โ€บNavigation
Closing reconciliation

Asked to wire two already-approved measurements of the shape of recent returns into real forex production columns. Came back with three stacked pull requests in which both features reproduce their reference numbers to within a billionth.

What it measured
QuantityValueMeaning
Features ported2 of 2 (cards #38, #7)both batch-3/4 cards landed
Parity rows per feature11,200 finite, equal to the oracle countnon-vacuous comparison, no silent NaN fill
Card #38 max deviation vs oracle1.4e-17 batch / 8.8e-16 streaming (tolerance 1e-9)about 8 orders inside tolerance
Card #7 max deviation vs oracle3.6e-12 batch / 1.2e-10 streaming (tolerance 1e-9)inside tolerance
Adversarial challenges18 attacks, 17 held, 1 conceded (doc staleness, fixed)the port survived attack
Card #7 sign witness9,097 of 11,200 negative; XAUUSD at 25 pinned -0.5838646869163802replaces the stale -0.605 in the contract
How it ran
FiringWhat it did
01Minted the v3 cohort, pinned numpy/scipy reference versions, added 4 anchor-pin tests
02Ported cumulative_residual_extropy (#38) into a Rust kernel and checked it against the reference
03Ported right_quantile_weight_tailweight (#7), reusing an already-shipped percentile primitive
What it produced
Carries forward

The page's handoff still specifies a squash-merge order for the three stacked pull requests; of those, only the last one's merge (2026-06-25T09:49:51Z) is evidenced in the folder. Operator note to preserve: the binding card-#7 witness is -0.5838646869163802, not the contract's -0.605.

Why it is closed

'The loop is complete: all three PRs are now OPEN as a stack (#124 -> #126 -> #127), every gate green and parity bit-exact-class.' (index.html; ledger adds mql5#127 merged 2026-06-25)

Lifecycle status is process state, not a judgement of the findings. Results are stated as numbers with their uncertainty. Generated between CLOSING-RECONCILIATION markers โ€” regenerate rather than edit.

Forex orthogonal features ยท production implementation ยท batch 3-4 ยท RQW + CREx
๐Ÿช™ FOREX 2ร— IMPLEMENT-TODAY STACKED-PR ยท MORE ROBUST 2026-06-24 ยท 3 PRs OPEN โ€” awaiting squash-merge

Forex Batch-3/4 โ€” RQW + CREx Feature Implementation (2ร—, MORE robust)

A live, autonomous run on bigblack that ports two batch-3/4 orthogonal features โ€” cumulative_residual_extropy (#38) and right_quantile_weight_tailweight (#7) โ€” into real fxview_cache.forex_bars columns. This is a stacked-PR run: PR1 bootstrap โ†’ PR2 #38 (stacked on bootstrap) โ†’ PR3 #7 (stacked on #38); the operator squash-merges in order. Repo terrylica/mql5. The loop is complete: all three PRs are now OPEN as a stack (#124 โ†’ #126 โ†’ #127), every gate green and parity bit-exact-class. The flowcharts below show every lane fully done.

In plain terms

Two genuinely-new "shape of the recent returns" measurements ship in this run. Both are computed on the same load-bearing substrate (a gap-aware trailing-200-bar log-return window), and both already passed the orthogonality KEEP test โ€” so this run is pure production wiring, not discovery.

The load-bearing substrate (the convention both features share)

Both kernels read the same gap-aware log-return window. The exact recipe was transcribed from multislice_orthogonality_probe.py (the k_crex / k_rqw paths) โ€” NOT the discovery harness, whose sign-dropped, no-gap-null variant would silently diverge.

  1. Trailing 200-bar window of closes.
  2. logret = diff(log(close)).
  3. Null every inter-bar gap > 8h (weekend / Ouroboros weekly reset) โ€” the gap-crossing return becomes NaN.
  4. Drop NaNs.
  5. Require โ‰ฅ 8 finite returns else the feature is NULL.

numpy pin: numpy 2.5.0 / scipy 1.18.0 (uv --python 3.14). Parity contract is VALUE-within-1e-9, not byte-sha โ€” the oracle is a pinned numpy reference, not a frozen blob.

2-feature ledger

FeatureCardSubstrateTier / TOLHardening witnessStatusParityChallenge-and-HeldPR
#38 cumulative_residual_extropy 38 gap-aware logret ฮต 1e-9 scale-witness RAW scale, CRExโ‰ค0; z-scoring FAILS parity PR OPEN 11,200 finite (== oracle, non-vacuous)
max-abs-dev 1.4e-17 batch / 8.8e-16 streaming
vs TOL 1e-9		 10 attacks ยท 10 held ยท 0 conceded #126
#7 right_quantile_weight_tailweight 07 gap-aware logret ฮต 1e-9 sign-witness leading minus; 9,097/11,200 negative; XAUUSD@25 pinned โˆ’0.5838646869163802; reuses shipped RollingPercentile HF7 (R2, bit-identical on 33,600 octile reads) PR OPEN 11,200 finite (== oracle, non-vacuous)
max-abs-dev 3.6e-12 batch / 1.2e-10 streaming
vs TOL 1e-9		 8 attacks ยท 7 held ยท 1 conceded (doc-staleness) FIXED #127

Orthogonality KEEP evidence (already satisfied, not re-run): #38 ฮพ 0.160 / 0.395 (close) independent; #7 ฮพ 0.135 / 0.260 independent (2026-06-12-xi-batch34-crossed-verdict.md).

โš  #7 witness value corrected. The contract's #7 sign-witness was stated as โˆ’0.605, but the binding multislice k_rqw on the FROZEN fixture gives โˆ’0.5838646869163802 (XAUUSD@25). No numpy quantile method reproduces โˆ’0.605 โ€” that figure was a stale full-production value. The binding value is pinned and the v3 anchor was re-locked.

โš‘ Operator morning handoff

All 3 PRs OPEN as a stack (#124 โ†’ #126 โ†’ #127). Squash-merge in order; the loop will offer to rebase each child onto main after its parent merges. All gates green; parity bit-exact-class. One operator note: #7 witness value corrected to โˆ’0.5838646869163802 (frozen fixture); contract's โˆ’0.605 was stale.

PR links โ€” squash-merge in order
  • PR1 (bootstrap โ€” v3 cohort, base=main) OPEN: #124 ยท feat/odb-ext/rqw-crex-bootstrap
  • PR2 (#38 cumulative_residual_extropy, base=PR1) OPEN: #126 ยท feat/odb-ext/cumulative_residual_extropy
  • PR3 (#7 right_quantile_weight_tailweight, base=PR2) OPEN: #127 ยท feat/odb-ext/right_quantile_weight_tailweight

Iteration log

One page per loop iteration. Each overview-flowchart lane below is clickable and links here; the loop overwrites the matching page when it reaches that iteration.

iterpagecoversstatus
01bootstrapv3 cohort (minted) ยท numpy-pin SSoT/tolerance ยท 3 hygiene fixes ยท dashboard placeholdersDONE ยท PR1 #124 OPEN
02cumulative_residual_extropyfeature #38 (stacked on bootstrap)PR OPEN ยท PR2 #126 OPEN
03right_quantile_weight_tailweightfeature #7 (stacked on #38)PR OPEN ยท PR3 #127 OPEN

Implementation flowcharts

One overview (both features as lanes) + one per feature. Every node is now green โ€” the full ยง5 pipeline (worktree โ†’ gap-aware oracle โ†’ kernel port โ†’ wire โ†’ SOTA ยง6+ยง6.RG โ†’ all-gates โ†’ no-split-brain โ†’ challenge-held โ†’ PR OPEN) is complete for each lane.

pending active done ๐Ÿ”ด blocker ๐ŸŸก robustness

Overview โ€” both features (stacked PRs)

flowchart LR
  B["iter 01 ยท bootstrap
v3 cohort minted ยท numpy-pin SSoT
placeholders ยท PR1 #124 OPEN"]:::done
  B --> F38["#38 cumulative_residual_extropy
gap-aware logret ยท scale-witness
stacked on bootstrap ยท PR2 #126 OPEN"]:::done
  F38 --> F7["#7 right_quantile_weight_tailweight
gap-aware logret ยท sign-witness ยท reuse HF7
stacked on #38 ยท PR3 #127 OPEN"]:::done
  F7 --> DONE(["all 3 PRs OPEN โ€” operator squash-merges
#124 โ†’ #126 โ†’ #127"]):::done
  click B "index_iter_01_bootstrap.html" "iter 01 ยท bootstrap"
  click F38 "index_iter_02_cumulative_residual_extropy.html" "iter 02 ยท #38"
  click F7 "index_iter_03_right_quantile_weight_tailweight.html" "iter 03 ยท #7"
  classDef pending fill:#334155,stroke:#475569,color:#e2e8f0;
  classDef active fill:#1d4ed8,stroke:#3b82f6,color:#fff;
  classDef done fill:#15803d,stroke:#22c55e,color:#fff;
  classDef blocker fill:#b91c1c,stroke:#ef4444,color:#fff;
  classDef robust fill:#b45309,stroke:#f59e0b,color:#fff;

#38 cumulative_residual_extropy iter 02 โ†—

flowchart LR
  W["worktree off
stacked on bootstrap"]:::done --> O["gap-aware oracle
CREx ยท ฮต 1e-9 ยท numpy 2.5.0"]:::done
  O --> K["kernel port
(Rust, causal)"]:::done
  K --> WIRE["wire: BarRow ยท COLUMN_NAMES ยท
BarDerivedV2 ยท schema.sql ยท feature_metadata"]:::done
  WIRE --> S["SOTA ยง6 + ยง6.RG
๐ŸŸก scale-witness: z-scoring FAILS parity
11,200 finite ยท max-dev 1.4e-17/8.8e-16"]:::done
  S --> G["all-gates
GREEN"]:::done
  G --> NS["no-split-brain
one PR"]:::done
  NS --> CH["challenge-held
10 attacks ยท 10 held ยท 0 conceded"]:::done
  CH --> PR(["PR2 #126 OPEN ยท stacked on bootstrap"]):::done
  classDef pending fill:#334155,stroke:#475569,color:#e2e8f0;
  classDef active fill:#1d4ed8,stroke:#3b82f6,color:#fff;
  classDef done fill:#15803d,stroke:#22c55e,color:#fff;
  classDef blocker fill:#b91c1c,stroke:#ef4444,color:#fff;
  classDef robust fill:#b45309,stroke:#f59e0b,color:#fff;

#7 right_quantile_weight_tailweight iter 03 โ†—

flowchart LR
  W["worktree off
stacked on #38"]:::done --> O["gap-aware oracle
RQW octiles ยท ฮต 1e-9 ยท numpy 2.5.0"]:::done
  O --> K["kernel port
reuse RollingPercentile HF7 (R2, bit-identical 33,600 reads)"]:::done
  K --> WIRE["wire: BarRow ยท COLUMN_NAMES ยท
BarDerivedV2 ยท schema.sql ยท feature_metadata"]:::done
  WIRE --> S["SOTA ยง6 + ยง6.RG
๐ŸŸก sign-witness: XAUUSD@25 โ†’ โˆ’0.5838646869163802
(contract's โˆ’0.605 stale) ยท 11,200 finite ยท max-dev 3.6e-12/1.2e-10"]:::done
  S --> G["all-gates
GREEN"]:::done
  G --> NS["no-split-brain
one PR"]:::done
  NS --> CH["challenge-held
8 attacks ยท 7 held ยท 1 conceded (doc) FIXED"]:::done
  CH --> PR(["PR3 #127 OPEN ยท stacked on #38"]):::done
  classDef pending fill:#334155,stroke:#475569,color:#e2e8f0;
  classDef active fill:#1d4ed8,stroke:#3b82f6,color:#fff;
  classDef done fill:#15803d,stroke:#22c55e,color:#fff;
  classDef blocker fill:#b91c1c,stroke:#ef4444,color:#fff;
  classDef robust fill:#b45309,stroke:#f59e0b,color:#fff;