This folder sets the rules for an automated hunt for one hundred brand-new candidate measurements that carry no hand-tuned settings and come with real open-source code, and the shared candidate registry now holds exactly one hundred entries filed under this hunt — but the folder itself never recorded a conclusion and none of the per-candidate paperwork it promised to keep here was ever written.
Lifecycle, not result. This says where the audit sits in its process — never whether what it found was good.
**Status:** DISCOVERY SWEEP (autonomous `/loop`, ~25 min cadence) · **Operator:** nasimubd
2026-08-17. Derived from folder evidence; the adversarial challenge pass CORRECTED the first reading. Quote verified verbatim at CLAUDE.md:3, but the proposal read the header's '~25 min cadence' as evidence the loop is still running — it is not. Three independent refutations. (1) The stop rule fired: CLAUDE.md:6 states 'Stop at 100 **or** after 2 dry iterations.' and candidate_ledger.jsonl now carries exactly 100 rows citing this audit_folder, ids cand-0001..cand-0100 (98 pending, 2 promoted). Tar
This audit has no separate verdict document. What it has is the folder's own record, reproduced in full below — presented as that and not dressed up as a conclusion it never wrote. Links inside it that point at unpublished files are shown as plain text rather than as links that would 404 here.
Status: DISCOVERY SWEEP (autonomous /loop, ~25 min cadence) · Operator: nasimubd Goal: accumulate 100 net-new, unique, pending, parameterless-by-inspection candidate features in the registry, each with a real FOSS implementation + pinned git SHA and a cited idea source, ready for evaluation by the legacy orthogonality probe. Stop at 100 or after 2 dry iterations.
> Discovery + registration ONLY — no ClickHouse, no orthogonality evaluation here (that is the > separate evaluation loop). This folder is the provenance home; the registry is the SSoT.
cand-NNNN + content-addressed dedup.spec_hash); never reused.provenance.source).provenance.foss_repo + provenance.foss_commit_sha (the FOSS is the kernel source that makes the candidate evaluation-ready later).order=/window=/q-grid/ scale schedule); closed-form or fully-derived only.unique per candidate_ledger.py --novelty (not already in the 44 + catalog).Written here as prov/<cand-id>.json:
{"id": "cand-00NN", "name": "...", "family": "...",
"source": "arXiv:.... / DOI ....", "foss_repo": "https://github.com/...", "foss_commit_sha": "<sha>",
"parameterless_justification": "closed-form; no tunable knobs because ...",
"novelty": {"status": "unique", "checked_against": "registry+catalog"},
"asset_origin": "crypto | forex (idea source)"}The structured triple (source + foss_repo + foss_commit_sha) is ALSO stored on the ledger row via --add --source --foss-repo --foss-sha, so the registry is queryable; this folder holds the fuller record.
python3 findings/evolution/shared_data/candidate_ledger.py --add \
--name "NAME" --family FAMILY \
--audit-folder findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/ \
--source "DOI/arXiv/URL" --foss-repo "REPO_URL" --foss-sha "COMMIT_SHA" \
--note "sweep=disc100; parameterless-by-inspection" \
&& python3 findings/evolution/shared_data/candidate_ledger.py --set-stage <id> pendingfindings/evolution/shared_data/candidate_ledger.jsonl (+ candidate_ledger.py)findings/evolution/shared_data/novelty_check.pyfindings/dashboard/spokes/2026-06-17-crypto-forex-parameterless-discovery-100/findings/dashboard/crypto-candidates/Every row pairs a claim with the file it came from and the verbatim text in that file. The sources sit above the deploy root, so the quote is embedded and the path is printed as text rather than linked — a link would resolve on a laptop and 404 here.
| Claim | Evidence |
|---|---|
| The folder is a contract for an autonomous discovery loop with an explicit numeric target and an explicit stop rule; it is not itself an evaluation. ASSERTED target 100 candidates; loop cadence ~25 minutes; stop rule = 100 reached or 2 dry iterations — declared plan, not measured results | **Goal:** accumulate **100 net-new, unique, pending, parameterless-by-inspection** candidate features in the registry, each with a real **FOSS implementation + pinned git SHA** and a cited idea source, **ready for evaluation** by the legacy orthogonality probe. Stop at 100 **or** after 2 dry iterations. findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/CLAUDE.md |
| Discovery and registration are the whole scope: no database access and no orthogonality evaluation happen in this folder by design. ASSERTED | > Discovery + registration ONLY — no ClickHouse, no orthogonality evaluation here (that is the > separate evaluation loop). This folder is the **provenance home**; the registry is the SSoT. findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/CLAUDE.md |
| Every registered candidate must carry five things: a minted stable id, a cited idea source, an open-source implementation pinned to a commit hash, a written justification that it has no tunable settings, and a novelty check against the existing catalogue. ASSERTED 5 mandatory contract fields per candidate | 3. **FOSS implementation + pinned git SHA** — `provenance.foss_repo` + `provenance.foss_commit_sha` (the FOSS *is* the kernel source that makes the candidate evaluation-ready later). 4. **Parameterless-by-inspection justification** — no tunable knobs (no `order=`/`window=`/`q`-grid/ scale schedule); closed-form or fully-derived only. findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/CLAUDE.md |
| The sweep did fire and reached its target exactly: the shared candidate registry holds one hundred rows attributed to this audit folder, the great majority of the whole registry. MEASURED 100 of 110 total ledger rows carry this audit_folder (n=110 rows); the other 10 come from two different audits; ids cand-0001 to cand-0100 spanning 31 families | "audit_folder": "findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/" findings/evolution/shared_data/candidate_ledger.jsonl |
| The registered cohort mostly satisfies the folder's own provenance contract but not entirely: every candidate has a cited idea source while six lack the required open-source repository and pinned commit hash. MEASURED 100/100 rows carry provenance.source; 94/100 carry both foss_repo and foss_commit_sha; 6/100 missing the pinned implementation (n=100) | "provenance": {"source": "J. R. M. Hosking (1990), \"L-Moments: Analysis and Estimation of Distributions Using Linear Combinations of Order Statistics\" findings/evolution/shared_data/candidate_ledger.jsonl |
| The novelty requirement holds across the cohort: every one of the hundred candidates is recorded as unique against the registry and catalogue. MEASURED 100 of 100 rows have dedup.status = unique (n=100) | "dedup": {"status": "unique", "mechanism": "none", "score": null findings/evolution/shared_data/candidate_ledger.jsonl |
| The cohort has largely not been consumed yet: nearly all of it still sits at the pending stage with no orthogonality verdict recorded. MEASURED 98 of 100 at stage=pending, 2 promoted; axis.orthogonal recorded for only 12 of 100 (5 BAN, 4 WATCH, 2 PASS, 1 PENDING, 88 null); axis.parameterless recorded for only 2 of 100 | "id": "cand-0100", "num": 100, "name": "Fraction of intrabar time above the open price", "aliases": [], "family": "occupation_arcsine", "stage": "pending" findings/evolution/shared_data/candidate_ledger.jsonl |
| The per-candidate provenance records the contract says live in this folder were never written — the folder contains a single file and no verdict. REFUTED 0 prov/*.json files present; folder contains 1 file in total (CLAUDE.md, 2,881 bytes); 0 verdict.md | ## Provenance record per candidate Written here as `prov/<cand-id>.json`: findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/CLAUDE.md |
| The folder's lifecycle state is contested: the dashboard twin still records the sweep as never having fired, which the registry contradicts, and an operator ruling later declared the campaign settled rather than re-deriving it from the files. OPEN dashboard manifest tally 0 of target 100 with 1 iteration page (iter 00); registry shows 100 registered; operator ruling dated 2026-08-14; prior recorded status READY | "_status_evidence_superseded": "Contract exists, discovery never fired. manifest 'target: 100, tally: 0'; the single row is 'index_iter_00_sweep_initialized.html, tally 0, registered 0'. findings/dashboard/campaigns/2026-06-17-crypto-forex-parameterless-discovery-100/LOOP_LEDGER.json |
| The registry, not this folder, is named as the single source of truth, with a specific registration command the loop must run verbatim. CONFIRMED | - Registry SSoT: `findings/evolution/shared_data/candidate_ledger.jsonl` (+ `candidate_ledger.py`) - Novelty gate: `findings/evolution/shared_data/novelty_check.py` findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/CLAUDE.md |
Source of record: findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/ — not published, so these are listed rather than linked.
| File | Role |
|---|---|
CLAUDE.md | The only file in the folder: declares the discovery contract (target 100, stop rule, five per-candidate requirements, provenance-record schema, verbatim registration command, cross-references). No results, no conclusion. |
findings/dashboard/campaigns/2026-06-17-crypto-forex-parameterless-discovery-100/
findings/dashboard/build_audits.py from findings/evolution/audits/2026-06-17-crypto-forex-parameterless-discovery/AUDIT_LEDGER.json — never hand-edited. Each quote was verified to occur in the file named beside it when the ledger was written.