Live Parity Status: 2026-08-13

Branch and release state

• Working branch: fix/soak-live-parity.
• Current head: 5e9169b0 (fix(scripts): accept full qkt commit in image checks).
• dev and testing already contain the prior attestation CLI change at the testing SHA a7cca220e974c819da2d464c328b1c2417341cab.
• No paper-soak attestation has passed on the trusted GitHub runner yet, so promotion to main remains correctly blocked.

Sealed live evidence

Gateway: MT5 demo login 436804390, server Exness-MT5Trial9, gateway 0.3.10. The account was flat after every case and the final account query was flat.

Evidence root: /var/tmp/qkt-validation/parity-a7cca220-full-20260813T0602Z.

• Four generated cases ran: EMA EURUSD, RSI GBPUSD, ATR EURUSD, and case-math GBPUSD.
• All four read-only warmup/tick/bar cases passed.
• All four armed market-bracket cases passed with real 0.01-lot demo orders, strategy-owned closes, zero final positions/orders, and balanced account/deal deltas.
• Every case has checksum-verified evidence and a sealed result.json.
• Every replay comparison passed: full-tick order journals, bar order timing, initial and broker-adjusted protection, and MT5-simulation fill parity.
• The generated capability catalog includes all registered indicators and numeric functions (including EMA, RSI, ATR, Z-score, lag, correlation, percentile/rank, regression, volatility, session, and candle functions).

Tooling fix

run-container-round-trips.sh previously accepted only the short SHA form in qkt --version. The exact testing image emits the full SHA, so a valid image was rejected before starting containers. The runner now accepts either short or full commit identity and still fails closed on a mismatch. bash -n passes.

The first concurrent Insights attempt also correctly caught two scenarios with different starting balances. Fresh equal-baseline scenarios were prepared; the collector health and causal contract probe passed. The armed concurrent run must be regenerated against the hardened branch/image before it can be called a pass.

Concurrent Insights evidence

The follow-up harness fix was merged to dev as PR #996. The exact local build and matching container image for commit f75a089df5626087e0db7631de431f0c1c540cd9 completed a real two-strategy same-account run:

• Evidence root: /var/tmp/qkt-validation/insights-f75-live-full.
• MT5 demo account: 436804390 / Exness-MT5Trial9; gateway 0.3.10.
• EURUSD and GBPUSD deployed concurrently with magic 919600 and 919601.
• Distinct real MT5 tickets: 3081426457 and 3081426434.
• Each strategy produced two fills/two deal legs, strategy-owned close, and zero final positions/orders.
• The base round-trip passed ownership, M1/M5 stream/evaluation, indicator traces, bracket protection, audit and transport checks.
• Insights result passed: two retained instances, isolated strategy attribution, no cross-owner causal leakage, no sequence gaps/regressions, and both books flat. SQLite retained 20 strategy events plus 8 account-state events per instance.
• The freshness evidence records the quiet-market behavior. The final quote was fresh and all samples stayed below the 60-second diagnostic ceiling; occasional older samples remain visible in tick-freshness-gate.jsonl rather than being hidden.

Higher-timeframe evidence

The exact f75a089d local build completed the read-only higher-timeframe probe at /var/tmp/qkt-validation/htf-f75-095216. It passed closed, aligned, unique bar retrieval for M15, H1, and H4 using one-hour, one-day, and two-day warmup sizes; the account remained unchanged, with zero positions, orders, and venue deals.

The four-container capability catalog was started at /var/tmp/qkt-validation/catalog-f75-run. All four daemons reached healthy running state with zero dropped ticks and produced per-case health, evaluation, volume-rejection, latency, and runtime-log evidence. Its aggregate result.json was not sealed after cleanup, so this catalog run is intentionally incomplete and must be rerun before it is used for attestation.

Exact testing-gate attempt

Testing now points to 50f59fc06a8b5f9334800a00022cd0aebde05713, and the published edge image was verified as ghcr.io/elitekaycy/qkt@sha256:aee5fe526f50d47fb1203ad5e93b46a56fa294b81a858ff0ac1776d1fb0d22a5 with the matching OCI revision. A real two-symbol run against that image opened and closed both demo positions, but did not seal an aggregate result: the local gateway produced recovered stale-data windows and both live feeds later logged a reconnect wait. This run is retained as failed/incomplete evidence, not promoted into an attestation.

The paper-soak workflow was dispatched on the exact testing SHA with a trusted runner. It failed closed because /var/lib/qkt/soak/attestation.json was absent. No attestation was fabricated. Main promotion remains blocked until a complete exact-SHA bundle is generated and placed at that runner path.

Next execution

Follow-up evidence (2026-08-13)

The exact 3f1873873acd8d60e02ab45fe09ea30dcdc71537 build has now passed the following additional local checks against the single local Exness demo gateway (127.0.0.1:5001, account 436804390):

• The static generated four-case parity verifier passed for ema-eurusd, rsi-gbpusd, atr-eurusd, and case-gbpusd. The generated catalog contains 59 indicator capabilities and 15 numeric functions. This is a preparer and contract check; it is not a claim that every capability has already been exercised with a real order.
• MT5BrokerIntegrationTest and MT5PositionPollerCloseTest both passed in a focused Gradle run. This covers the protection-registration race, pending and partial fills, cancellation, ambiguous outcomes, recovery, and poller close attribution regressions.
• The isolated static risk matrix passed five causal rejection cases; the stateful matrix passed four restored-state halt cases. Both remained flat and produced zero transport mutations for rejected decisions.
• The controlled margin-floor fixture passed. It observed a real 0.01-lot demo opener, rejected the probe before transport while below the dynamic floor, accepted the same probe after headroom recovered, and flattened the recovered position. The account was flat after cleanup.
• The exact-build higher-timeframe warmup probe passed M15, H1, and H4 one-hour, one-day, and two-day requests with aligned unique closed bars and no account mutations.

The follow-up harness change fix(scripts): accept full qkt image revisions (9c8f13e0) is merged to dev as PR #1002. Dev CI for the current docs head 865ef81e770dd47182f1b92a85ab363b5932ce32 is green. testing and main still point at the prior promoted code (3f187387... and b478062b... respectively); any promotion of a newer SHA requires a newly built exact image and regenerated exact-SHA evidence.

Remaining notes-matrix gaps

The following are deliberately still open and must not be represented as attested: explicit live limit/stop-limit/stop-order fills, cancellation and partial-fill waves, timed exits and re-entry across every strategy family, cross-book portfolio attribution/isolation, and a complete live-vs-backtest comparison for every DSL/indicator/math capability. The catalog's prior live run also recorded a dropped tick during a gateway disconnect; this is retained as a feed-resilience observation to classify and rerun, not hidden as a pass.

Exact testing-SHA follow-up (2026-08-13)

The exact testing image was refreshed at 11d5793e4f0ff56a2f1da10918b7e539a17d4ef7 (ghcr.io/elitekaycy/qkt@sha256:c285abfb6ddc6748467285d450770936d9c199bd839d038c94858288a125bc08). The four-container read-only catalog ran locally against the demo gateway for the required 330-second window. Numeric/candle, cross-symbol/multi-timeframe, session/history, and volume-negative cases all emitted their expected traces and the account stayed flat. The run failed closed because the cross-symbol case reported one dropped tick after a gateway disconnect/recovery; no catalog result was sealed and it is not promotion evidence. The metric includes late pipeline ticks as well as ingress shedding, so this remains an explicit stress observation to rerun and classify, not a fabricated zero.

A fresh exact-SHA two-container armed wave then placed and closed real 0.01-lot EURUSD/GBPUSD demo positions and reconciled the account, but the runner failed closed on a false runtime error: the MT5 position poller observed the expected fill-anchored SL/TP normalization before the asynchronous broker callback had registered the protection change. This exposed a real ordering race. The fix is on the current feature branch: register expected protection before both sync and async venue requests and remove it on request failure. Focused MT5BrokerIntegrationTest and MT5PositionPollerCloseTest suites pass; the live wave must be rerun from the resulting promoted image before attestation.

1. Build/publish a QKT image for the hardened commit, or run the concurrent runner from a clean checkout at that exact commit.
2. Prepare two equal-baseline scenarios and run the armed shared-account Insights round-trip. Require both daemons to complete, isolated instance and strategy attribution in SQLite, and a flat account.
3. Run the read-only catalog and higher-timeframe warmup suites against the same exact commit/image.
4. Assemble canonical health, journal, reconciliation, coverage, parity, and insights artifacts from those sealed outputs; run qkt soak report and scripts/verify-paper-soak-attestation.py locally.
5. Obtain a successful exact-SHA paper-soak workflow on the trusted [self-hosted,qkt-paper-soak] runner, then promote dev -> testing -> main through the protected PR gates. Do not fabricate or bypass the attestation.

Safety

The live lock is account-scoped. Keep the gateway on the demo account, arm only with the explicit local approval token, verify flatness before and after each run, and never treat a partial runner directory as evidence.

Main promotion and follow-up audit (2026-08-13)

The hardened protection-ordering fix is now merged to main through promotion PR #1000 at merge commit b478062bb63a483bc1d019cd1b2530c8938276d1. Required build, integration, Docker/runtime-smoke, Windows packaging/install, GitGuardian, and exact-image paper-soak checks passed. Main post-merge integration run 31702307730 also passed.

Exact testing SHA 3f1873873acd8d60e02ab45fe09ea30dcdc71537 evidence:

• Armed EURUSD/GBPUSD live round trip plus full-ticks-paper, full-ticks-MT5, and bars-paper replay: /var/tmp/qkt-validation/insights-3f-live and /var/tmp/qkt-validation/final-3f-replay-{a,b}; both strategies opened and closed real 0.01-lot demo positions and the account finished flat.
• Insights causal round trip passed with two isolated instances, no sequence gaps/regressions, and no cross-owner attribution leakage.
• Restarted local-gateway read-only load soak passed at /var/tmp/qkt-validation/load-3f-122354-610042/evidence/result.json: 650 seconds, two containers, four symbols, eight 1m/5m streams, controlled restart, state restore, 0 dropped ticks, and zero gateway mutations.
• Paper-soak workflow 31701168665 passed using the immutable QKT digest sha256:f69102f813e79831e79a41f51223c3aae4e1b70c07bd749e8edac7981e097f4c.

The follow-up four-case catalog was rerun after the gateway restart. Numeric, cross/multi-timeframe, session/history, and volume-negative cases emitted their expected traces and remained financially read-only, but the aggregate run again failed closed because the numeric case recorded one dropped tick during a feed disconnect. This is not promotion evidence and remains an open gateway/feed resilience gap. The numeric trace itself covered EMA, SMA, WMA, DEMA, TEMA, HMA, RSI, standard deviation/variance, z-score, regression, percentile rank, skew, efficiency ratio, lag, MACD, Bollinger, extrema, math functions, ATR, Williams R, CCI, stochastic, Keltner, DI, and ADX. Evidence root: /var/tmp/qkt-validation/catalog-3f-125922-704966-live.

The broader notes matrix is therefore still incomplete: order-bearing coverage for limit/stop/cancel/partial-fill/risk-halt paths, higher-timeframe H1/H4 boundary parity, and portfolio/book isolation remain to be executed and sealed.

Final Promotion Record (2026-08-13)

The preceding sections are historical snapshots. The authoritative current release state is:

• dev is 7c400906453cf6d75d3de10601cdbf285b3c7a90 and testing is c818c0b44d7d4cf7070edc0bc9b22dea12a98a37. No dev commit is absent from testing; the Windows promotion-gate fix and all preceding source/evidence fixes are present in both.
• Exact testing image: ghcr.io/elitekaycy/qkt@sha256:9d21d5ec9009c16da94ac376afc5af1fccc221360ab09aac0226e0e59109a664, revision c818c0b44d7d4cf7070edc0bc9b22dea12a98a37.
• The exact-image armed EMA run passed against the single local Exness demo account 436804390 via 127.0.0.1:5001: real 0.01-lot entry and timed close, two accepted/fill events, one order post and one close post, zero risk rejections, and final flat positions/orders. One stale/recovery episode was recorded and retained as feed evidence.
• The exact-image read-only soak passed for 310 seconds with 31 health samples, 105 live ticks, 160 warmup ticks, 29 candles, 11 stream candles, zero dropped ticks, zero venue deals, unchanged flat account, and latency tracking enabled (tick-processing p99 3.73 ms, max 26.34 ms). M1 and M5 traces were observed.
• Local attestation verification passed for the exact SHA/image. Trusted paper-soak workflow 31738623043 passed and uploaded immutable evidence.
• Promotion PR #1006 passed build, integration, runtime-smoke, build-and-push, Windows packaging/install, and GitGuardian checks, then merged to main. Current main is 8ad1c109fc3231e1f0dab0e94e0b4d475a5b083a.
• Main post-merge docs, integration, and Docker runtime-image smoke workflows passed (31740355885, 31740355930, and 31740355830).

The local gateway required one restart after an MT5 IPC timeout and recovered to healthy/connected. This is an environment/session recovery observation, not a QKT code fix. Future soak reports must retain stale/recovery and restart evidence rather than relabeling the run as uninterrupted health.

Notes Matrix Follow-up Evidence (2026-08-13)

The first re-entry/risk-gate slice of the previously unsealed matrix is now sealed against the exact c818c0b4 image and the single local demo account:

• /var/tmp/qkt-validation/reentry-c818: live reentry passed with two EMA-driven entries and two timed exits, four accepted/fill events, two entry posts, two closes, zero risk rejections, and final flat state. Its golden replay passed at /var/tmp/qkt-validation/reentry-c818-replay.
• /var/tmp/qkt-validation/reentry-blocked-c818-2: live reentry_blocked_max_trades passed with one real entry/exit and one MaxTradesPerDay rejection before transport. It produced one entry post, one close, one risk rejection, and final flat state.
• Both runs used generated EMA 3/5 M1 plus 5-minute streams, warmup ticks, real MT5 fills, engine/transport journals, and final account reconciliation. Each recorded stale-feed episodes that recovered before completion; these remain gateway polling evidence and are not counted as continuously healthy.

Remaining unsealed notes-matrix work is explicit: live pending limit/stop and stop-limit trigger fills, cancellation/partial-fill waves, the other re-entry and risk-halt recovery variants, portfolio/book isolation, and complete live-vs-backtest coverage for every registered DSL/indicator/math capability.

Pending Order Boundary Evidence (2026-08-13)

The local QKT CLI was used against the single local Exness demo account with a unique magic and bounded 0.01-lot intents. Resting BUY_LIMIT and BUY_STOP orders were accepted by MT5, observed through the venue orders endpoint, cancelled by their native tickets, and left zero pending orders, zero positions, and unchanged balance/equity. A BUY_STOP_LIMIT request initially exposed a real gateway defect: the gateway rejected the stoplimit field as unknown before reaching MT5.

The gateway fix is isolated in mt5-gateway PR #81, merged into gateway dev as 17f5fe8f608b910fb1f486832e1c566ba74f060b after test and Docker CI passed. The fix accepts both native stop-limit order types, validates the limit price, and forwards stoplimit into the MT5 trade request. Focused gateway tests pass (14), and the rebuilt local image accepted a valid BUY_STOP_LIMIT (retcode 10009), reported native type 6, then cancelled it successfully. Final account state was flat with zero pending orders and unchanged balance 99992.61.

This proves placement and cancellation only. Stop-limit trigger-to-fill parity, partial-fill waves, released gateway image promotion, and QKT image refresh remain unsealed and must not be inferred from this evidence.

Limit Trigger Probe

/var/tmp/qkt-validation/pending-limit-trigger-c818-20260813T220239Z placed a BUY_LIMIT at 1.15305 while the quoted ask was 1.15306. MT5 accepted the pending order (ticket 3085335993), and the next poll observed the pending order gone and one open 0.01 position at the same price. QKT then closed it and reconciled the account to zero positions and zero pending orders. The gateway deal range filtered by magic 948500 contains the entry at 1.15305 and the exit at 1.15297, with net demo PnL -0.08 and zero commission/swap. This is live trigger evidence, but it is not yet a full QKT golden/replay parity bundle; the harness cleanup interruption and that replay comparison remain open work.

Current Exact-SHA Promotion and Four-Case Wave (2026-08-13)

The authoritative current release state supersedes older snapshots above:

• testing SHA e271822d4895cf16a662338c1f38090f77965485 was promoted through PR #1010 and merged to main at f275065947c9b967b85843ba1c12a8e16f90d587.
• The exact testing runtime image was ghcr.io/elitekaycy/qkt@sha256:1dbef063538c29e37a880e1544c3a07aad66d57c369d9377f86bbad1ca54ac23, whose OCI revision is the exact testing SHA. Main post-merge Docker, integration, and docs workflows were dispatched on the merge commit.
• Fresh exact-image live evidence was generated at /var/tmp/qkt-validation/exact-testing-e271822d-20260814-rerun2. The local gateway was 127.0.0.1:5001, MT5 demo login 436804390, and the account was flat before and after the wave. The first attempted run was interrupted by an external probe and is invalid evidence; only rerun2 is sealed.
• The four generated cases were EMA EURUSD, RSI GBPUSD, ATR EURUSD, and case-math GBPUSD. All read-only cases passed warmup, live ticks, M1/M5 bars, stale recovery, zero dropped ticks, journals, and flat reconciliation. ATR and case-math each produced a real 0.01-lot market entry and strategy-owned close; both had two live fills and passed full-ticks-paper, full-ticks-mt5, and bars-paper replay. Current result files identify QKT commit e271822d and status: passed.
• A fresh six-artifact live-parity attestation was generated and locally verified, then trusted workflow 31754382341 passed. The attestation is retained on the trusted runner at /var/lib/qkt/soak/attestation.json; generated artifacts are disposable and are not committed to source.

This wave is promotion evidence, not completion of the full notes matrix. Still unsealed are stop/stop-limit trigger-to-fill golden replay, cancellation races and partial fills, every re-entry and halt/cooldown recovery variant, H1/H4 live boundary parity, portfolio/book isolation, QKT Insights attribution for this exact wave, and exhaustive live-vs-backtest coverage for every registered indicator, numeric function, DSL construct, order type, and strategy template/deployment mode.

Higher-Timeframe Warmup Probe (2026-08-14)

The read-only exact-image probe at /var/tmp/qkt-validation/htf-e271822d-1786665912 passed against the same local gateway and demo account. It exercised nine closed-bar warmup requests: M15 at 1-hour/1-day/2-day windows, H1 at 1-hour/1-day/2-day windows, and H4 at 4-hour/1-day/2-day windows. All bars were aligned, unique, and closed; the account remained unchanged with zero positions, pending orders, and venue deals. The probe is warmup/bar-ingest evidence only; it does not seal higher-timeframe strategy signal or live-vs-backtest order parity.

Parallel Static Risk-Rejection Matrix (2026-08-14)

The exact promoted image ghcr.io/elitekaycy/qkt:edge (QKT commit e271822d4895cf16a662338c1f38090f77965485) passed the five-container local demo matrix at /var/tmp/qkt-validation/risk-rejection-e271822d-20260814-run. Maximum quantity, maximum notional, far-price collar, measured-usage, and operator-halt cases each produced a causal risk rejection before broker transport. The run verified zero mutating gateway requests, zero order events, zero fills, and no venue-state change while the five containers launched and deployed in parallel.

The result deliberately reports only preTransportStaticRejectionsPassed as true. Stateful daily-loss, drawdown, loss-streak, and margin-floor fixtures remain unpassed and are not inferred from this matrix.

Stateful Risk-Halt Matrix (2026-08-14)

The exact promoted image also passed the four-container stateful matrix at /var/tmp/qkt-validation/stateful-risk-e271822d-20260814-run. Persisted strategy-daily-loss, global-daily-loss, global-drawdown, and loss-streak state each tripped on live bars/ticks and produced the required causal halt and risk rejection before broker transport. The run required zero mutating gateway requests, zero order events, zero fills, and unchanged flat venue state.

The result reports dailyLossPassed, drawdownPassed, and lossStreakPassed as true while marginFloorPassed remains false. Margin-floor requires an owned live exposure and deterministic venue margin; it is still an explicit unsealed fixture rather than an inferred pass.

Exact-Image Insights Attribution (2026-08-14)

The fixed exact testing image 8d5b0cdde48e956db155793106e4dcf74bb1418c (OCI digest sha256:3e7787c0751c862b0d8f6a039c042aa86515b09c4af5dfe142d6ad7333047bb9) passed the local demo attribution runner at /var/tmp/qkt-validation/insights-8d5b0cdd-20260814-final3. The bundle covers an M1/M5 read-only sibling and one bounded 0.01-lot armed strategy. It recorded warmups, live ticks, matched M1/M5 evaluations, a real entry and strategy-owned exit, and final flat reconciliation.

Insights attribution sealed with two rule decisions, two decision-to-order links, two submitted/accepted/filled orders, two trades, two accounted fills, zero rejected events, zero dropped envelopes, and no duplicate event IDs. An intentional collector outage queued 330 envelopes; restart replay drained the journal. The open MT5 position was observed with the owning strategy before the strategy close. Final pending orders and positions were zero.

This is exact-wave Insights and causal round-trip evidence, not exhaustive coverage of every portfolio, deployment mode, indicator, DSL construct, or order boundary in notes.txt.

Order-Gateway Restart Evidence (2026-08-14)

The exact testing image 8d5b0cdde48e956db155793106e4dcf74bb1418c passed the order-bearing gateway restart runner at /var/tmp/qkt-validation/order-restart-8d5b0cdd-20260814-runtime/evidence/result.json. The bounded 0.01-lot owner strategy opened a real EURUSD position, the local MT5 gateway container was restarted while the position was open, and the same strategy closed the persisted venue ticket after reconnect. The bundle records two accepted/fill-accounted orders, six stale events with six recoveries, zero dropped ticks, final flat state, zero pending orders, and deal-net equality with the account balance delta (0.02). This seals restart/reconnect ownership and accounting for this exact wave; it does not seal every order type or every replay mode.

Native Stop-Limit Boundary Evidence (2026-08-14)

QKT commit d60ff76f changes the MT5 protocol capability set to advertise native STOP_LIMIT after the gateway support landed in mt5-gateway. Focused protocol, translator, and broker integration tests passed, and the change merged into dev as a82a94d4 through PR #1015. A localhost QKT CLI probe using that build placed a bounded BUY_STOP_LIMIT (retcode 10009, ticket 3086018968), showed the working order through the MT5 orders endpoint, and cancelled it through qkt bot cancel; the final account was unchanged and flat. The retained probe artifacts are at /var/tmp/qkt-validation/qkt-stop-limit-native-d60ff76f-20260814.

The probe did not trigger a stop-limit into a fill, and therefore does not claim stop-limit trigger-to-fill or live-vs-backtest parity. A separate exact-image strategy run must still capture the trigger, fill, cancellation races/partial fills, and golden replay before this order boundary can be promoted to main.

The temporary deployed-strategy probe at /var/tmp/qkt-validation/strategy-stop-limit-native-d60ff76f-20260814 also confirmed warmup, DSL compilation, and live deployment. Its first stop-limit decision occurred during a measured stale-feed episode, so QKT rejected it before broker transport; the feed recovered and the account remained flat with no pending order. This is evidence that stale-order suppression is fail-closed, not evidence of uninterrupted feed health or a native stop-limit fill.

A separate QKT CLI trigger observation at /var/tmp/qkt-validation/qkt-stop-limit-trigger-d60ff76f-20260814 placed a native BUY_STOP_LIMIT at 1.15392 with 1.15382 stop-limit price and polled the real venue for 180 seconds. The market remained below the trigger, so no deal or position occurred; QKT cancelled ticket 3086032248 and the account finished unchanged and flat. This seals the resting-order/cancel boundary only; the trigger-to-fill path is still unsealed.

Exact Testing-Image Four-Case Parity Wave (2026-08-14)

The corrected generated wave was run against testing revision 47e64f9372a32e611c0680e99123763e743848e4, image ghcr.io/elitekaycy/qkt@sha256:d27dd965850866ea1ac4374b86c77a6fa3ccb9bb2551846c4631b8150662258b, and local gateway 0.3.9 on demo account 436804390 / Exness-MT5Trial9. Evidence root: /var/tmp/qkt-validation/parity-47e6-retry-gb2.

All four read-only cases passed with 310-second captures, exact symbol routing, warmup/live tick and M1/M5 candle journals, golden captures, zero fills, zero dropped ticks, zero queue depth, and unchanged flat account state:

• atr-eurusd: 156 live ticks, 160 warmup ticks, 2/2 stale episodes recovered.
• ema-eurusd: 159 live ticks, 160 warmup ticks, 2/2 stale episodes recovered.
• case-gbpusd: 263 live ticks, 160 warmup ticks, 2/2 stale episodes recovered; warmup and audit symbols are EXNESS:GBPUSD.
• rsi-gbpusd: 259 live ticks, 160 warmup ticks, 1/1 stale episode recovered; warmup and audit symbols are EXNESS:GBPUSD.

Each armed case then passed with one real strategy-owned entry and one strategy- owned exit, two accepted and filled events, zero final positions/orders, and golden replay comparison passed:

• atr-eurusd/armed-live: ticket 3086526183, balance/deal delta +0.04.
• ema-eurusd/armed-live: ticket 3086540418, balance/deal delta -0.11.
• rsi-gbpusd/armed-live: ticket 3086556383, balance/deal delta -0.22.

Replay results are under each armed-live/replay/result.json and all report status: passed with the same testing SHA. The final gateway account snapshot was flat with zero margin. This wave proves the corrected four-case warmup, tick/bar, indicator, DSL order path, fill accounting, stale recovery, and golden replay on the exact image. It does not replace the remaining notes matrix for stop-limit trigger fills, cancellation/partial-fill races, risk/margin fixtures, portfolios/books, full indicator/math/DSL catalog, or Insights attribution.

Insights Verifier Follow-Up (2026-08-14)

The exact-image Insights run at /var/tmp/qkt-validation/insights-47e6-wave/cases/atr-eurusd completed the collector contract probe, read-only sibling, outage/restart replay, real entry, strategy-owned exit, and ticket/deal attribution checks. It was not sealed because the verifier wrote an empty duplicate-event-ids.json when SQLite returned no duplicate rows; the subsequent integer comparison treated the empty string as an error. This is a test-harness defect, not a trading-runtime or collector finding. The partial run is explicitly not promotion evidence.

PR #1019 (fix(scripts): normalize empty insights duplicate output) normalizes that empty query result to []. Linux CI is green and Windows CI is pending; the exact-image Insights run must be rerun after the fix is in the promoted testing image before Insights attestation is claimed.

Exact Testing-Image Higher-Timeframe Warmup (2026-08-14)

Clean testing-worktree probes passed for both EXNESS:EURUSD and EXNESS:GBPUSD at /var/tmp/qkt-validation/htf-47e6-clean-EURUSD/evidence/result.json and /var/tmp/qkt-validation/htf-47e6-clean-GBPUSD/evidence/result.json. Both report testing SHA 47e64f9372a32e611c0680e99123763e743848e4, qktDirty: false, nine probes, and zero final positions/orders. Every M15/H1/H4 one-hour, one-day, and two-day request returned closed, aligned, unique bars.

The attempted XAUUSD probe was not sealed: the local demo gateway returned no reviewed closed M15 one-hour bar set. This is classified as symbol/data availability, not a QKT warmup pass or code failure; XAUUSD remains unproven.

Exact Testing-Image Read-Only Catalog (2026-08-14)

The four-container catalog was rerun against the same testing image and local gateway at /var/tmp/qkt-validation/catalog-47e6-final-live/evidence/result.json. It passed with revision 47e64f9372a32e611c0680e99123763e743848e4, image digest sha256:d27dd965850866ea1ac4374b86c77a6fa3ccb9bb2551846c4631b8150662258b, 360 seconds, four parallel containers, five parallel tick symbols, 500 ms tick polling, and no JVM or Docker resource restrictions.

The catalog covered numeric/candle indicators and math, cross-symbol M1/M5 mapping, session/history/stateful functions, and the volume-capability negative case. All four cases passed their warmup bars, readiness vectors, live ticks, constructed bars, joined evaluations, and financially read-only assertions. Aggregate counts were zero gateway mutations, order events, fills, and venue deals; account state stayed unchanged. The run recorded 7 stale episodes and recovered all 7, with zero in-window disconnect warnings, zero unexpected errors, and no dropped-tick failure. The earlier catalog dropped-tick result is therefore superseded for this exact image, while the gateway stale/recovery events remain an operational observation.

Current Promotion Reconciliation (2026-08-14)

The remote refs were refreshed after the historical sections above were written. The currently authoritative release refs are:

• dev: f8d91366f413331a8d208c2c4cb73aa2c7262555, including PR #1019, fix(scripts): normalize empty insights duplicate output.
• testing: daf5b4639c53d3244672207d710db78f38c91e3c, promoted from that dev revision. Its exact QKT image is ghcr.io/elitekaycy/qkt@sha256:e7eb41cfe6300b7ec599b83db7c642e1aa66b7a18ef8f506d0e59abacbf15b6c.
• main: f275065947c9b967b85843ba1c12a8e16f90d587, the last completed testing-to-main promotion. It does not yet contain PR #1019.

The corrected Insights verifier must be rerun against the immutable testing image above before another promotion. Earlier Insights evidence generated from a different image is historical and cannot attest this revision. The notes matrix also remains open for stop-limit trigger/fill replay, cancellation and partial-fill races, margin-floor, portfolio/book isolation, and exhaustive capability coverage.

Corrected Exact-Testing Insights Attestation (2026-08-14)

The corrected verifier passed against QKT testing revision daf5b463 and the immutable QKT image ghcr.io/elitekaycy/qkt@sha256:e7eb41cfe6300b7ec599b83db7c642e1aa66b7a18ef8f506d0e59abacbf15b6c. Evidence is retained at /var/tmp/qkt-validation/insights-daf5-prep2/evidence/result.json.

The run recorded M1/M5 warmup, live ticks, matched evaluations, a real bounded entry and strategy-owned exit, and final flat reconciliation. Insights retained two rule decisions, two decision-to-order links, two submitted/accepted/filled orders, two trades, two accounted fills, zero rejected events, zero dropped envelopes, and a maximum duplicate-attempt count of one. An intentional collector outage queued 338 envelopes and restart replay drained them completely. The final state had zero pending orders and zero positions.

This seals the Insights verifier fix for daf5b463; it does not close the other notes-matrix gaps listed above.

Focused Order and Persistence Regression (2026-08-14)

On the current checkout, the focused JUnit suite passed for MT5 simulator order boundaries, stop-limit activation/fill, partial-fill slicing, cancellation races, DSL stop-limit compilation/rendering, and concurrent state-file reads/writes. StateFileWriterTest.concurrent reads while writing never see a torn file passed; the run emitted expected slow-write warnings on the contended filesystem but no torn reads or failed writes. Command: ./gradlew test --tests 'com.qkt.trade.BotActionCompilerTest' --tests 'com.qkt.trade.BotDslRendererTest' --tests 'com.qkt.persistence.StateFileWriterTest' --tests 'com.qkt.broker.MT5BrokerSimulator*'.

These are regression/unit results and do not substitute for the still-open localhost MT5 stop-limit trigger/fill and live partial-fill evidence.

Margin-Floor Rerun Status (2026-08-14)

A clean exact-daf5b463 localhost fixture was exercised with the single Exness demo account. It opened a real 0.01 EURUSD position, derived the dynamic floor from the live margin level, rejected the probe before broker transport, and recovered a probe position after the opener was flattened. The runner did not finish its final journal/result sealing phase, so these captures remain unsealed and are not promotion evidence. The account was explicitly verified flat after cleanup. A sealed margin-floor result is still required.

Sealed Exact-Testing Margin-Floor Fixture (2026-08-14)

The corrected runner passed against testing revision ba1809217ad8f6c9db494054d8f0641a48a6704c and immutable image ghcr.io/elitekaycy/qkt@sha256:35c82798dab413fe39dde48ff2dc81f8c62fd5f5e40c2a028d564887fdce6636. Evidence is retained at /var/tmp/qkt-validation/margin-ba1809-live/evidence/result.json.

The fixture opened one bounded 0.01 EURUSD position with the opener role, derived the floor from observed live margin level, rejected the probe before any mutating gateway request, then allowed and filled the probe after opener flattening and headroom recovery. Both strategy-owned positions were flattened; venue history and account balance reconciled, with zero final positions and pending orders. The aggregate result reports marginFloorPassed: true and productionReadiness: false (the latter remains intentionally conservative).

Exact d55 Live Matrix and Insights (2026-08-14)

All evidence below uses testing revision d55f8f51d70831006615132cf86a36938b3a46f3 and QKT image ghcr.io/elitekaycy/qkt@sha256:083052c7ce1467b01fe604b4634ce2ed33d081f1d3e4f33e71801295ee74402d. Generated evidence is disposable and retained outside the source tree.

• Four-case ATR, case-math, EMA, and RSI parity wave passed read-only warmup, live demo bracket, and full-tick/bar/MT5 replay comparisons at /var/tmp/qkt-validation/parity-d55-live. The account finished flat.
• Four-case catalog passed at /var/tmp/qkt-validation/catalog-d55-live/evidence/result.json, covering numeric/candle, cross-symbol and multi-timeframe, session/history, and volume capability rejection with zero mutations and zero dropped ticks.
• Five-case pre-transport risk rejection passed in isolation at /var/tmp/qkt-validation/riskreject-d55-retry/evidence/result.json.
• Four-case restored-state risk matrix passed in isolation at /var/tmp/qkt-validation/stateful-d55-retry/evidence/result.json.
• QKT Insights attribution passed at /var/tmp/qkt-validation/insights-d55-scenario/evidence/result.json: 341 outage-queued envelopes replayed completely; two decisions/orders/fills were attributed to the owning strategy; no drops or duplicate attempts; final flat.

The intentional nine-daemon aggregate probe failed because the local gateway's Waitress queue saturated and live tick feeds disconnected, causing deploy timeouts. Each matrix passed when run alone. This is an explicit gateway capacity limit, not a claimed production-scale pass, and needs a gateway/load fix or an enforced concurrency bound before unrestricted multi-container use.

d55 Attestation Dispatch

A six-artifact d55 attestation passed the local verifier and is retained at /var/tmp/qkt-validation/attest-d55-final. Trusted workflow 31800477867 was dispatched against testing d55, but its sole qkt-paper-soak runner is offline, so it remains queued. Local verification cannot substitute for the trusted workflow; main promotion remains blocked until the runner executes successfully.