chiki2bum2/Centaur_Quant_Architecture
Watch
0
0
Fork
You've already forked Centaur_Quant_Architecture
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
Centaur_Quant_Architecture/00_CONCEPT/EXP003B_RESULTS/EVIDENCE_PROTOCOL_v0.1.md

31 KiB
Raw Permalink Blame History

EVIDENCE PROTOCOL v0.1 — EXP-003B / GATE B

CENTAUR QUANT ARCHITECTURE — EXP-003B · Governance / Evidence Specification · Version: 0.1 (DRAFT — AWAITING HUMAN RATIFICATION) · Created: 2026-08-15 · Status: REQUIRED REPOSITORY ARTIFACT per ratified rule AE+8

0. Document Identity and Status

  • This document is a governance / evidence specification only.
  • It is the repository artifact required by ratified rule AE+8 of 00_CONCEPT\EXP003B_RESULTS\ANALYSIS\gate_b_human_ratification_decision_COMPLETED.md ("Evidence Protocol v0.1 is required as a repository artifact before execution").
  • It is a DRAFT for human ratification. It does not become authoritative until explicitly ratified by the human principal.
  • This document grants NO authorization — specifically, it grants no instrumentation authorization, no source-modification authorization, no compilation authorization, no tester-execution authorization, and no Gate B execution authorization (see §22).
  • It reproduces only already-ratified Gate B rules from the authoritative human-ratification record and its verified supporting audits. It introduces no new statistical rule, no new threshold, no new sample requirement, no new OOS rule, no new stop condition, and no new execution permission.
  • Where ratified governance does not specify a value, this document marks the item: UNSPECIFIED — HUMAN DECISION REQUIRED. No assumption is invented.

0.1 Authority chain (precedence)

  1. gate_b_human_ratification_decision_COMPLETED.md — authoritative human-ratification record (highest precedence for Gate B governance).
  2. post_ratification_governance_lock_audit.md — verified governance lock (2026-08-15 14:41 +07).
  3. gate_b_pre_execution_instrumentation_readiness_audit.md — 33-field evidence schema and stop/abort evidence mapping (2026-08-15 14:46 +07; verified consistent with the ratification record).
  4. gate_b_protocol_addendum_audit.md — operational specification the human ratified (2026-08-15 14:24 +07).
  5. EXP-003_SCANNER_ACTIVATION.md (ratified at commit f81d85d) — experiment objective, threshold 55.0, Phase-4 intent, diagnostic mode.
  6. Step 1–8 artifacts (gate_a_evaluation.md, observation_unit_definition.md, oracle_reconciliation.md, C0_quality_report.md, etc.) — verified evidence chain.

Any future change to this protocol requires a new explicit human governance decision.

1. Purpose and Scope

1.1 Purpose

Define, as a governance specification, the complete evidence handling required to make a future Gate B evaluation auditable and reproducible: the mandatory evidence lifecycle, every mandatory evidence field, field provenance, referential integrity, timestamp/event ordering, identity rules, missing- and invalid-evidence treatment, R-multiple reconstruction, partial-exit handling, cost representation, zone clustering, dataset-snapshot identity, binary/config provenance, manifest/freeze procedure, evidence preservation, and the mapping of the 18 ratified stop/abort conditions to the evidence required to detect them.

1.2 Scope

  • In scope: evidence specification for the execution stage of EXP-003B / Gate B (the stage that produces order attempts, positions, management events, exits, and round-trips). C0 scanner-stage evidence handling (Steps 1–8) is already executed and frozen; this protocol adopts its PRESERVED → HASHED → FROZEN pattern as precedent.
  • Out of scope: statistical inference rules (ratified separately; see §2 reference), parameter optimization, OOS-barrier definition (ratified Option A; formal barrier deferred to Phase 5/6), live trading, and any source modification. Instrumentation implementation is explicitly NOT authorized by this document.

1.3 Current binding state (unchanged, verified)

GATE A = PASS
B1 = RETIRED — ORIGINAL CONTINGENCY INVALIDATED
GATE B GOVERNANCE = RATIFIED
GATE B GOVERNANCE LOCK = VERIFIED / COMPLETE
AE+6 POWER CALCULATION = DEFERRED
INSTRUMENTATION = NOT AUTHORIZED
SOURCE MODIFICATION = NOT AUTHORIZED
COMPILATION = NOT AUTHORIZED
GATE B EXECUTION = NOT AUTHORIZED
RAW EVIDENCE = FROZEN
BASELINE = IMMUTABLE

2. Ratified Gate B Governance Reference (reproduced verbatim in substance)

The following are already-ratified rules. This section is a reference only; nothing here is new.

Rule Ratified value Source
Primary endpoint net R per completed round-trip record A
Primary comparison E[net R | fused ≥ 55] vs E[net R | fused < 55]; threshold 55.0 fixed record B; C (N/A — existing)
Effect-size floor Cohen's d ≥ 0.20 (a priori) record D
CI method cluster bootstrap, 10,000 resamples, structural-zone clusters record E
Significance α = 0.05, secondary to effect size record F
Minimum sample ≥ 50 completed round-trips per fusion class AND ≥ 50 distinct structural zones record G
Interim analysis None record H
Stopping One terminal decision after evidence freeze record I
Insufficient sample NOT EVALUABLE (never converted to PASS/FAIL) record J
Signal unit structural zone (direction, zone_low, zone_high) record K (N/A — Step 7 existing)
Outcome unit completed round-trip record L
Dependence round-trips clustered by structural zone; never treated as independent record M
Same-zone re-entry allowed only when each round-trip complete and traceable record N
First-entry-only rejected record O
Lifecycle CANDIDATE → DECISION → ORDER ATTEMPT → ACCEPT/REJECT → POSITION OPEN → MANAGEMENT → POSITION CLOSE → ROUND-TRIP COMPLETE record P
Identifiers zone_id, attempt_id, round_trip_id (= position ticket) record Q
Missing evidence absence = governance event; never guessed/reconstructed record R
OOS OPTION A: in-window confirmatory; no new split; formal barrier deferred to Phase 5/6; no frozen dataset/barrier modified record S, §6
Dataset identity pre-run identity mandatory record T
Dataset comparison compare vs C0 (11,639 bars / 696,947 ticks); discrepancy reported record U
Execution mode diagnostic MT5 Strategy Tester only; no live/VPS/broker transmission record V
Intervention prohibited except defined emergency abort record W
R-multiple R fixed to initial risk at entry; partial exits handled consistently record X
Monetary P&L secondary until independent tick-value reconciliation record Y
Transaction costs preserved separately, normalized to initial-risk units record Z
Risk scaling R is lot-size invariant; no reweighting to compensate for risk scaling record AA
Binary provenance binary identity + SHA-256 captured before execution record AB
Manifest/freeze evidence manifested and frozen before analysis record AC
Stop/abort 18-condition matrix authoritative (reproduced in §18) record AD
Reproducibility target; all stated prerequisites satisfied before execution record AE
Censoring end-of-window open positions censored, counted, explicitly reported record AE+1
Invalid outcomes excluded only with documented reason and preserved evidence record AE+2
Zero-R R=0 retained; zero-risk cases invalid record AE+3
No reweighting R observations never reweighted for risk scaling record AE+4
Arm isolation future arms isolated; no pooling without new protocol + ratification record AE+5
Power calculation DEFERRED until empirical variance, expected effect size, zone ICC known (non-blocking; prerequisite only for future power claim) record AE+6
Evidence schema ratified as a governance requirement; instrumentation separately unauthorized record AE+7
Protocol artifact Evidence Protocol v0.1 required as repository artifact before execution record AE+8

3. Evidence Lifecycle (ratified P — operational definition)

Every terminal state must be reconstructable from structured records. The mandatory lifecycle is:

CANDIDATE → DECISION → ORDER ATTEMPT → ACCEPT/REJECT
         → POSITION OPEN → MANAGEMENT → POSITION CLOSE
         → ROUND-TRIP COMPLETE
Stage Definition (evidence-level)
CANDIDATE Scanner produces a structural zone in proximity; candidate row recorded (CSV pattern, C0 precedent).
DECISION Fusion classifies candidate PASS/SKIP vs fixed threshold 55.0; decision row recorded.
ORDER ATTEMPT An order request is submitted for a PASS candidate; attempt record created.
ACCEPT/REJECT Broker/tester response recorded (retcode, description, deal ref) — accept or reject.
POSITION OPEN Accepted order becomes a position; position-open record with ticket and actual fill.
MANAGEMENT Every SL/TP modification (break-even, trailing) recorded.
POSITION CLOSE Exit event recorded (timestamp, price, reason, deal id).
ROUND-TRIP COMPLETE Round-trip summary record: initial risk R0, gross R, cost_R, net R, monetary P&L (if reconcilable), reconciliation status.
  • A record must exist for every stage. Absence of a mandatory record is a governance event (ratified R) and triggers the stop/abort mapping (§18).
  • A candidate that is SKIPped does not proceed beyond DECISION; its records remain part of the evidence (C0 precedent: PASS and SKIP both logged).

4. Mandatory Evidence Fields (operational schema)

The mandatory field set below operationalizes ratified lifecycle rule P, identifier rule Q, and evidence-schema rule AE+7. The field set is the 33-field schema (A–AG) from the verified gate_b_pre_execution_instrumentation_readiness_audit.md §5, which was audited as consistent with the ratification record. No field is considered implemented merely because equivalent information appears in Print logs (see §14).

Legend — Provenance: CSV = exists in C0 candidate CSV pattern · INSTR = requires new structured instrumentation record (NOT AUTHORIZED by this document) · DERIVED = computed from recorded fields · REPORT = from preserved tester report · ABSENT = currently absent from evidence. Identity: zone_id · attempt_id · round_trip_id.

ID Stage Mandatory field Provenance Type Identity
A CANDIDATE Candidate identity CSV string zone_id
B CANDIDATE Structural zone (direction, zone_low, zone_high) CSV string zone_id
C CANDIDATE Timestamp CSV datetime —
D CANDIDATE Direction (LONG/SHORT) CSV enum —
E DECISION Fused score CSV double —
F DECISION Threshold (55.0) CSV double —
G DECISION Decision (PASS/SKIP) + reason CSV enum+string —
H ORDER ATTEMPT Attempt ID INSTR ulong attempt_id
I ORDER ATTEMPT Attempt timestamp INSTR datetime —
J ORDER ATTEMPT Order type INSTR enum —
K ORDER ATTEMPT Requested volume INSTR double —
L ORDER ATTEMPT Requested entry CSV double —
M ORDER ATTEMPT Initial SL CSV (+ comment) double —
N ORDER ATTEMPT Initial TP CSV double —
O ACCEPT/REJECT Broker/server response description INSTR string —
P ACCEPT/REJECT Retcode INSTR uint —
Q POSITION OPEN Position ticket INSTR ulong round_trip_id
R POSITION OPEN Deal ticket(s) INSTR ulong[] —
S POSITION OPEN Actual fill price INSTR double —
T POSITION OPEN Actual volume INSTR double —
U POSITION OPEN Position-open timestamp INSTR datetime —
V MANAGEMENT Management events INSTR (batched) rows ticket
W MANAGEMENT SL/TP modifications INSTR (batched) rows ticket
X POSITION CLOSE Exit timestamp INSTR datetime —
Y POSITION CLOSE Exit price INSTR double —
Z POSITION CLOSE Exit reason INSTR enum —
AA POSITION CLOSE Closure/deal identity INSTR ulong —
AB ROUND-TRIP Round-trip identity INSTR ulong round_trip_id
AC ROUND-TRIP Initial risk R0 (entry − SL) DERIVED double —
AD ROUND-TRIP Realized R (gross) DERIVED double —
AE ROUND-TRIP Net R (gross R − cost_R) DERIVED double —
AF ROUND-TRIP Monetary P&L REPORT double —
AG ROUND-TRIP Reconciliation status DERIVED enum round_trip_id
  • Fields marked INSTR cannot be produced by the current source state (verified: order-attempt, position, management, exit, round-trip records are Print-only or absent). Producing them requires separately authorized instrumentation. This document does not authorize that.
  • Fields V/W (management, SL/TP modifications) fire in OnTick and therefore require mandatory batching (see §5.3 and §21) to avoid per-tick I/O explosion.

5. Field Provenance Rules

  1. CSV fields (A–G, L, M, N): already produced by the C0 candidate record pattern (AppendSetupCSV); values must be byte-traceable to the frozen C0 schema semantics (31 columns; ob_time_age_bars = seconds despite name).
  2. INSTR fields (H–K, O–U, V, W, X–AB): must be written by dedicated structured evidence records (attempt log, position-open log, management log [batched], exit log, round-trip summary). Print-only equivalents do not satisfy the schema (§14).
  3. DERIVED fields (AC–AE, AG): computed only from recorded structured fields per §10 (R-multiple) and §11 (costs); derivations must be reproducible from the recorded evidence alone.
  4. REPORT field (AF): monetary P&L from the preserved tester report only; secondary metric (ratified Y). Never reconstructed from prices without independent tick-value reconciliation.
  5. Every field must be deterministic given the recorded inputs and the ratified rules — no field may require discretionary judgment.

5.1 Referential integrity (ratified Q)

  • zone_id = (direction, zone_low, zone_high) — foreign key from every execution-stage record back to the CANDIDATE/DECISION CSV rows.
  • decision_id = zone_id + decision timestamp (per verified addendum §7).
  • attempt_id = sequential per (zone_id, decision_id); a rejected/failed attempt gets its own attempt row with retcode and no round-trip.
  • round_trip_id = position_ticket; every deal under the round-trip (deal_ids[]) attaches to it.
  • Referential chains must be fully resolvable: zone_id → decision_id → attempt_id → position_ticket → deal_ids[].

5.2 Timestamp / event ordering requirements

  • Event timestamps recorded in tester simulated time (C0 convention: YYYY.MM.DD HH:MM:SS, 1-second granularity).
  • Event order must conform to the §3 lifecycle for every round-trip: candidate ≤ decision ≤ attempt ≤ accept/reject ≤ open ≤ close (≤ window end).
  • Where events share the same simulated timestamp (C0 observed same-second re-emissions), ordering is established by the referential identity chain (attempt sequence, ticket/deal linkage), never by guessing.
  • Wall-clock time may be recorded additionally for operational auditing but must never replace simulated-time ordering.

5.3 Batching of management events (V/W)

  • Management-event records V/W must be batched (mandatory requirement to control per-tick I/O; see instrumentation readiness audit §5/§6 and §21).
  • Exact batch interval/count parameters: UNSPECIFIED — HUMAN DECISION REQUIRED.

6. Identity Rules (ratified Q — operational)

Identifier Definition Created at Notes
zone_id (direction, zone_low, zone_high) CANDIDATE exists in CSV; Step-7 signal unit
decision_id zone_id + decision timestamp DECISION one per PASS/SKIP evaluation (verified addendum §7)
attempt_id sequential per (zone_id, decision_id) ORDER ATTEMPT rejected attempts have attempt_id, no round_trip_id
round_trip_id position ticket POSITION OPEN parent identity for all deals of the round-trip
  • Same-zone re-entry produces distinct position tickets → distinct round_trip_ids (ratified N: permitted when each round-trip is complete and traceable).
  • A round-trip is traceable iff its full §3 lifecycle chain resolves via the identifiers above.

7. Missing Evidence Treatment (ratified R)

  1. Absence of any mandatory record/field is a governance event.
  2. Missing evidence is never filled by:
    • Print/log lines, tester summaries, later position state, timestamps, P&L values, assumptions, or synthetic events.
  3. Every missing-evidence occurrence must be:
    • recorded (which field, which stage, which round-trip/attempt),
    • preserved (partial evidence retained as-is),
    • mapped to the stop/abort action in §18 (e.g., missing mandatory event → ABORT; missing exit / open at end → CENSOR+REPORT; missing initial SL → INVALID OUTCOME / CONTINUE).
  4. No reconstruction, no imputation, no "normalization" of missing evidence.

8. Invalid Outcome Treatment (ratified AE+2, AE+3, X)

  • Invalid outcomes (e.g., missing initial SL/entry/exit, or failed reconciliation) are excluded from the primary endpoint only with:
    • a documented reason, and
    • the preserved underlying evidence.
  • Zero-risk cases (SL-distance = 0 or entry = exit) are invalid (ratified AE+3).
  • R = 0 outcomes are retained as valid observations when risk0 is valid (ratified AE+3).
  • Exclusions and their reasons must be reported in the analysis manifest; they must never be silently dropped.

9. Censored Outcomes (ratified AE+1)

  • Positions still open at window end are censored, counted, and explicitly reported — never silently treated as completed round-trips.
  • Censored positions do not enter the primary endpoint (completed round-trips).

10. R-Multiple Reconstruction (ratified X, AA; AE+3)

Definition (operational form of ratified rule X, per verified addendum §11):

LONG :  R = (exit_price − entry_price) / risk0
SHORT:  R = (entry_price − exit_price) / risk0
risk0 = |entry_price − initial_SL|     (price units, fixed at entry)

Rules:

  • entry_price = actual fill price at position open (field S).
  • initial_SL = zone.sl (source of truth; also persisted in comment CEN:<score>:<sl>); source for round-trip R.
  • Break-even and trailing moves do not redefine risk0 (R = realized / initial risk by definition).
  • exit_price = volume-weighted mean exit price across partial-exit deals (see §10.1).
  • Missing initial SL → invalid outcome (§8).
  • Zero-risk → invalid (§8).
  • R is lot-size invariant (both numerator and denominator scale with volume) → anti-veto risk scaling does not bias E[net R] per class; no reweighting is permitted to compensate for risk scaling (ratified AA, AE+4).
  • Position reversal: NOT APPLICABLE (no reversal logic in source).

10.1 Partial exits

  • Multiple closure deals under the same round_trip_id are aggregated by volume-weighted mean exit price: exit_price = Σ(deal_volume_i × deal_price_i) / Σ(deal_volume_i).
  • Each partial-exit deal must be individually recorded (fields R, X–AA) so the weighted mean is auditable.

11. Cost Representation (ratified Z)

  • Transaction costs are preserved separately and normalized to initial-risk units (ratified Z).
  • Operational form (per verified addendum §4):
cost_R = (spread_points + commission_points + slippage_points + swap_points)
         / initial_risk_points
net R  = gross R − cost_R
  • Each cost term is defined against the broker/tester symbol specification at execution time.
  • Gross R and each cost component (spread, commission, slippage, swap) must be analyzed and reported separately (robustness), with net R as the primary endpoint.
  • Actual numeric cost values (spread/commission/slippage/swap): UNSPECIFIED — HUMAN DECISION REQUIRED (run-configuration, to be ratified at execution planning; protocol guidance: conservative/pessimistic values, per EXP-003 protocol §9).

12. Zone Clustering / Dependence (ratified E, M)

  • Cluster = structural zone (direction, zone_low, zone_high).
  • Multiple round-trips from the same zone are correlated and must never be treated as independent observations (ratified M).
  • Primary CI: cluster bootstrap, 10,000 resamples, clusters = structural zones (ratified E).
  • Zone-level mean net R (equal weight per zone) is reported as a secondary robustness view (per verified addendum §6).
  • Decision assignment: each round-trip inherits the fused score recorded for the candidate that produced its successful order attempt (entry-time score).

13. Dataset Snapshot Requirements (ratified T, U)

A pre-run dataset identity record must be captured before run start and frozen with the run. Required fields (operational form of ratified T/U, per verified addendum §9):

Field Value basis
Dataset path (terminal data folder) actual, at run start
Symbol / timeframe XAUUSD / M15
Model / tick mode Model=1 (1-minute OHLC)
Date range 2026-02-01 → 2026-07-31
Bar count / tick count compare vs C0: 11,639 bars / 696,947 ticks; discrepancy → ABORT (U)
Source identity HFMarketsGlobal-Demo terminal
Snapshot timestamp recorded
Cryptographic hash of preserved dataset-identity artifacts SHA-256
MT5 build 6090
Tester agent identity recorded
Broker/server identity recorded
  • Freeze point: identity captured BEFORE run start; no history download/sync during the run.
  • Exact artifact set to be hashed for the dataset snapshot: UNSPECIFIED — HUMAN DECISION REQUIRED (verified addendum proposed history-quality log + journal excerpt).

14. Binary / Configuration Provenance (ratified AB)

Before any execution, the exact authorized binary and configuration identity must be captured:

Item Requirement
EX5 path recorded
EX5 file size recorded
EX5 SHA-256 recorded (mandatory; C0 precedent never hashed — gap to close)
Build timestamp recorded
Source commit expected bf19250 + ratified addendum
.mqproj state must be resolved or formally recorded (current unstaged modification is a documented pre-existing drift; disposition is a human decision)
.set SHA-256 recorded (C0: 732D2230…)
Compiler/build environment recorded if available
  • Binary mismatch vs provenance → ABORT (ratified §18).
  • No binary exists today that satisfies this requirement for Gate B; no compile is authorized by this document.

15. Manifest and Freeze Requirements (ratified AC)

Adopting the C0 chain pattern (PRESERVED → HASHED → FROZEN) as precedent:

  1. PRESERVED: raw execution evidence copied to EXP003B_RESULTS\GATE_B\RAW\ — run config, binary, dataset identity, logs, tester report, deal/position history, telemetry, CSV, attempt log, statistical specification.
  2. HASHED: SHA-256 manifest of all preserved files.
  3. FROZEN: manifest status set to FROZEN before analysis; after freeze, no modification is permitted; only addition of derived artifacts with hash-stamped provenance.

Manifest must identify: original evidence · derived evidence · execution logs · tester report · deal history · position history · telemetry · configuration · binary · dataset identity · statistical specification · governance metadata.

  • Exact file naming/layout under GATE_B\RAW\: UNSPECIFIED — HUMAN DECISION REQUIRED (folder location per verified addendum §14; file names to be specified at execution planning).
  • Tester report format to preserve (HTML/XML): UNSPECIFIED — HUMAN DECISION REQUIRED.
  • Deal/position history export method: UNSPECIFIED — HUMAN DECISION REQUIRED.

16. Evidence Preservation

  • Preservation must occur before any analysis and before any freeze transition.
  • On abort (§18), immediate RAW copy + hashing per the manifest procedure.
  • Full logs are gitignored and may remain at original locations, but the preserved evidence package must include the mandatory structured records and tester report (C0 precedent: full logs not preserved; excerpts identified as excerpts).
  • No evidence may be deleted, overwritten, normalized, or "cleaned".

17. Print-Only Evidence Is Not Evidence (ratified P, R, AE+7)

  • Print-only records cannot substitute for mandatory structured evidence. Order-attempt, position-open, management, exit, and round-trip events are currently Print-only or absent in the source (verified in gate_b_pre_execution_instrumentation_readiness_audit.md §4/§5). They do not satisfy the schema.
  • Structured records are mandatory for every §3 lifecycle stage; absence is a governance event (§7) — never filled from Print lines, tester summaries, later position state, timestamps, P&L, or assumptions.

18. Stop / Abort Matrix — Evidence Required to Detect Each Condition

All 18 conditions are ratified (record §12, AD) with their actions, evidence preservation, and human escalation. This table adds the evidence required to detect each condition (per verified instrumentation readiness audit §13) — no condition, action, or rule is added or changed.

# Condition Ratified action Evidence required to detect
1 Evidence corruption / hash mismatch ABORT manifest recheck; SHA-256 of all preserved files
2 Missing mandatory event ABORT full schema conformance check (§4/§3)
3 Dataset mismatch ABORT dataset snapshot vs C0 (bars/ticks) at run start
4 Binary mismatch ABORT binary SHA-256/size vs provenance (§14)
5 Unauthorized parameter change ABORT frozen .set vs run config
6 OOS violation ABORT OOS statement (Option A) audit
7 Unexpected live/broker connectivity ABORT runtime terminal-state monitor
8 Order-rejection cascade CONTINUE + FLAG / rate-limit attempt log (retcode surge)
9 Duplicate position / stacking ABORT position records (ticket uniqueness)
10 Impossible position state ABORT deal history vs position records
11 Missing exit / open at end CENSOR + REPORT / CONTINUE position/exit records; end-of-window census
12 Unreconcilable deal history ABORT deal history vs tester report P&L
13 Missing initial SL INVALID OUTCOME / CONTINUE CSV zone.sl + comment (candidate side)
14 Invalid R calculation ABORT numeric audit of entry/SL/exit fields (S, M, Y)
15 Uncontrolled repeated attempts CONTINUE + FLAG / rate-limit attempt log cadence
16 Telemetry failure CONTINUE + FLAG telemetry heartbeat/tick census
17 Tester interruption ABORT tester journal
18 Partial evidence loss ABORT evidence inventory audit
  • Nine conditions (2, 7, 8, 9, 10, 11, 12, 14, 15) are not implementable with current evidence; they become detectable only after separately authorized instrumentation. This protocol does not authorize that.

19. Diagnostic-Only Execution Boundary (ratified V, W)

  • Execution, when eventually authorized, is limited to diagnostic MT5 Strategy Tester execution only: no live account, no VPS/live terminal, no broker order transmission, no real financial risk.
  • InpEnableTrading=true is permissible only within the isolated tester context under a separate, explicit execution authorization.
  • Manual intervention is prohibited except under the defined emergency-abort procedure (kill tester + preserve all evidence).
  • Emergency-abort operational detail beyond "kill tester, preserve evidence": UNSPECIFIED — HUMAN DECISION REQUIRED.

20. Reproducibility (ratified AE)

  • Governance reproducibility: satisfied by the ratified rules (all deterministic and fully specified).
  • Execution reproducibility: achieved only after the record §15 preconditions are completed and audited (protocol artifact ratification; separate instrumentation authorization; evidence-field implementation and verification; binary/config/dataset freeze + hash; dataset snapshot + C0 comparison; binary/.mqproj drift resolution; tester-only isolation verification; separate execution authorization review).
  • This document is precondition #1 only; it does not satisfy the others.

21. Ratified Requirement: B-Class Timing-Risk Acceptance and Batching

  • The proposed evidence instrumentation is classified B — observation with timing risk (per-event file/telemetry I/O) and requires explicit governance acceptance (verified instrumentation readiness audit §6).
  • Management-event batching is mandatory for fields V/W (§5.3).
  • Acceptance of the B-class classification and batching parameters: UNSPECIFIED — HUMAN DECISION REQUIRED (a separate human decision; not granted by this document).

22. Explicit Non-Authorization

This document grants NO authorization. In particular, it does NOT authorize:

  • source modification (.mq5/.mqh);
  • .mqproj modification or disposition;
  • instrumentation implementation (attempt/position/management/exit/round-trip logging);
  • telemetry extension;
  • compilation;
  • tester execution;
  • Gate B execution;
  • parameter optimization or sweep;
  • dataset/OOS modification;
  • oracle rerun;
  • live/VPS/broker execution;
  • Git commit or push.

Ratification of this protocol (AE+8) does not constitute instrumentation authorization (AE+7 separation). Creation of this artifact is not authorization to implement anything.

23. UNSPECIFIED — HUMAN DECISION REQUIRED (complete list)

Item Context
Management-event batching interval/count (V/W) §5.3, §21
Rate-limit values for order-rejection cascade / repeated attempts §18 conditions 8, 15
Actual spread/commission/slippage/swap values for cost model §11
Dataset snapshot hash artifact set §13
GATE_B\RAW\ file naming/layout §15
Tester report format to preserve (HTML/XML) §15
Deal/position history export method §15
Emergency-abort operational procedure detail §19
B-class timing-risk acceptance (explicit human decision) §21
.mqproj disposition (accept / revert metadata / new ratified config) §14
AE+6 power calculation (future, when assumptions known) §2 (record AE+6)

No item on this list is resolved by inference.

24. Consistency Statement

This protocol was drafted by reproducing ratified rules from gate_b_human_ratification_decision_COMPLETED.md and its verified supporting audits (post_ratification_governance_lock_audit.md, gate_b_pre_execution_instrumentation_readiness_audit.md, gate_b_protocol_addendum_audit.md, Step 1–8 artifacts, EXP-003 protocol). No ratified rule was weakened, extended, overridden, or re-interpreted. No new statistical rule, threshold, sample requirement, OOS rule, stop condition, or execution permission was introduced. Open items are explicitly marked UNSPECIFIED — HUMAN DECISION REQUIRED.

25. Final Status Block

EVIDENCE PROTOCOL v0.1 = CREATED (DRAFT — AWAITING HUMAN RATIFICATION)

GOVERNANCE RATIFICATION = READY FOR REVIEW (human disposition of AE+8 required)

INSTRUMENTATION = NOT AUTHORIZED
SOURCE MODIFICATION = NOT AUTHORIZED
COMPILATION = NOT AUTHORIZED
GATE B EXECUTION = NOT AUTHORIZED

RAW EVIDENCE = FROZEN
DATASET = IMMUTABLE
OOS = IMMUTABLE

NO TESTER EXECUTION
NO OPTIMIZATION
NO ORACLE RERUN
NO GIT COMMIT
NO GIT PUSH

STOP.