The Forex candidate intake standard: one ledger, one preflight, zero double-evaluation

The forex twin of the crypto intake standard. Same discipline — one append-only ledger row per candidate, one preflight, a hard rule that no idea is evaluated twice — but forex-scoped and resident in the mql5 repo, where the forex audit folders live. The crypto shared-data pool links to it; it does not host it.

In plain English. Until now, forex feature ideas were tracked the hand-written way: a scoreboard file (STATUS.md), a verdict file (keepers.json), a deferral register (DEFERRALS.md), one per campaign. That works until two campaigns quietly re-test the same idea. This standard gives forex the same single, machine-readable logbook crypto already has — one row per candidate, stamped the moment it is named, that can only gain notes, never lose them — plus a one-command preflight that catches duplicates before any expensive evaluation.

What is this?

A single SSoT ledger — forex_candidate_ledger.jsonl in mql5/findings/shared_data/ — driven by the CLI forex_candidate_ledger.py. Every forex candidate gets exactly one row at identification time and walks a fixed set of named stages, so you can always see at a glance where any idea stands and never re-evaluate one already on file.

identified → pending → evaluating → evaluated → promoting → promoted

Three are terminal — dead ends an idea can land in and never leave:

deferred — set aside (substrate-blocked or parameter-flagged); may return when its blocker clears.
banned — ruled out as redundant (a high-correlation connector dropped to keep its partners).
out_of_scope — not part of what this effort studies.

The standard flow for a NEW forex candidate

Step	What you run / produce	What it guarantees
1	`forex_candidate_ledger.py --check "<name>"`	No double evaluation. Exact + ≥8-char containment match over all names & aliases — seeded with every prior forex batch, so a re-discovery is caught instantly.
2	`--add --num N --name … --family … --audit-folder findings/audits/<campaign>/`	One row at birth — stage `identified`, pointing at its mql5 audit folder.
3	`--preflight "name1,name2,…"`	Per candidate: dedup verdict + ledger stage + audit-folder-on-disk. (Axis-2 parameterless + knob-floor are a forex extension point — wire a forex parameterless probe here when one exists.)
4	the evaluation loop runs → per-cell records	Orthogonality verdicts via Spearman \|ρ\| + Chatterjee ξ across the crossed regime panel; slates of ≥2 cross-check siblings.
5	`--set-stage fxcand-XXXX <stage> --note "…"`	The outcome is written back — the next campaign's Step-1 `--check` sees it and avoids repeating the work.

Stage + status at intake (this discovery campaign)

A pure discovery pass — find → de-dup → provenance → testability — is not an evaluation, so candidates land conservatively:

Input stage = identified (forced by --add).
End-of-discovery stage = pending — row + audit folder + preflight green; loop-ready, not evaluated.
Status = the axis triad. parameterless = inspection/preflight verdict; agnostic and orthogonal stay null (untested) until the evaluation loop runs. Discovery never lands a candidate at evaluated/promoted.

Real-data testability gate (forex substrate)

Every candidate must be computable from real forex bars on bigblack (read-only ClickHouse):

Table: fxview_cache.forex_bars FINAL — ReplacingMergeTree, ouroboros_mode='week', 201 columns.
Columns: the locked FOREX_COLUMNS = 42 populated, non-forward-looking columns (time / OHLC / bid-ask quote / spread / session_label / 9 intra-bar / 10 lookback / active microstructure: realized · bipower · Parkinson · Garman-Klass vol, Amihud, Lee-Ready, VPIN, realized skew/kurt, Lee-Mykland jumps). Forward-looking [LABEL] columns are excluded (leakage). See the archived forex availability probe.
Symbols: EURUSD · GBPUSD · XAUUSD · XAGUSD (expansion path documented in the probe).

Per-candidate exit criteria

#	Requirement
1	Provenance — a research finding with a stable paper ID (arXiv / DOI / SSRN / ISBN).
2	SOTA FOSS implementation — a reference open-source implementation found via web search, with its SPDX license.
3	Git-hash de-dup anchor — the commit hash of the chronicle snapshot the candidate was confirmed novel against.
4	Real-data testability — the exact `FOREX_COLUMNS` / fields it is computed from.
5	Parameterless flag — parameterless-by-inspection true/false (preferred, not required).

Seeded from prior art

The ledger is bootstrapped with every forex candidate already evaluated or queued across the 2026-06-03 discovery and 2026-06-07 implementation campaigns (batches 1–10) — so the new brute-force sweep --checks against real history, not a blank slate.

Ledger row per candidate

Steps in the flow

6 + 3

Normal + terminal stages

FOREX_COLUMNS substrate

Relationship to crypto

This ledger is separate from the crypto candidate_ledger.jsonl (BTCUSD-only). The forex shared-data layer was archived from active scope on 2026-05-30; this standard reintroduces forex with its own ledger in mql5 rather than reversing the crypto-only decision in opendeviationbar-py. The two tracks run in parallel and share only the pattern, not the ledger file.

Bottom line. Forex now has the same one-ledger / one-preflight / zero-double-evaluation discipline as crypto — append-only, machine-readable, seeded with prior art, and resident where forex actually lives (mql5).

Shared Data Pool · forex finding · source: mql5/findings/shared_data/forex-candidate-intake-standard.md · ← back to the pool · 2026-06-16