
Symbols
3742
Hits
467
Kind
market_close
US weekly
2026-08-07 23:55
Candle
2026-08-07
Symbols
12215
Hits
4141
Kind
market_close
US weekly
2026-08-02 03:31
Candle
2026-07-31
Symbols
12217
Hits
4783
Kind
market_close
Workspace Tool
Signal Desk
Signal Desk is the scanner workspace. It shows raw daily and weekly pattern signals by market, timeframe, and pattern group.
Pattern Signals Shown In Signal Desk
These are the scanner labels shown in Signal Desk. Strict means cleaner geometry or candle quality; loose means still valid, but weaker.
| Pattern | Family | Side / Style | Rule | Strict vs Loose | Removal / Weakness |
|---|---|---|---|---|---|
|
Hammer
Backtest scored
|
Candlestick |
Buy
Contrarian
|
Requires at least three prior bars of measured downside pressure plus an ATR-material lower-wick rejection inside a recent-low zone; an exact new 10/20-bar low earns extra credit but is not the only valid support retest. | Strict requires stronger wick/body cleanliness, close location, ATR significance from 20 prior bars, recent-low-zone location, and a full 8-bar daily or 6-bar weekly downtrend context; shorter histories are capped at Loose. | Weakens quickly if price closes below the hammer low or fails to reclaim nearby resistance. |
|
Shooting Star
Backtest scored
|
Candlestick |
Sell
Contrarian
|
The Shooting Star rule requires at least three prior bars of measured upside pressure plus an ATR-material upper-wick rejection inside a recent-high zone, scored symmetrically with Hammer. | Strict symmetrically requires stronger wick/body cleanliness, close location, ATR significance from 20 prior bars, recent-high-zone location, and a full 8-bar daily or 6-bar weekly uptrend context; shorter histories are capped at Loose. | Weakens quickly if price closes above the shooting-star high or fails to lose nearby support. |
|
Bullish Engulfing
Backtest scored
|
Candlestick |
Buy
Contrarian
|
Requires at least three bars of measured downside pressure before the two-candle pattern, a meaningful prior bearish body, substantial bullish body coverage and expansion, ATR-material range, and a strong close location. | Strict requires a full body engulf with body expansion, meaningful prior and current bodies versus an established 20-prior-bar ATR, a strong close, and confirmed downside context across the full 8-bar daily or 6-bar weekly window; Loose permits near-engulfing coverage or shorter qualifying history. | Weakens if the engulfing low is lost or the next candles fail to follow through. |
|
Bearish Engulfing
Backtest scored
|
Candlestick |
Sell
Contrarian
|
Requires at least three bars of measured upside pressure before the two-candle pattern, a meaningful prior bullish body, substantial bearish body coverage and expansion, ATR-material range, and a strong close location. | Strict requires a full body engulf with body expansion, meaningful prior and current bodies versus an established 20-prior-bar ATR, a strong close, and confirmed upside context across the full 8-bar daily or 6-bar weekly window; Loose permits near-engulfing coverage or shorter qualifying history. | Weakens if the engulfing high is reclaimed or the next candles fail to follow through. |
| Rising Support | Trendline |
Buy
Trend Following
|
Searches chronologically from each older confirmed swing-low wick to later confirmed swing lows inside a timeframe-specific horizon. Daily lines require the first and newest confirmed anchors to span at least three calendar months and normally use an oldest anchor no more than 18 months old; a currently interacting or freshly broken 18-to-24-month line is an extended fallback only when no standard Daily line qualifies, and 24 months is the absolute Daily maximum. Weekly lines require at least a 12-month confirmed-anchor span and normally use an oldest anchor no more than 10 years old; a currently interacting or freshly broken 10-to-15-year line is an extended fallback only when no standard Weekly line qualifies. Origins older than the applicable absolute maximum are excluded. Two independently spaced anchors are sufficient; a third, fourth, or later contact increases confidence and score but is not required to form the line. Projection time after the newest anchor does not make a younger line mature. The exact anchor line must rise by at least 4% Daily or 5% Weekly and remain at or below every candle low both between the anchors and after the newest anchor. Nearby contacts are recognized inside a bounded volatility band of about one ATR, capped at 1.0% Daily or 1.5% Weekly, but that band never moves the exact wick boundary or excuses a crossing. Scanner and backtest evidence additionally require current-price interaction or a fresh held breakdown. A held breakdown is eligible on the break period and the next completed period only; at age two or more it is stale and excluded. An intact standard-horizon line that is farther away may remain as chart context. For active candidates in the same current support zone, the longest, earliest still-clear origin inside the preferred horizon is primary. Intentional product decision, not missing data or a scanner error: cash-like and mechanically accruing funds are omitted from Trendlines and directional Channels because their normal drift adds little trend-discovery value and can crowd out more informative market moves. Before scanner ranking and Top-10 selection, a shared structural-trend eligibility filter persistently excludes funds whose apparent trend is largely built into a cash-management, ultrashort or short-duration, Treasury-bill, low-risk floating-rate, box-spread, short-duration TIPS, or near-maturity mandate. Fund names and primary business descriptions catch future equivalents, while a small audited override list covers SHY, BOXX, TBIL, TFLO, STIP, RAVI, and FLRN. Recent volatility does not temporarily re-admit an excluded instrument. Ordinary equities and economically rate- or credit-sensitive funds such as HYG, SPHY, BIV, TIP, IEF, and CMF remain eligible. Eligibility is separate from Top Charts diversity: an eligible fund may be absent from one Trendlines or directional Channels list when a higher-ranked fund already has a persistently near-identical return path. Filtering happens before chart places are filled, so the next eligible assets take the open positions, with display diversity preferring distinct return paths; stored prices and historical report artifacts are not deleted. Multi-ticker Top Charts cards use a compact recent-detail window capped at 220 Daily or 130 Weekly candles. When the first anchor predates that window, the exact scored line is projected into view and the true first and newest confirmed anchor dates are shown; the Full Anchor Chart link opens the Research page, which begins at the oldest retained validated anchor and includes every subsequent bar through the current interaction without unrelated earlier candles. This display-only crop never changes detection, anchors, score, or invalidation. Active and broken lines are selected separately. Horizontal levels belong to range/support analysis, not trendlines. A terminal-wick refit remains chart-only until that swing confirms. | Both Strict and Loose must pass the structural-trend eligibility filter and require at least two independent anchors, a three-calendar-month Daily or 12-month Weekly confirmed-anchor span, an origin inside the applicable 18-month Daily or 10-year standard Weekly horizon, material upward slope, current interaction or a fresh held break, and complete wick clearance with zero crossings. An 18-to-24-month Daily or 10-to-15-year Weekly origin is fallback-only when no standard line qualifies. Two- and three-anchor lines remain Loose; Strict additionally requires at least four independent contacts, at least three confirmed reactions, strong line fit, and strong anchor consensus. | Any wick below the exact projected line between the anchors invalidates that pair. Any non-recent wick crossing after the newest anchor also retires it as active support; a close-confirmed breakdown may appear separately only while held, on the break period or the next completed period. At two or more periods since the break it is stale and excluded. A Daily period is one trading session; a Weekly period is one completed week. Daily origins older than 24 months and Weekly origins older than 15 years are invalid. An 18-to-24-month Daily or 10-to-15-year Weekly line is also withheld whenever a standard-horizon line qualifies or the older line is not interacting with current price. A live board's scored formation must pass today's stored-OHLC validation and render exactly; corrected bars that cross it retire the current recommendation, and an unrelated support line can never replace it. Delayed and archived Top Trades boards instead preserve the ranked rows and AI reviews stored when that report was published, so a later rule revision does not rewrite the historical record. |
| Falling Resistance | Trendline |
Sell
Trend Following
|
Searches chronologically from each older confirmed swing-high wick to later confirmed swing highs inside a timeframe-specific horizon. Daily lines require the first and newest confirmed anchors to span at least three calendar months and normally use an oldest anchor no more than 18 months old; a currently interacting or freshly broken 18-to-24-month line is an extended fallback only when no standard Daily line qualifies, and 24 months is the absolute Daily maximum. Weekly lines require at least a 12-month confirmed-anchor span and normally use an oldest anchor no more than 10 years old; a currently interacting or freshly broken 10-to-15-year line is an extended fallback only when no standard Weekly line qualifies. Origins older than the applicable absolute maximum are excluded. Two independently spaced anchors are sufficient; a third, fourth, or later contact increases confidence and score but is not required to form the line. Projection time after the newest anchor does not make a younger line mature. The exact anchor line must fall by at least 4% Daily or 5% Weekly and remain at or above every candle high both between the anchors and after the newest anchor. Nearby contacts are recognized inside a bounded volatility band of about one ATR, capped at 1.0% Daily or 1.5% Weekly, but that band never moves the exact wick boundary or excuses a crossing. Scanner and backtest evidence additionally require current-price interaction or a fresh held breakout. A held breakout is eligible on the break period and the next completed period only; at age two or more it is stale and excluded. An intact standard-horizon line that is farther away may remain as chart context. For active candidates in the same current resistance zone, the longest, earliest still-clear origin inside the preferred horizon is primary. Intentional product decision, not missing data or a scanner error: cash-like and mechanically accruing funds are omitted from Trendlines and directional Channels because their normal drift adds little trend-discovery value and can crowd out more informative market moves. Before scanner ranking and Top-10 selection, a shared structural-trend eligibility filter persistently excludes funds whose apparent trend is largely built into a cash-management, ultrashort or short-duration, Treasury-bill, low-risk floating-rate, box-spread, short-duration TIPS, or near-maturity mandate. Fund names and primary business descriptions catch future equivalents, while a small audited override list covers SHY, BOXX, TBIL, TFLO, STIP, RAVI, and FLRN. Recent volatility does not temporarily re-admit an excluded instrument. Ordinary equities and economically rate- or credit-sensitive funds such as HYG, SPHY, BIV, TIP, IEF, and CMF remain eligible. Eligibility is separate from Top Charts diversity: an eligible fund may be absent from one Trendlines or directional Channels list when a higher-ranked fund already has a persistently near-identical return path. Filtering happens before chart places are filled, so the next eligible assets take the open positions, with display diversity preferring distinct return paths; stored prices and historical report artifacts are not deleted. Multi-ticker Top Charts cards use a compact recent-detail window capped at 220 Daily or 130 Weekly candles. When the first anchor predates that window, the exact scored line is projected into view and the true first and newest confirmed anchor dates are shown; the Full Anchor Chart link opens the Research page, which begins at the oldest retained validated anchor and includes every subsequent bar through the current interaction without unrelated earlier candles. This display-only crop never changes detection, anchors, score, or invalidation. Active and broken lines are selected separately. Horizontal levels belong to range/resistance analysis, not trendlines. A terminal-wick refit remains chart-only until that swing confirms. | Both Strict and Loose must pass the structural-trend eligibility filter and require at least two independent anchors, a three-calendar-month Daily or 12-month Weekly confirmed-anchor span, an origin inside the applicable 18-month Daily or 10-year standard Weekly horizon, material downward slope, current interaction or a fresh held break, and complete wick clearance with zero crossings. An 18-to-24-month Daily or 10-to-15-year Weekly origin is fallback-only when no standard line qualifies. Two- and three-anchor lines remain Loose; Strict additionally requires at least four independent contacts, at least three confirmed reactions, strong line fit, and strong anchor consensus. | Any wick above the exact projected line between the anchors invalidates that pair. Any non-recent wick crossing after the newest anchor also retires it as active resistance; a close-confirmed breakout may appear separately only while held, on the break period or the next completed period. At two or more periods since the break it is stale and excluded. A Daily period is one trading session; a Weekly period is one completed week. Daily origins older than 24 months and Weekly origins older than 15 years are invalid. An 18-to-24-month Daily or 10-to-15-year Weekly line is also withheld whenever a standard-horizon line qualifies or the older line is not interacting with current price. A live board's scored formation must pass today's stored-OHLC validation and render exactly; corrected bars that cross it retire the current recommendation, and an unrelated resistance line can never replace it. Delayed and archived Top Trades boards instead preserve the ranked rows and AI reviews stored when that report was published, so a later rule revision does not rewrite the historical record. |
| Double Bottom | Tops/Bottoms |
Buy
Contrarian
|
Requires two prominent, independently spaced lows inside an ATR-adjusted touch zone, a body-backed intervening reaction neckline, orderly prior downside pressure, and either a current support rejection or a confirmed break that still holds. The selected pivot wicks define the zone; its lower outer edge is the displayed support boundary, and every candle must clear it with its full range before any separately labelled failed-zone breakdown. | Complete full-candle clearance of the support boundary before any failed-zone break is mandatory for both Strict and Loose. Strict additionally requires a currently held, confirmed break plus strong pivot similarity, reaction depth, prominence, prior trend, and zone integrity; an unconfirmed near-support rejection is capped at Loose. | Any wick below the lower outer edge retires normal long evidence. A failed-zone breakdown may appear separately only when its break candle is the first boundary crossing, occurs strictly after the final pivot, is confirmed by two consecutive closes or one decisive volume-backed close, remains beyond the exact edge, and is no older than five Daily bars or three Weekly bars. |
| Double Top | Tops/Bottoms |
Sell
Contrarian
|
Requires two prominent, independently spaced highs inside an ATR-adjusted touch zone, a body-backed intervening reaction neckline, orderly prior upside pressure, and either a current resistance rejection or a confirmed break that still holds. The selected pivot wicks define the zone; its upper outer edge is the displayed resistance boundary, and every candle must clear it with its full range before any separately labelled failed-zone breakout. | Complete full-candle clearance of the resistance boundary before any failed-zone break is mandatory for both Strict and Loose. Strict additionally requires a currently held, confirmed break plus strong pivot similarity, reaction depth, prominence, prior trend, and zone integrity; an unconfirmed near-resistance rejection is capped at Loose. | Any wick above the upper outer edge retires normal short evidence. A failed-zone breakout may appear separately only when its break candle is the first boundary crossing, occurs strictly after the final pivot, is confirmed by two consecutive closes or one decisive volume-backed close, remains beyond the exact edge, and is no older than five Daily bars or three Weekly bars. |
| Triple Bottom | Tops/Bottoms |
Buy
Contrarian
|
Requires three prominent lows in one ATR-adjusted touch zone plus two separate, meaningfully deep and reasonably balanced reaction legs. The lower outer edge remains at or below all three pivot wicks, and a qualifying triple subsumes overlapping double-bottom evidence from the same zone. | Complete full-candle clearance of the support boundary before any failed-zone break is mandatory for both Strict and Loose. Strict additionally requires balanced spacing and balanced depth across both independent reaction legs; the third touch receives no automatic score bonus. | Invalid if either intervening reaction leg is shallow or unbalanced, if any wick crosses the support boundary before the break candle of a valid failed-zone breakdown, or if a claimed neckline or failed-zone break is stale or no longer held. |
| Triple Top | Tops/Bottoms |
Sell
Contrarian
|
Requires three prominent highs in one ATR-adjusted touch zone plus two separate, meaningfully deep and reasonably balanced reaction legs. The upper outer edge remains at or above all three pivot wicks, and a qualifying triple subsumes overlapping double-top evidence from the same zone. | Complete full-candle clearance of the resistance boundary before any failed-zone break is mandatory for both Strict and Loose. Strict additionally requires balanced spacing and balanced depth across both independent reaction legs; the third touch receives no automatic score bonus. | Invalid if either intervening reaction leg is shallow or unbalanced, if any wick crosses the resistance boundary before the break candle of a valid failed-zone breakout, or if a claimed neckline or failed-zone break is stale or no longer held. |
| Ascending Channel | Channel |
Buy
Trend Following
|
Pairs support and resistance over one shared interval with volatility-adjusted tolerances, stable parallel width, independently spaced touches, confirmed reactions, and either a respected support test or a fresh break that still holds; a dual-edge candle cannot anchor or count as a touch on both boundaries. The same persistent structural-trend eligibility filter used by trendlines removes structurally carry-driven funds before ranking; recent volatility does not temporarily re-admit them. | Both quality levels must first pass the structural-trend eligibility filter. Strict then requires at least four independent touches on both boundaries, at least two confirmed reactions per boundary, strong slope agreement, high containment, and clean pre-break geometry. | A fund removed by the structural-trend eligibility filter is withheld before scanner ranking. Otherwise a break must occur after the final channel anchor and remain held by two closes or a decisive volume-backed close. Re-entry removes the break state; revised or later price history must still preserve the stored pre-break geometry. |
| Descending Channel | Channel |
Sell
Trend Following
|
Pairs resistance and support over one shared interval with volatility-adjusted tolerances, stable parallel width, independently spaced touches, confirmed reactions, and either a respected resistance test or a fresh break that still holds; a dual-edge candle cannot anchor or count as a touch on both boundaries. The same persistent structural-trend eligibility filter used by trendlines removes structurally carry-driven funds before ranking; recent volatility does not temporarily re-admit them. | Both quality levels must first pass the structural-trend eligibility filter. Strict then requires at least four independent touches on both boundaries, at least two confirmed reactions per boundary, strong slope agreement, high containment, and clean pre-break geometry. | A fund removed by the structural-trend eligibility filter is withheld before scanner ranking. Otherwise a break must occur after the final channel anchor and remain held by two closes or a decisive volume-backed close. Re-entry removes the break state; revised or later price history must still preserve the stored pre-break geometry. |
| Range Box | Channel |
Contextual
Contextual
|
Uses the same horizontal support/resistance levels for qualification, scoring, validation, and display; it requires independently spaced reactive touches and contained price action, while dual-edge candles are excluded from both touch tallies and treated as directionally ambiguous. | Strict requires at least four independent touches and two confirmed reactions at both exact displayed levels, strong flatness, high containment, and clean pre-break geometry. | Near-edge evidence is invalidated when its boundary fails. Breakout/breakdown evidence must occur after the final anchor and remain beyond the exact box level; re-entry into the box removes the directional break state. |
|
Bull Flag
Backtest scored
|
Flag |
Buy
Trend Following
|
Continuation setup after an impulse up, followed by compact controlled consolidation or breakout. | Strict requires an actual breakout, stronger pole, limited retracement, and compact flag structure. | Invalid if price loses the flag support/box after the setup. |
|
Bear Flag
Backtest scored
|
Flag |
Sell
Trend Following
|
Continuation setup after an impulse down, followed by compact controlled consolidation or breakdown. | Strict requires an actual breakdown, stronger pole, limited retracement, and compact flag structure. | Invalid if price reclaims the flag resistance/box after the setup. |
| 100/200 MA Support | Moving Average |
Buy
Trend Following
|
Evaluates the 100-day/week and 200-day/week averages independently after each has a full integrity lookback, then classifies an ATR-scaled support interaction as near, touch, genuine support bounce, or fresh support reclaim. | Strict is reserved for a genuine support bounce: a touched MA, no recent MA breaches, limited ATR-scaled overshoot, a meaningful bullish reaction candle that closes away from the average, an aligned multi-bar slope in the selected average, and a supportive non-flat regime. Near, touch, and reclaim states remain Loose. | Invalid if price closes back below the selected moving average, exceeds the ATR-scaled wick-overshoot or extension limit, or the average has too many recent soft/hard breaches. |
| 100/200 MA Resistance | Moving Average |
Sell
Trend Following
|
Evaluates the 100-day/week and 200-day/week averages independently after each has a full integrity lookback, then classifies an ATR-scaled resistance interaction as near, touch, genuine resistance rejection, or fresh resistance reclaim. | Strict is reserved for a genuine resistance rejection: a touched MA, no recent MA breaches, limited ATR-scaled overshoot, a meaningful bearish reaction candle that closes away from the average, an aligned multi-bar slope in the selected average, and a supportive non-flat regime. Near, touch, and reclaim states remain Loose. | Invalid if price closes back above the selected moving average, exceeds the ATR-scaled wick-overshoot or extension limit, or the average has too many recent soft/hard breaches. |
Archive And Latest Tape
0. Full delayed board dates
Current-access users get one complete date list. For Daily boards, it combines the latest board in the website database, the two prior boards in the compact protected archive, and older boards from R2. For Weekly boards, it combines the latest board in the website database with older boards from R2. Free users see only boards that are at least 3 exchange sessions behind the latest completed board. Selecting a Signal Desk date loads that exact stored board. It does not substitute a different date. The archive keeps the ranked names for all saved US and Hong Kong Daily and Weekly board dates that the viewer can access.
1. Top Signal Archive
The Top Signal Archive saves the final top 10 signals for each market, Daily-or-Weekly timeframe, pattern, and board date when the website files are published. Strict and Loose signals compete together. If fewer than 10 valid signals exist, it saves the smaller number. The newest saved list is in a small website file, the two prior Daily dates are in the protected archive, and older dates are in R2. A reader request loads these exact saved rows. It does not load 80 candidates or rank the list again.
2. Latest overall tape
The Signal Desk also keeps the latest raw scanner tape for every market and timeframe, limited to the top 100 rows per board. Its own pattern, market, Daily-or-Weekly timeframe, and quality filters narrow the current feed without changing the archive filters. Older non-winning tape rows are not retained by this view.
Workspace Tool
Top Trades
Top Trades is the ranking workspace. It converts scanner evidence into symbol-plus-direction trade candidates, then sorts those candidates by rule-based score or AI-adjusted score.
How A Ranked Candidate Is Built
1. Source evidence
Candidate construction starts from scanner-detected daily and weekly pattern signals.
2. Candidate grouping
Signals are grouped into one symbol plus one direction, such as US AAPL buy or HK 0700 sell.
3. Rank inputs
The score rewards pattern quality, daily plus weekly agreement, multiple same-side pattern families, strict labels, and eligible backtest contribution. Daily and Weekly freshness are admission rules and add no rank points.
4. Conflicts and removal
Stale or broken evidence is removed. Active opposite-side evidence stays visible as conflict and reduces the composite score.
5. Trendline rules
Candidate anchors are tested oldest-to-newest inside bounded timeframe horizons. Daily requires a three-month confirmed-anchor span and normally caps the oldest anchor at 18 months; a currently interacting or freshly broken 18-to-24-month line is used only when no standard Daily line qualifies, and 24 months is the absolute Daily maximum. Weekly requires a 12-month confirmed-anchor span and normally caps the oldest anchor at 10 years; a currently interacting or freshly broken 10-to-15-year line is used only when no standard Weekly line qualifies, and nothing older than 15 years is eligible. At least two independent swing-wick anchors are required, and a third or later contact strengthens the score without being mandatory. Total displacement must be at least 4% Daily or 5% Weekly; flatter structures are horizontal levels. Near-reactions use an approximately one-ATR recognition band capped at 1.0% Daily or 1.5% Weekly, while the exact line may not cut through any intervening or later candle wick. A current interaction or fresh held break is required. Scanner evidence may include the break period and age one only; backtests retain their separate next-period hold confirmation but use the same age-two expiry; farther intact standard-horizon lines are chart context only. Multi-ticker Top Charts cards show at most 220 Daily or 130 Weekly candles and project the exact scored line into that recent-detail window, with the true first and newest confirmed anchor dates shown when the origin is clipped. The linked Full Anchor Chart starts on the oldest retained validated anchor and keeps every later bar through the current interaction without unrelated pre-anchor candles. The compact crop is display-only. Exact endpoint geometry is stored at sufficient precision to reconstruct the validated wick boundary. Only geometry that satisfies today's scanner rules may be published as a signal—an unrelated fallback line cannot rescue it.
6. Double/triple reversal-zone rules
ATR-adjusted zones group prominent independent touches, while selected pivot wicks define the lower support edge or upper resistance edge. Every active full candle range must clear that outer boundary; a crossing forces refit or rejection unless it is the first candle of a separately labelled fresh, held failed-zone break. Break confirmation requires two consecutive closes or one decisive volume-backed close, followed by no close back through the exact break level. Body-backed necklines, two balanced reaction legs for triples, and volume-neutral scoring when volume is unavailable remain in force.
7. Candlestick rules
Hammers, Shooting Stars, and engulfings require ATR-material candles, directional-pressure gates, meaningful body and close-location checks, symmetric long/short scoring, recent-extreme zones for single-bar rejections, and optional volume confirmation that is neutral when unavailable. Every candidate needs at least three measured pre-pattern trend bars. Strict requires the full 8-bar daily or 6-bar weekly trend window plus 20 prior bars for ATR20; intermediate qualifying histories are capped at Loose.
8. Moving-average rules
The 100- and 200-period averages qualify independently with full integrity history, ATR-scaled interaction and overshoot limits, multi-bar selected-average slope, and mutually exclusive near, touch, bounce/rejection, or fresh reclaim states.
9. Channel/range-box rules
Channels and range boxes use one consistent paired geometry, ATR-adjusted tolerances, independently spaced reactive touches, exclusion of dual-edge candles from both boundary touch tallies, pre-break containment, held breaks only after the final anchor, and unbiased rejection of candles near both box edges.
10. Why some funds are intentionally absent
This is an intentional product decision—not missing data or a scanner error. Cash-like and mechanically accruing funds are omitted from Trendlines and ascending/descending Channels because their normal drift adds little trend-discovery value and can crowd out more informative equity, rate, and credit moves. Trendlines and ascending/descending channels are removed before scanner ranking, downstream score composition, and Top-10 selection whenever the instrument has a structurally carry-driven mandate. The classifier combines registered fund names and primary business descriptions with a small audited override list for SHY, BOXX, TBIL, TFLO, STIP, RAVI, and FLRN. Flexible mandate parsing also catches future equivalents: Treasury-bill, box-spread, cash-management, money-market, ultra-short, short-maturity, short-duration, short-duration TIPS, low-risk floating- or variable-rate, AAA CLO, and near-maturity fixed-income funds. An excluded instrument remains excluded during a volatility spike; recent price behaviour is not an exception. Ordinary equities, intermediate- and long-duration bonds, broad TIPS and municipal funds, and credit-sensitive high-yield or loan funds remain eligible, so HYG, SPHY, BIV, TIP, IEF, and CMF are not removed merely for being fixed income. Descriptions that mention cash or bills only as collateral for an equity, option, buffer, protection, or autocallable strategy do not qualify, and leveraged or inverse products remain eligible for their own economically informative moves. Range boxes are not subject to this filter. Filtering occurs before chart places are filled. A separate display-only diversity pass walks the ranked candidate pool for each Trendlines or ascending/descending Channels list. When two eligible funds have at least 60 aligned daily returns, correlation of at least 0.975 across the trailing 126 sessions and at least 0.970 across the recent 63 sessions, and volatility within 1.5 times each other in both windows, only the higher-ranked representative receives a list place. All comparisons stop at the displayed board date; ordinary equities are never merged, and funds with insufficient history or materially different risk remain separate. The next distinct qualified asset fills the open place. This diversity pass does not remove either fund's detection, change its score, composite evidence, AI evidence, journal or backtest status, or hide its Research page. An eligible asset can therefore be absent from one Top Charts list without being excluded from the scanner. An excluded directional structure creates no scanner rank, composite input, AI pattern evidence, journal admission, or new current-rule backtest trade. Stored prices, saved hits, and historical report artifacts are not deleted or rewritten.
Candidate Evidence Rules
These rules decide whether scanner evidence can still support a current Top Trades candidate.
1. Daily evidence age
Every daily pattern used by Top Trades—including candles, flags, channels, trendlines, tops/bottoms, ranges, and moving-average setups—must be detected on the latest completed market session.
2. Broken and stale setup removal
Trendline and top/bottom evidence is removed when any later candle, including its wick, crosses the active support or resistance boundary. A separately labelled, still-held trendline break remains eligible on the break period and the next completed period only; at age two or more it is removed. Top/bottom breaks keep their separate five-bar Daily or three-bar Weekly freshness window. Re-entry, expiry, or any older breach removes the evidence. Channel, flag, and moving-average evidence follows its stated invalidation rule.
2a. Ordinary carry-structure removal
A trendline or directional channel removed by the shared structural-trend eligibility filter is excluded before downstream score composition. It therefore cannot become primary or confirming evidence, add trend-following context, influence the AI verdict, enter either trade journal, or contribute a new current-rule backtest trade. Stored prices and historical report artifacts are not deleted or rewritten.
3. Opposite-side conflict
Active evidence on the opposite side is kept visible as a conflict and subtracts from the composite score.
4. Primary timeframe split
Daily and weekly Top Trades tabs are separated by the candidate's primary signal timeframe, not by every supporting signal.
5. Historical board date
When a Board Date is selected under Top Trades, the desk loads only that exact board date. It never uses an older board for an exact selection. If more than one report version has that board date, a complete exact-source repair wins over an incomplete or mixed-date original. A repair with evidence that stops on the exact board date wins over a repair that also contains later evidence. An exact-date historical repair combines US and HK only when original exports exist for both markets on that date. If only one market has an original export, the repair is market-only and does not carry the other market forward or backward. Original reports remain stored. For Top AI, the desk first selects a complete reviewed version for that exact timeframe board date. The version must render exactly five reviewed trades and all five must have a fundamental review. If either set is short, the exact board shows AI review unavailable and no partial AI rows. Reviews stay linked to their own report rows and are never attached to a different repair version. The desk reproduces the selected report's ranked rows and linked stored AI reviews; current scanner freshness, geometry, and structural-trend eligibility rules are not applied retroactively. Each market's price and chart inputs stop at that market's own saved source date. An original stored AI review is never replaced. If an archived report has no complete AI review, an owner-approved historical repair can apply the current Codex review only to the frozen report rows and date-truncated charts. Daily fundamental sources must stop at that market's saved Daily source date. Weekly fundamental sources must stop at the exact Weekly board date. A repair never creates a trade-journal entry. The date selector indexes the full stored report archive. If that exact board does not exist, the desk shows that it is unavailable. A grey date has a hover message when the selected market was closed, so the calendar explains the closure instead of showing only an unavailable date.
Top Composite Score Components
This applies to Top Composite, Top Liquid, Top Trend Following, Top Contrarian, Top Buy, Top Sell, Top US, and Top HK tabs. The composite score is rule-based, capped at 100, and is not an AI opinion or target return. Top Liquid ranks candidates using 70% composite score and 30% liquidity percentile within each market and timeframe.
| Component | Limit | Rule |
|---|---|---|
| Primary signal | Up to 60 | The chosen detector score contributes 0.62 points per score point, capped at 60, on the same scale for Daily and Weekly evidence. Weekly importance remains in evidence ordering and cross-timeframe confirmation; it is not multiplied again inside the capped primary component. |
| Confirmation | Up to 28 | Adds credit for extra aligned signals, multiple pattern families, daily+weekly agreement, and strict-quality labels. |
| Freshness | Admission only | Daily evidence must come from the latest completed session and Weekly evidence from the latest completed week. Freshness adds no score points; the retained compatibility field remains zero in stored score breakdowns. |
| Hammer/Shooting Star | Up to 4 | Adds reversal weight when hammer or shooting-star evidence is present, with extra credit for weekly or strict evidence. |
| Backtest contribution | -14 to +18 | Uses eligible daily candle, engulfing, and confirmed-flag signals generated by today's scanner rules, with at least 30 matured five-session outcomes, scored from 1D, 3D, and 5D average returns and win rates. Weekly studies, moving averages, and structural studies remain display-only until separately calibrated. MA100/MA200 and setup states remain separate. |
| Conflict penalty | Up to -22 | Subtracts points when the same symbol has active opposite-direction evidence. |
Top Trades Tab Filters
These tabs start from the same current scored candidate universe. Most rank by composite score; Top Liquid and Top AI use the alternative rules stated below.
Top Composite Trades By Score
All scored candidates are ranked by composite score and separated by Daily or Weekly primary timeframe. The board displays the top five for the selected timeframe. Every applicable category must contain exactly five Daily trades and five Weekly trades before it can be saved or published. A confirmed fully closed regional market has zero rows and is labelled Market closed. A short, oversized, mixed-limit, or unconfirmed closed-market category fails with its category, timeframe, expected count, actual count, and cause. An exact-date historical reconstruction is rejected rather than padded with an ineligible trade. The same check applies to normal generation and repair.
Top Liquid Trades
Ranks one best direction per symbol using 70% composite score plus 30% liquidity percentile. Liquidity percentile is calculated separately within each market and primary timeframe; composite score and raw 20-day traded value remain visible.
Top Trend Following Trades
Continuation setups from flags, trendlines, channels, and moving-average support/resistance. Reversal patterns can move here if they also have same-side trend context or breakout/breakdown state.
Top Contrarian Trades
Reversal setups from hammers, shooting stars, engulfings, and double/triple tops or bottoms when they do not have continuation context.
Top Buy / Top Sell Trades
Same composite universe filtered by long or short side.
Top US / Top HK Trades
Same composite universe filtered by market. HK can publish after HK coverage is ready; global views wait for the required combined universe.
Top AI Recommended Trades
AI Desk layer over the current composite report. The chart-review AI adjusts the score after viewing daily and weekly evidence; the application then enforces the verdict-specific adjustment range. The computed trade side and setup state are authoritative action semantics and take precedence over the pattern family name: for example, a held breakout through a Double Top can be Long, while broken Rising Support can be Short. The AI must assess whether the chart supports that supplied state and side rather than reversing the trade from its label alone. Each scheduled report has a fixed, deterministic review set. A completed AI board must have exactly five fully reviewed trades for each applicable timeframe; a short or oversized final board fails publication and records the expected and actual count. An exact archived board also stays empty and shows AI review unavailable until all five displayed trades have linked fundamental reviews. The configured owner is either the built-in API batch or checkpointed Codex scheduled passes. Codex technical and fundamental lanes use GPT-5.6 Terra with high reasoning. Codex mode never calls the external LLM API for scheduled chart or fundamental reviews, while on-demand AI Ticker Review keeps its existing API path. Time limits and individual failures leave a checkpoint: the next fixed scheduled run or locally checkpoint-triggered recovery skips stored reviews and resumes only the missing technical and fundamental reviews. A current local checkpoint never opens production storage, and a non-failing incomplete AI stage waits at least one hour before another recovery attempt. After the scanner, backtest, archive, and snapshot checks pass, the verified Signal Desk and rule-based Top Trades boards publish without waiting for Top AI. The exact Top AI board stays empty and shows that its review is in progress. A second checked snapshot publishes Top AI only after every configured technical and fundamental review is present. A checkpointed recovery selects one exact archived composite generation through its indexed timestamp instead of reading the full report archive; its technical charts, fundamental shortlist, source cutoffs, completion checks, and journal-admission policy remain bound to that frozen report even when a newer report is the public latest. Historical repair uses only chart data available by that report. Daily reviews use each market's Daily board date as the source cutoff. Weekly reviews use the exact Weekly board date. Every cited source must have a clear publication date no later than its cutoff. Undated or later saved fundamental rows do not count as complete coverage. Historical repair never creates retrospective journal entries. Model changes are not promoted from aggregate verdict counts: the opt-in shadow-calibration command compares two models on the same candidates, prompt, and chart payload and writes a calibration artifact without publishing reviews or creating journal entries. A separate versioned outcome dataset is rebuilt after scheduled coverage completes; it deduplicates actual historical reviews, enters at the first market open after each saved chart cutoff, measures anomaly-filtered 1D/3D/5D/10D/21D returns and a fixed 22-session MFE/MAE window, and never simulates target fills or creates journal trades. Scheduled board AI does not use user quota.
AI Verdict And Score Adjustment
New AI chart reviews use these non-overlapping enforced ranges. The raw model response is retained for audit, while the displayed and operational adjustment is constrained to the range matching its verdict.
| Verdict | Allowed Adjustment | Meaning |
|---|---|---|
| Strong | +3 to +10 | Clear chart support with coherent structure, levels, and trade plan. |
| Acceptable | -2 to +2 | Tradable but not exceptional; neutral or only modest score movement. |
| Weak | -10 to -3 | Setup exists but is messy, late, crowded, extended, or unclear. |
| Reject | -20 to -11 | Chart structure contradicts the candidate or invalidates its main rationale. |
Workspace Tool
AI Ticker Review
AI Ticker Review is the on-demand single-ticker workspace. It is different from the scheduled Top AI Recommended Trades board.
- Inputs: market and ticker; the system builds daily and weekly chart context for that symbol.
- History source: the page uses the read-only market snapshot when it has at least 30 valid Daily bars. It does not call the provider only because the snapshot is behind. If the ticker is missing or has too little history, provider bars are held in memory for that request only and are never written to the market snapshot or main market database.
- Data cutoff: unfinished current-session candles are removed. The result shows the exact completed-data cutoff used by the charts, scanner checks, and AI, with a warning when it is older than the latest expected completed market session.
- Scanner context: hammers, engulfings, trendlines, flags, channels, double/triple tops and bottoms, moving averages, daily evidence, and weekly evidence where available.
- Two score modes are kept distinct. When a current scanner setup exists, its composite score remains the base and the verdict applies the normal bounded AI adjustment. When no current Daily or Weekly scanner setup exists, there is no zero placeholder and no automatic Buy direction: the AI independently selects Buy, Sell, or No Trade and assigns a standalone 0-100 chart-tradeability score.
- Standalone AI chart scores must agree with the verdict bands: Strong 80-100, Acceptable 60-79, Weak 40-59, and Reject 0-39. No Trade is capped at 59 and normally has no entry, stop, or target. A standalone score measures the tradeability of the reviewed chart direction, not company quality, and never enters scanner ranking or backtest evidence.
- Direction semantics: the computed trade side and setup state override any usual directional implication of the pattern name; the AI checks that supplied interpretation against the chart and never flips the trade from the label alone.
- Output: the AI evaluates the computed or user-selected buy/sell side and returns a verdict, thesis, levels, observations, and risks.
- AI boundary: this review is technical and chart-only. It uses GPT-5.6 Luna with medium reasoning. The scheduled Top AI model settings do not change.
- Account boundary: the manual review does not create a trade-journal entry. My Tracked Trades remains writable through the separate account store.
- Quota: on-demand reviews count against the user plan quota; scheduled Top AI board reviews do not.
Workspace Tool
Trade Journal
Trade Journal is the tracking workspace. It records selected trade plans and outcomes; it does not feed back into scanner ranking.
- Stores selected AI-reviewed trades with thesis, entry, stop, target, status, notes, and later outcome fields.
- The Overall view provides matching summary panels and cumulative journal P/L charts by trade source, Daily versus Weekly timeframe, and US versus HK market. The AI Trades and My Tracked Trades tabs isolate either source while retaining all statuses. Overall and both source tabs order open positions first, planned entries next, and closed history last. Every journal view places a compact latest-five table directly below its charts, scoped to that active category. Only open or closed trades with an executable entry contribute to charts; planned trades wait until entry.
- Automatic journal entry requires an Acceptable or Strong AI verdict, adjusted score of at least 95, and a valid original AI-supplied entry, stop, target, and holding timeframe. Measured from the conservative edge of the entry zone, the original TP1 must provide at least 1R and the original final target must provide at least 2R; a missing or weaker target is not repaired to gain admission.
- AI trade plans should prefer visible chart structure when available: long stops below meaningful support and targets before meaningful resistance; short stops above meaningful resistance and targets before meaningful support. If no credible support/resistance level exists, a normal volatility/risk-defined plan is acceptable.
- After a trade has been selected manually, journal normalization may still supply risk-defined 1R/2R tracking targets when needed. This convenience is separate from automatic admission and cannot make an initially poor AI plan eligible.
- Trade State combines the lifecycle status, such as planned, open, partially closed, or closed, with target and stop outcome events.
- Journal entries do not change scanner scores or Top Trades ranks.
- User-added tracked trades save the exact click time in the relevant market timezone and enter at the first stored regular-market open after it: the same-day open only when saved before the 9:30 a.m. market-local open, otherwise the next stored trading-day open. An older signal date never causes an earlier entry.
- My Tracked Trades can be deleted individually or cleared together from the Trade Journal.
- Outcome fields track realized return and 1D/3D/5D/10D/21D follow-through using the same next-open daily-session convention as the backtester: 1D is the entry session close, 3D is the third eligible session close, and so on. Maximum favorable/adverse excursion uses a separate fixed 22-session diagnostic window. The journal stores the observed bar count, end date, and completion flag so immature excursion samples are not mixed into calibration.
- When a trade has two profit targets, the first target is treated as a partial/progress target: the trade remains in the open bucket, displays as partially closed, and closes only at the final target or stop.
- Journal P/L and mark-to-market displays include an estimated 0.10% transaction cost per side. Partially closed rows assume 50% was realized at the first target and show open MTM only for the remaining 50%.
Workspace Tool
Backtests
Backtests measures what happened after past signals that satisfy today’s scanner rules. Eligible results also provide the calibrated contribution used by Top Trades scoring.
- Each backtest signal is formed only with data available through the signal candle close; the simulated entry is the next stored market-session open.
- Provider OHLC is never repaired or invented for a backtest. Relative-shock continuity breaks split each ticker into intact clean segments: no signal, entry, liquidity window, Weekly aggregate, horizon return, or AI calibration outcome may cross a flagged Daily candle, and a saved hit must still match the currently stored signal OHLC.
- Daily Backtest Evidence reports 1D, 3D, 5D, and 10D exits at the first, third, fifth, and tenth trading-session close beginning with the entry session. Weekly evidence instead reports 1W, 3W, 5W, and 10W exits at completed weekly closes beginning with the entry week; holiday-shortened weeks use their final open session and an unfinished week never counts as matured.
- Returns are side-adjusted, so positive means favorable for both Buy and Sell studies. Results are gross of fees and slippage, and each horizon shows its own matured sample size.
- Backtest Evidence includes only signals generated by the scanner rules used today. Results from retired detector rules are excluded and are never combined with the displayed samples or Top Trades scoring.
- Backtest rebuilds fail closed before replacing published evidence: every exported and staged row must carry the current backtest-rule stamp, while missing or retired stamps and old rebuild checkpoints are rejected.
- Hammer, shooting-star, and engulfing studies use the same candle rules as today's scanner. Flag evidence requires a confirmed breakout or breakdown and counts a repeated formation once; forming flag consolidations are not included.
- Moving-average evidence tests MA100 and MA200 independently and publishes actionable support bounces/reclaims and resistance rejections/reclaims separately; observational near/touch states are not headline trades. Repeated same-period/state detections count as one continuous episode until there has been a gap of at least five daily signal bars or two weekly signal bars.
- Trendline evidence replays every signal using only history available through that date and applies the same bounded horizons as the scanner: Daily uses a three-month minimum confirmed-anchor span and 18-month standard maximum, with an interacting 18-to-24-month fallback only when no standard Daily line qualifies; Weekly uses a 12-month minimum span and 10-year standard maximum, with an interacting 10-to-15-year fallback only when no standard Weekly line qualifies. It requires at least two independent confirmed wick anchors, material 4% Daily or 5% Weekly displacement, exact wick clearance between and after anchors, and a current interaction or fresh held break. Backtest breaks require the next completed period to confirm that the break remains held, are eligible at age one only, and are excluded from age two onward. Extra contacts and reactions strengthen the score but are not prerequisites for a Loose line. The approximately one-ATR contact band, capped at 1.0% Daily or 1.5% Weekly, counts nearby confirmations without moving the exact boundary. Horizontal levels, crossed lines, re-entered breaks, stale breaks, distant chart-only context, and chart-only terminal refits are excluded from ranks and backtests. Channels and range boxes separate support/resistance holds from held upper/lower breaks, so the trade side follows the state rather than the pattern name alone.
- Double/triple tops and bottoms require full-candle boundary integrity and separate zone rejections, confirmed neckline breaks, and fresh held failed-zone breaks. Structural results count a saved formation/state once so persistent detections do not inflate the sample.
- Rows remain separate by pattern, setup state, side, Strict/Loose quality, timeframe, and moving-average period so unlike setups are never pooled. Symbol-level evidence cards are withheld when the available aggregate cannot preserve those boundaries.
- Top Trades backtest contribution remains based on eligible daily candle, engulfing, and confirmed-flag 1D, 3D, and 5D evidence with at least 30 matured five-session outcomes. Weekly-period evidence, moving averages, structural families, and 10D/10W results are display evidence and do not alter ranking until separately calibrated.
- Backtests use the currently stored symbol universe, so survivorship bias may remain. A positive result adds research context and does not make a trade automatic.