QUANTSMART 1.2.0-rc.7 — synchronized source, portable build and documentation

QUANTSMART is a local quantitative research preview for building modules, composing Strategy Cards and BEASTs, and inspecting reproducible experiments.

This release synchronizes the public source, Windows portable package and documentation at commit 7179c5c.

• Includes the completed-candle availability fixes for Atlas, Conductor routing and Replay pivots. Recompute older affected Atlas/routing reports.
• Restores truncated README text, uses one canonical tutorial index and adds project status and remaining work.
• Includes the timing correction, status page and tutorial index in the offline package.
• Keeps the default address http://localhost:5090/. A busy port causes startup to stop; no alternate port is chosen automatically.
• Packages no account credentials, private workspace or market history. Demo execution defaults remain off.

Windows: download QUANTSMART-1.2.0-rc.7-windows-x64.zip, verify SHA256SUMS.txt, extract the entire archive into a writable folder and open start.bat. Historical mode works without installed Node/npm/Python. MT5 requires your Windows terminal and Python 3.11+ 64-bit.

Upgrading: stop the existing launcher before starting a new copy. Preserve local configuration, saved Cards, imports and workspace data. They are not included in this release.

Verification: production build and the public regression/runtime lane passed; 70 bridge tests and 2 streaming-import tests passed. The Windows ZIP was extracted into a path containing spaces and Unicode and started through start.bat at port 5090 with host Node/npm/Python absent from PATH. Package file hashes and bounded private-path/credential-pattern checks passed. Public installation CI passed on Windows, macOS and Ubuntu for the release changes.

These checks establish software behavior within their stated scope, not profitable strategy edge. Issues #4, #25 and #26 remain open. Native automatically generated Demo entry/exit acceptance and predictive qualification are unfinished. Research remains SHADOW_ONLY / NO_DECISION; the separately enabled adapter supports Demo accounts only. This is a prerelease, not a real-account trading release.

QUANTSMART 1.2.0-rc.6 — Canvas cleanup and accurate research labels

This preview removes the two empty-lane warning banners in Canvas L5/L6 while preserving validation checks.

Research reports now derive path and cost labels from their execution receipt. Cost-fragility text names the failing scenario instead of incorrectly claiming that only LOW cost succeeds. Playbook has a clear missing-collection state and a working route to CSV research.

Verification: production build, full public lane, four new report-label regressions, 28 bridge tests, and Windows portable start.bat at the default port 5090 passed locally. The live browser opened all 17 workspaces, previewed Custom L1–L6 on saved CSV, ran a 5,000-bar H1 Card and reopened its immutable evidence. Responsive checks covered effective widths from 521 to 1920 CSS pixels with no document-level horizontal overflow.

Scope: 17-page coverage is a smoke test, not exhaustive verification of every control. No MT5 terminal was running on the test machine, so no live account/tick-stream acceptance is claimed. No broker orders were sent. Market history and credentials are not bundled.

Extract the entire Windows ZIP and run start.bat. Default URL: http://localhost:5090/. Occupied ports stop startup rather than selecting a different port. See SHA256SUMS.txt for download checksums.

Fixes: #23 and #24. Source: 391b6a0

QUANTSMART 1.2.0-rc.5 · decision evidence

This preview adds optional, immutable decision/outcome evidence in Microscope and records the assessment of the owner's Drive integration handoff.

Decision and outcome evidence

Select a saved CSV and Card in Microscope, open Decision & outcome evidence, then build the journal. One optional canonical simulation joins observed signals with policy blocks, modeled fills and completed trades. Saved reports reopen after restart and include source, definition, configuration and engine hashes. No-entry and still-open observations keep null outcomes rather than zero-R labels. Content-addressed storage is idempotent and validates integrity on read.

The journal covers 61–5,000 selected design bars with equal design/execution timeframes. Observation IDs are not canonical persistent structural-setup IDs. Silent pre-signal veto/abstain branches remain unobserved. Excursions use complete design bars before the exit bar and are lower bounds in price units; exact excursion R is unavailable.

Signature record correction

The legacy entry-record helper incorrectly relabeled net R as MFE/MAE and read the entry candle's high/low/close at an entry-open decision. It now uses the completed signal bar, preserves canonical net_R and returns nullable mfe_R/mae_R. Consumers of that helper must accept null excursion values.

Drive assessment

Read the full assessment. The 28 linked objects were reviewed as source evidence. Raw Drive Python, credentials, account state, runtime policy and PID controls were not imported. The lifecycle source is primarily position governance; MAM persistent tracking is separate from its clustering engine; the Adaptive impact report is proxy-only; later QI shadow wiring exists despite the old P2 report. No calibrated-probability or independent-outcome advantage was established for those engines, so MAM, Adaptive and QI activation remains deferred.

Verification

Production build, public regression lane and 28 bridge tests passed locally. The built HTTP service verified deterministic replay, one simulator call per journal request, idempotence, restart/reopen, corruption refusal, missing outcomes and unchanged canonical trades/costs. The installed preview joined 459 observations to 305 closed modeled trades on an existing 5,000-row XAUUSD H1 research import, with no unjoined trades and no Card-library mutation. The research tape's broker provenance is unverified; these counts demonstrate wiring, not an edge.

The final Windows ZIP was launched at default port 5090 in a Unicode/spaces path with system Node/npm/Python absent from PATH, without build/install/Vite. New-panel TypeScript, production build and server-rendered controls were checked; fresh browser visual acceptance was unavailable and is not claimed.

Install

Extract the Windows x64 ZIP fully and double-click start.bat. URL: http://localhost:5090/. Startup stops if the port is occupied; it never chooses an alternate port. No market history, private workspace or credentials are bundled.

17 existing video tutorials remain available with English narration and Vietnamese captions. They do not claim to film this new journal panel.

Research only: SHADOW_ONLY / NO_DECISION / live_apply_allowed=false. The canonical simulator and MT5 order paths were not changed. Independent human-label verification remains open in #4. Issues #21 and #22 track the record fix and evidence integration.

Exact-source public installation checks passed on Windows, Ubuntu and macOS: https://github.com/Jokejoker-designer/QUANTSMART/actions/runs/41339511091

QUANTSMART 1.2.0-rc.4 · custom layers and BEAST workflows

Superseded by v1.2.0-rc.5: https://github.com/Jokejoker-designer/QUANTSMART/releases/tag/v1.2.0-rc.5

QUANTSMART 1.2.0 release candidate 4 repairs the CSV-backed custom-authoring workflow and preserves BEAST operating context.

• Custom L1–L6 studios share an explicit saved CSV and evaluation timeframe. L1 output, replay and role checks use the same interval. L4/L5 individual-bar inspectors use their parent dataset.
• Expanded template and JSON-import controls remain in their scrollable columns. Library-only saves no longer attach modules to the current draft.
• Portable definitions are validated before mutation. Missing inputs, malformed graphs/ASTs, conflicting IDs and unsupported anchors are refused. Untrusted performance claims are reset. Failed durable saves restore local state.
• Rule previews share canonical RSI/ATR calculations. L5 sample distributions exclude warmup consistently. CSV results include the last closed observation and use source-correct labels.
• Custom L6 reference zones appear on the main chart. Evidence series are computed once per preview rather than once per candle.
• BEAST member Cards must share symbol/design/execution context. The CSV role analysis and exported Card preserve that context. Mixed or unavailable execution data is refused.

Downloads

Extract the Windows x64 ZIP completely and double-click start.bat. Historical mode requires no separately installed Node, npm or Python. Default URL: http://localhost:5090/. An occupied port stops startup; the launcher does not choose another port. No market history, private workspace or credentials are bundled.

Tutorials

17 recorded lessons on Drive: English synthetic narration, Vietnamese captions, plus an offline ZIP, subtitle files and example definitions. Drive access remains owner-controlled. Lesson links and scope.

Verification

Local production build, public regression lane and 28 bridge tests passed. The packaged start.bat was tested at port 5090 in a path containing Unicode and spaces, with system Node/npm/Python absent from PATH and no install/build/Vite step. Runtime tests cover import/export/reopen, real HTTP persistence, restart, invalid definitions and rollback after HTTP 503/quota errors. All six layers were also checked on the same 2,000-row tail of an authorized 44,640-row exchange CSV.

Actual browser recordings cover CSV file-chooser import, chart opening, per-layer prompt/JSON import, Card runs, matched-control refusal/results and BEAST composition/role analysis/export. The final additional BEAST-to-Microscope browser recording was unavailable. Its exported H1/H1 context and actual installed-worker Card run were verified separately, without claiming filmed evidence.

Detailed verification and limitations. Fixes #16, #17, #18, #19 and #20. CSV browser acceptance for #12 was exercised. The independent human-label benchmark in #4 remains open.

Research only: SHADOW_ONLY / NO_DECISION / live_apply_allowed=false. OHLC simulation does not prove tick fills, live readiness, an independent holdout or profitability. MT5 order paths were not changed.

Public installation CI passed on Windows, Ubuntu and macOS for this exact source commit: https://github.com/Jokejoker-designer/QUANTSMART/actions/runs/41337398821

QUANTSMART 1.2.0 preview — Default port 5090

Superseded by v1.2.0-rc.4: https://github.com/Jokejoker-designer/QUANTSMART/releases/tag/v1.2.0-rc.4

Default-port contract update: fresh source setup, .env.example, the standalone production server, Vite development mode, README, installation guide and portable start instructions now agree on http://localhost:5090/.

The launcher already refused occupied ports; the previous local 5193 preview was an explicit test override. This update fixes the old 5190 defaults. Occupied 5090 fails clearly and preserves the existing owner. No automatic alternate-port search occurs. A user-supplied --port or MFE_PORT remains an explicit advanced override. On an older installation, set MFE_PORT=5090 in .env to adopt this default.

Acceptance exercised the actual packaged start.bat in a Unicode/space path, with host Node/npm/Python absent from PATH and without --port. It served 5090, printed http://localhost:5090/, and recorded port 5090 in the launcher state. The documented hostname reached the same engine as the loopback health endpoint. A separate occupied-port test refused startup without announcing another ready URL and kept the unrelated listener alive.

The local preview was moved to 5090; the saved Card library and two imported datasets were preserved by hash/readback. Old 5093/5190/5193 endpoints are no longer serving this preview.

Default-port issue #15 · installation guide · CI.

This remains a preview: CSV manual browser acceptance (#12) and independent human rejection labels (#4) are separate pending items. No broker orders were sent. Research authority remains SHADOW_ONLY / NO_DECISION / live_apply_allowed=false.

Download the Windows x64 portable ZIP or source ZIP below. The portable contains verified Node.js and uses the production build. No market history, credentials or private workspace is bundled.

QUANTSMART 1.2.0 preview — CSV research and matched controls

Port update notice: use 1.2.0-rc.3 for the default http://localhost:5090/ URL. This older asset retains its 5190 default. Existing local .env values should be set to MFE_PORT=5090 to adopt the new default.

Preview of durable CSV research and bounded strategy attribution. Automated build/API checks passed on Windows, macOS and Linux. Manual browser file-chooser acceptance is pending because the browser tool reported Local browser use disabled; this preview does not claim that interaction was verified.

CSV now parses in a backend worker, validates numbers/calendar/OHLC strictly, preserves supported quoted/MT5/locale/epoch formats, reports duplicate/gap handling, saves content-hashed local datasets and reopens them after restart. No silent price repair or history truncation. The same imported source feeds chart aggregation and Card jobs; unsupported timeframe/asset contexts fail explicitly.

Execution fixes: final stop-floor-adjusted R:R is checked; unsupported ambiguous full-equity policies refuse instead of silently booking stop-first; incomplete/duplicate M1 paths refuse instead of time compression; canonical simulation refuses missing/invalid/mismatched asset specifications.

Matched controls provide A real+fixed hold, B matched random+fixed hold, C the same random+Card exits, D real+Card exits as isolated events, plus a separate actual full-system reference. The protocol matches supported eligibility/regime/direction/UTC hour/chronological third and exposes seed/event identities and exposure. Custom/stateful/mixed-timeframe Cards are refused. No independent OOS, p-value, DSR, exposure equivalence or profitability is implied.

Validation retained 44,640 actual BTCUSDT M1 records from the official Binance January 2025 archive and produced 744 H1 bars, after checking the provider checksum. Unknown BTCUSDT monetary specifications were refused. A separate existing local XAUUSD research file was exercised with explicitly unverified broker provenance. Neither dataset is bundled.

Import guide · validation and limits · CI. Issues #9–#14 track the changes; #12 remains open for final browser acceptance, and #4 remains open for independent human labels.

Source ZIP and Windows x64 portable ZIP include no market history, credentials or account data. Import your authorized files. Preserve your workspace/imports when updating. Research authority: SHADOW_ONLY / NO_DECISION / live_apply_allowed=false.

QUANTSMART 1.1.2 — Causal entry and execution corrections

CSV/execution follow-up notice: this version has the stop-floor, ambiguous-policy, M1-integrity and missing-specification defects documented in issues #9, #10, #11 and #14. The 1.2.0 preview contains corrections and automated checks; manual browser acceptance remains pending. Validation and scope. Original release assets are preserved.

This patch corrects three independently reproduced backtest defects. Previous results from 1.1.0 and 1.1.1 require reruns.

• #6: ATR and Auto R:R entry levels now use completed signal-time data, not the future fill candle high/low/close. Invalid target direction after a fill gap is rejected.
• #7: Closed execution signals fill at the next observed child open. The remaining child path is processed immediately, so same-design-candle stops/targets are not skipped. Pre-entry extremes and retroactive reentries are excluded. Missing suffix data fails explicitly.
• #8: OHLC/M1 facades cannot claim TICK_PATH_EXACT from hasTickBidAsk metadata. The truth builder refuses exact labels while no tick executor exists.

Model details are explicit: entry-time ATR is the provisional slippage scale; within-candle order remains ambiguous. Engine manifests use realistic-sim-1.1.2 independently of saved configuration strings.

Verification includes red/green tests against 05cdaeb, long/short mutation and chronology cases, actual M1 resampling, built-worker API truth checks, the existing measurement suite, 28 bridge tests and Windows portable startup/shutdown in a Unicode/space path without host Node/npm/Python.

In twelve fixed 5,000-bar local-history comparisons, all net results decrease after these corrections. For example, the H1 middle-window Trend Pullback result changes from +40.9286R to -23.9404R. This is an engineering comparison under provisional costs, not evidence of a profitable edge. See the exact aggregate table rather than extrapolating from one result.

Audit, observations and limits. Independent human-label benchmark remains open.

Download the Windows x64 portable ZIP or source ZIP. Import your own authorized data. No market history or credentials are bundled. MT5 is optional. No broker order path was changed or exercised. Research authority: SHADOW_ONLY / NO_DECISION / live_apply_allowed=false.

QUANTSMART 1.1.1 — Research engine corrections

Execution correction notice: this version contains next-open ATR/Auto R:R look-ahead, missed same-design-candle intrabar exits and metadata-based tick-exact promotion. Use 1.1.2 and rerun prior experiments. Reproduction and limits. Existing release assets are preserved.

Fixes reproduced quantitative-research defects in gap execution, final-bar exits and custom-module ablation. Adds a durable journal of observed search attempts and canonical simulator calls.

• Gap stops/targets use the first observed open when the level was skipped. Ambiguous OHLC paths remain explicitly modeled.
• Ablation removes executable custom definitions, pins coefficients and symbol, retains losses from sparse windows, and rejects an unsupported mixed-timeframe execution path.
• Search counts survive restart and cache pruning. Prior/external trials remain unknown; no multiple-testing correction is claimed.
• Dataset-free regression suite: 41,280 full-series versus prefix comparisons, gap/final-bar cases, ablation parity, journal persistence and built-worker integration.
• A read-only comparison across 12 existing local-history runs changed 9 results. This is regression evidence, not strategy qualification.

Read the audit and its limits. Issues #1, #2 and #3 track the confirmed research defects. Issue #4 remains open for independent human rejection labels.

Existing results must be rerun against the new engine. Version 1.1.0 uses the previous engine. The Windows x64 ZIP includes verified Node.js and launches offline in historical mode. Import your own authorized data; no market history or credentials are bundled. MT5 remains optional. This patch does not establish live-trading readiness or profitability.

Research authority: SHADOW_ONLY / NO_DECISION / live_apply_allowed=false.

QUANTSMART 1.1.0 — easier local installation

Execution correction notice: this version contains next-open ATR/Auto R:R look-ahead, missed same-design-candle intrabar exits and metadata-based tick-exact promotion. Use 1.1.2 and rerun prior experiments. Reproduction and limits. Existing release assets are preserved.

Windows x64 portable release: extract the entire ZIP, open start.bat and choose historical CSV research or optional MT5. Node.js and the production build are included. Historical use needs no installed Node/npm/Python.

Source setup now includes an installation doctor, cached production startup, first-run CSV import, public dataset-free test lanes and cross-platform CI. Optional MT5 setup requires Windows and Python 3.11+ 64-bit. No broker account is logged in by the installer, and manual Demo execution remains off by default.

Windows portable entry-point, Unicode/spaced extraction paths, real public server/worker tests and 28 isolated bridge/Demo tests were checked locally. See docs/INSTALLER_VALIDATION.md for scope and GitHub Actions for cross-platform results.

This is a portable local app, not a signed native installer. No native macOS/Linux binary or phone installation is supplied. No market histories, account credentials or personal workspaces are bundled. Verify downloads against SHA256SUMS.txt.

Installation guide: https://github.com/Jokejoker-designer/QUANTSMART/blob/e17448af4e1f4f93414b8ee512600b9e1bc2e634/docs/INSTALLATION.md

Cross-platform source CI passed on Windows, macOS and Ubuntu for this exact commit: https://github.com/Jokejoker-designer/QUANTSMART/actions/runs/41276817759