Ficta Meridian Group
A diversified financial services group serving corporates, institutions and private clients worldwide — supported by a single operations and control platform.
Our role in society
We believe finance can be a positive force. Across our business, colleagues are helping clients navigate the transition, strengthening communities and making considered choices for the long term.
One operations platform
Four jurisdictions · under administration
Same-day rate
Strategic advisory, capital raising and financing solutions for corporates and institutions.
Sales and trading across equities, rates, credit and foreign exchange, including Delta One.
Cash management, trade finance, clearing and custody for institutional clients.
Comprehensive wealth planning and discretionary management through Ficta Meridian PB.
Institutional strategies across traditional and alternative asset classes.
The unit behind every trade: booking, allocation, reconciliation, collateral, settlement and release.
Senior management is accountable for the group's operating platform and control framework.
Reports to Z. Quarry
Responsible for technology across the operations platform, including the systems that book, margin and settle every trade.
Reports to T. Pellow
Oversees market and credit risk for the trading businesses — limits, exposures and counterparty review.
Reports to M. Marchant
Responsible for middle office control across the group: trade support, confirmations and the daily control framework.
Institutional decomposition // technical buyer entry
This inspection surface connects 42 released coverage primitives to 69 source-grounded Episodes, task specifications, adjudication rubrics and measured recovery probes.
Public: canonical definitions, schemas and adjudication rubrics. Private: occurrence instances and the Record plane. Examples below are Cast-only synthetic counterparts.
BOCG v0.6.2 · Operating reference 1.0.0
Explore the work a desk performs, the public provisions relevant to that work, and what the current evaluation can support.
Shared definitions from the released model-consensus grid. Named institutional units, reporting lines, books, systems and local settings are held in Finexhaust's private reference register. The released catalogue contains 0 curated control-point definitions; proposed cells do not count as curated.
42 divisions match
Current capability surface · rubric projected onto BOCG
Each tile is one released coverage primitive of BOCG v0.6.2. Tile colour advances only on ratified evidence: from vocabulary, to candidate mappings (shown, never counted), to owner-ratified Episode evidence, to a ratified task, to a measured model deficit on that task.
Candidate mappings come from lexicon and ontology proposals and are displayed as candidates only. Nothing below an owner or SME ruling counts as coverage, supports an SKU, or authorises an eval or rubric build. Counts are partitioned by source profile so that no institution's evidence is silently pooled with another's.
Evidence ledger: BOCG v0.3.0. Current catalogue: v0.6.2. Compatibility is verified from identical coverage-matrix bytes; past rulings retain their original release binding.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface4 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Candidates by source profile (not counted)2
Candidate receipts (shown, not counted)ep-d1f7a4178d3flexicalep-c7ecccfb6c55lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 1 ratified Episode mapping → 0 ratified tasks → 0 measured
Released binding surface7 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Ratified by source profile1
Candidates by source profile (not counted)7
Rulings preserved1 rejected · 0 deferred
Deferred and rejected receiptsep-5b36c3cf4c3frejected
Ratified Episode receiptsep-35f4f3c650dbqualifiedno released binding yet
Candidate receipts (shown, not counted)ep-e4ef3f320de2lexicalep-f85b563dc25flexicalep-d1f7a4178d3flexicalep-a0422483cf7blexicalep-803acc3afcf0lexicalep-cdcf1fcafb73lexicalep-c7ecccfb6c55lexical
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface4 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Rulings preserved0 rejected · 1 deferred
Deferred and rejected receiptsep-809f543d075adeferred
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface4 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Candidates by source profile (not counted)5
Rulings preserved6 rejected · 0 deferred
Deferred and rejected receiptsep-879ead7f9542rejectedep-492944c0a778rejectedep-fac89cbcbdc2rejectedep-1a99aa27e757rejectedep-7d7d85832429rejectedep-b187a2047913rejected
Candidate receipts (shown, not counted)ep-c721e4aec245lexicalep-10dd9254f2a2lexicalep-a1a164af56f3lexicalep-38623105fc33lexicalep-47ea021cd3a7lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface24 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 0.00 · DeepSeek 0.00 · Moonshot 0.67 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)9
Rulings preserved3 rejected · 1 deferred
Deferred and rejected receiptsep-0e54b9240435rejectedep-35f4f3c650dbrejectedep-b0d631c496d8rejectedep-55cae80f0b6cdeferred
Candidate receipts (shown, not counted)ep-a5e133bf3e3blexicalep-4d2102bdd658lexicalep-c721e4aec245lexicalep-e4ef3f320de2lexicalep-54e572ccb554lexicalep-b4259da48cc8lexicalep-f9e417f355e0lexicalep-e3c4f8f9b456lexicalep-535ed66e02c5lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface39 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 1.00 · DeepSeek 0.00 · Moonshot 1.00 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)6
Candidate receipts (shown, not counted)ep-9a4cdf0acb6clexicalep-e4ef3f320de2lexicalep-d1f7a4178d3flexicalep-b4259da48cc8lexicalep-bae4d9434967lexicalep-e3c4f8f9b456lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface16 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.67 · Anthropic 0.33 · DeepSeek 0.00 · Moonshot 0.33 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)2
Candidate receipts (shown, not counted)ep-3205a4c9a0a2lexicalep-38623105fc33lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface4 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Candidates by source profile (not counted)5
Rulings preserved1 rejected · 0 deferred
Deferred and rejected receiptsep-0e54b9240435rejected
Candidate receipts (shown, not counted)ep-b4259da48cc8lexicalep-5f3dcc195bbclexicalep-44d4930b0ea6lexicalep-1d324296e084lexicalep-47ea021cd3a7lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 1 ratified Episode mapping → 0 ratified tasks → 0 measured
Released binding surface16 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 1.00 · DeepSeek 0.00 · Moonshot 0.33 · Qwen 0.00 · weak tier
Ratified by source profile1
Rulings preserved0 rejected · 1 deferred
Deferred and rejected receiptsep-55cae80f0b6cdeferred
Ratified Episode receiptsep-6e7926c8c378qualifiedno released binding yet
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface4 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface40 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 0.67 · DeepSeek 1.00 · Moonshot 1.00 · Qwen 0.00 · moderate tier
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 1 ratified Episode mapping → 1 ratified task → 0 measured
Released binding surface44 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 1.00 · DeepSeek 1.00 · Moonshot 1.00 · Qwen 0.00 · moderate tier
Ratified by source profile1
Candidates by source profile (not counted)1
Rulings preserved1 rejected · 0 deferred
Deferred and rejected receiptsep-b187a2047913rejected
Ratified Episode receiptsep-879ead7f9542qualifiedprice a cash bond or credit derivative
Candidate receipts (shown, not counted)ep-a0422483cf7blexical
PROPOSED-SKU-401 · ratified tasktp_aa36bf63cadc
SKU candidatesPROPOSED-SKU-401
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface9 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.33 · Anthropic 0.00 · DeepSeek 0.67 · Moonshot 0.00 · Qwen 0.00 · weak tier
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface8 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 0.67 · DeepSeek 0.00 · Moonshot 0.00 · Qwen 0.00 · weak tier
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface21 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.33 · Anthropic 0.67 · DeepSeek 0.67 · Moonshot 0.33 · Qwen 0.00 · weak tier
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface17 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.33 · Anthropic 0.33 · DeepSeek 0.67 · Moonshot 0.33 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)3
Rulings preserved2 rejected · 0 deferred
Deferred and rejected receiptsep-eb0566f0899drejectedep-92bcff13b5c8rejected
Candidate receipts (shown, not counted)ep-3205a4c9a0a2lexicalep-a1a164af56f3lexicalep-38623105fc33lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface10 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Candidates by source profile (not counted)1
Candidate receipts (shown, not counted)ep-e3c4f8f9b456lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface18 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 0.00 · DeepSeek 0.67 · Moonshot 1.00 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)1
Candidate receipts (shown, not counted)ep-a1a164af56f3lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface50 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 1.00 · DeepSeek 1.00 · Moonshot 1.00 · Qwen 0.67 · moderate tier
Candidates by source profile (not counted)2
Rulings preserved0 rejected · 1 deferred
Deferred and rejected receiptsep-809f543d075adeferred
Candidate receipts (shown, not counted)ep-10dd9254f2a2lexicalep-47ea021cd3a7lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface24 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 1.00 · DeepSeek 0.00 · Moonshot 1.00 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)1
Candidate receipts (shown, not counted)ep-3205a4c9a0a2lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface6 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 0.00 · DeepSeek 0.00 · Moonshot 0.00 · Qwen 0.67 · weak tier
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface39 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.33 · Anthropic 1.00 · DeepSeek 0.67 · Moonshot 1.00 · Qwen 0.33 · weak tier
Candidates by source profile (not counted)11
Rulings preserved3 rejected · 0 deferred
Deferred and rejected receiptsep-73d0298ebcf8rejectedep-6e7926c8c378rejectedep-26df4ea7ff06rejected
Candidate receipts (shown, not counted)ep-9a4cdf0acb6clexicalep-4d2102bdd658lexicalep-c721e4aec245lexicalep-2aff03025f42lexicalep-5f3dcc195bbclexicalep-c78bff3c8433lexicalep-415fd55f4047lexicalep-535ed66e02c5lexicalep-cdcf1fcafb73lexicalep-54e12622c53dlexicalep-8aa0728ace30lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface6 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 0.00 · DeepSeek 0.67 · Moonshot 0.00 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)2
Candidate receipts (shown, not counted)ep-803acc3afcf0lexicalep-1d324296e084lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface4 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface24 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 1.00 · DeepSeek 0.00 · Moonshot 1.00 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)1
Rulings preserved1 rejected · 0 deferred
Deferred and rejected receiptsep-a8bc8e087a39rejected
Candidate receipts (shown, not counted)ep-1d324296e084lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface3 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface12 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.67 · Anthropic 0.00 · DeepSeek 0.33 · Moonshot 0.00 · Qwen 0.33 · weak tier
Candidates by source profile (not counted)2
Rulings preserved1 rejected · 0 deferred
Deferred and rejected receiptsep-a8bc8e087a39rejected
Candidate receipts (shown, not counted)ep-c78bff3c8433lexicalep-535ed66e02c5lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 1 ratified Episode mapping → 1 ratified task → 0 measured
Released binding surface16 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 1.00 · DeepSeek 0.00 · Moonshot 0.33 · Qwen 0.00 · weak tier
Ratified by source profile1
Rulings preserved5 rejected · 0 deferred
Deferred and rejected receiptsep-492944c0a778rejectedep-a8bc8e087a39rejectedep-1a99aa27e757rejectedep-7d7d85832429rejectedep-b187a2047913rejected
Ratified Episode receiptsep-f304ed77d58dqualifiedproduce daily var expected shortfall using approved model
PROPOSED-SKU-402 · ratified tasktp_867ed76cefec
SKU candidatesPROPOSED-SKU-402
Evidence chainBOCG v0.6.2 → 5 ratified Episode mappings → 5 ratified tasks → 0 measured
Released binding surface12 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 0.00 · DeepSeek 0.00 · Moonshot 0.00 · Qwen 0.00 · weak tier
Ratified by source profile5
Candidates by source profile (not counted)1
Rulings preserved2 rejected · 6 deferred
Deferred and rejected receiptsep-f304ed77d58drejectedep-7d7d85832429rejectedep-7ce3a57c84e3deferredsource correction pendingep-2642f0c6f012deferredsource correction pendingep-30e5f94401ebdeferredsource correction pendingep-094febc7b9eedeferredsource correction pendingep-0aa38bfb1452deferredep-8f870687ba6adeferred
Ratified Episode receiptsep-492944c0a778qualifiedgate model change before production deploymentep-1a99aa27e757qualifiedgate model change before production deploymentep-a8bc8e087a39qualifiedgate model change before production deploymentep-fac89cbcbdc2qualifiedcalculate daily market risk measureep-b187a2047913qualifiedcalculate daily market risk measure
Candidate receipts (shown, not counted)ep-44d4930b0ea6lexical
PROPOSED-SKU-403 · ratified tasktp_ac52146bbcc3
PROPOSED-SKU-403 · ratified tasktp_6f344aca5dd4
PROPOSED-SKU-403 · ratified tasktp_eae00841210b
PROPOSED-SKU-402 · ratified tasktp_f8352b9cd454
PROPOSED-SKU-402 · ratified tasktp_45c5020b8c7f
SKU candidatesPROPOSED-SKU-402 · PROPOSED-SKU-403
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface4 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Rulings preserved3 rejected · 0 deferred
Deferred and rejected receiptsep-9324ce618d7crejectedep-879ead7f9542rejectedep-fac89cbcbdc2rejected
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface3 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Candidates by source profile (not counted)1
Candidate receipts (shown, not counted)ep-803acc3afcf0lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 2 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface54 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 1.00 · DeepSeek 1.00 · Moonshot 1.00 · Qwen 1.00 · moderate tier
Ratified by source profile2
Rulings preserved0 rejected · 2 deferred
Deferred and rejected receiptsep-55cae80f0b6cdeferredep-809f543d075adeferred
Ratified Episode receiptsep-6e7926c8c378qualifiedno released binding yetep-26df4ea7ff06qualifiedno released binding yet
Audit only: measured on tasks not yet ratifiedFD-1197 BARE 19/20, DOCTRINE 20/20, RAW 16/20 n/a · FD-2833 UNION_PACK_FIRST minus RAG_2X 0.2 · FD-2833 UNION_PACK_FIRST minus WRONG_LOCUS_UNION 0.3056 · FD-2833 DOCTRINE minus BARE (descriptive) 0.454
SKU candidatesFD-1197 · FD-2833
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface16 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 1.00 · DeepSeek 0.00 · Moonshot 0.33 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)2
Candidate receipts (shown, not counted)ep-54e572ccb554lexicalep-44d4930b0ea6lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface45 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.67 · Anthropic 1.00 · DeepSeek 1.00 · Moonshot 1.00 · Qwen 0.33 · moderate tier
Candidates by source profile (not counted)8
Rulings preserved2 rejected · 0 deferred
Deferred and rejected receiptsep-73d0298ebcf8rejectedep-26df4ea7ff06rejected
Candidate receipts (shown, not counted)ep-9a4cdf0acb6clexicalep-4d2102bdd658lexicalep-2aff03025f42lexicalep-5f3dcc195bbclexicalep-415fd55f4047lexicalep-cdcf1fcafb73lexicalep-54e12622c53dlexicalep-8aa0728ace30lexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 2 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface12 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 1.00 · DeepSeek 0.00 · Moonshot 0.00 · Qwen 0.00 · weak tier
Ratified by source profile2
Candidates by source profile (not counted)1
Rulings preserved1 rejected · 0 deferred
Deferred and rejected receiptsep-9324ce618d7crejected
Ratified Episode receiptsep-eb0566f0899dqualifiedno released binding yetep-92bcff13b5c8qualifiedno released binding yet
Candidate receipts (shown, not counted)ep-10dd9254f2a2lexical
Evidence chainBOCG v0.6.2 → 2 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface21 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.33 · Anthropic 0.67 · DeepSeek 0.00 · Moonshot 0.67 · Qwen 0.00 · weak tier
Ratified by source profile2
Ratified Episode receiptsep-73d0298ebcf8qualifiedno released binding yetep-b0d631c496d8qualifiedno released binding yet
Audit only: measured on tasks not yet ratifiedFD-2833 UNION_PACK_FIRST minus RAG_2X 0.2 · FD-2833 UNION_PACK_FIRST minus WRONG_LOCUS_UNION 0.3056 · FD-2833 DOCTRINE minus BARE (descriptive) 0.454
SKU candidatesFD-2833
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface42 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 1.00 · Anthropic 1.00 · DeepSeek 1.00 · Moonshot 1.00 · Qwen 0.00 · moderate tier
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface13 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.67 · Anthropic 0.33 · DeepSeek 0.33 · Moonshot 0.00 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)1
Candidate receipts (shown, not counted)ep-a5e133bf3e3blexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 6 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface30 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.67 · Anthropic 0.67 · DeepSeek 0.33 · Moonshot 0.33 · Qwen 0.67 · weak tier
Ratified by source profile6
Candidates by source profile (not counted)2
Rulings preserved2 rejected · 1 deferred
Deferred and rejected receiptsep-5b36c3cf4c3frejectedep-708e2317ddc4rejectedsource correction pendingep-b323a337107edeferred
Ratified Episode receiptsep-0e54b9240435qualifiedno released binding yetep-35f4f3c650dbqualifiedno released binding yetep-23c427505a14ratifiedno released binding yetep-9324ce618d7cqualifiedno released binding yetep-eb0566f0899dqualifiedno released binding yetep-92bcff13b5c8qualifiedno released binding yet
Candidate receipts (shown, not counted)ep-6eaffba8c160lexicalep-f85b563dc25flexical
Audit only: measured on tasks not yet ratifiedFD-1197 BARE 19/20, DOCTRINE 20/20, RAW 16/20 n/a · FD-1162 UNION_PACK_FIRST minus RAG_2X 0.0417 · FD-1162 UNION_PACK_FIRST minus WRONG_LOCUS_UNION 0.1319 · FD-1162 DOCTRINE minus BARE (descriptive) 0.454
SKU candidatesFD-1162 · FD-1197
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface8 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 0.67 · DeepSeek 0.00 · Moonshot 0.00 · Qwen 0.00 · weak tier
Current gapNo owner-ratified Episode mapping; candidates never count.
Evidence chainBOCG v0.6.2 → 1 ratified Episode mapping → 0 ratified tasks → 0 measured
Released binding surface3 terminality tasks in the release; no curated control point in v0.6.2
Model consensusNot named in the admitted consensus board
Ratified by source profile1
Candidates by source profile (not counted)5
Ratified Episode receiptsep-b0d631c496d8qualifiedno released binding yet
Candidate receipts (shown, not counted)ep-a0422483cf7blexicalep-e40cd14e85e5lexicalep-415fd55f4047lexicalep-8aa0728ace30lexicalep-c7ecccfb6c55lexical
Evidence chainBOCG v0.6.2 → 0 ratified Episode mappings → 0 ratified tasks → 0 measured
Released binding surface30 terminality tasks in the release; no curated control point in v0.6.2
Model consensusOpenAI 0.00 · Anthropic 1.00 · DeepSeek 0.33 · Moonshot 1.00 · Qwen 0.00 · weak tier
Candidates by source profile (not counted)3
Candidate receipts (shown, not counted)ep-a5e133bf3e3blexicalep-d0354e74a3e8lexicalep-96ca6675a8fclexical
Current gapNo owner-ratified Episode mapping; candidates never count.
Retained without a significance mappingep-2ed4ea44c91aep-db380fb691d7ep-e14eca188e21ep-cba54510857cep-91ceb9ec0173ep-b357cb3d6de8ep-3ed599d35542ep-0fbe103df718ep-3d9c431d59c9ep-59bb6ef64466ep-8ca8759137efep-e1bd3f52c7fbep-4b049d4bf964ep-4f7ab4a077ddep-6aad2703b159ep-57cb0cbafe0bep-7d7d85832429ep-11d122264e3f
Select any tile for its combined evidence and model-consensus receipt. Vendor support counts a model family when its valid-sample fraction is at least 0.50. Prompt efd90b74 · aliases 61025b9f · 2026-08-26. Coverage semantics: candidates are displayed, never counted as evidence; only owner or SME ratified ACCEPT/QUALIFY mappings are evidence coverage; rejected and deferred preserved; measured deficit only on ratified tasks with receipts. Explicit status: no model training claim.
Public-eval overlay · SPEC-09
The same 42 admitted divisions, painted only by third-party public rubrics, benchmarks and evals. Each mapping carries a likelihood band (low, medium, high), a named matching rule, a rationale and a dated ruling. Saturation is the sum of band weights under a division; a division with none is a void.
This is a supplement, not a validation of the grid, and a void is not a claim that the work is under-served. BOCG's own control-point cells, their citations and its own rubrics never paint: they are listed below for audit and excluded by rule.
| Public artifact | Class | Ruling | Rule · band | Rationale |
|---|---|---|---|---|
| Harvey LAB task format (documents + task.json match-criteria rubrics) | peer task shape | qualifiedpaints | R-PEER-SHAPEL1 · Division · Trade lifecycle operations | CONFORMANCE.md records two banking tasks accepted in the Harvey LAB schema. Public peer task-shape only; content loci are not republished here, so mapping stays division-grain L1. This is not BOCG cell coverage.ruled 2026-09-25 · owner |
| BOCG episode-feasibility rubric (own artifact; audit only) | own rubricown artifact | deferreddoes not paint | R-REJECT | In-repo BOCG episode-feasibility rubric. Own artifact; excluded from public-eval saturation by channel=own_artifact.no ruling · audit listing only |
| BOCG five-families adjudication rubric (own artifact; audit only) | own rubricown artifact | deferreddoes not paint | R-REJECT | In-repo BOCG rubric. Own coverage/adjudication artifact; excluded from public-eval saturation by channel=own_artifact.no ruling · audit listing only |
| BetterBench benchmark-assessment criteria (46 criteria) | peer framework | rejecteddoes not paint | R-PEER-SHAPE | BetterBench assesses benchmark-design criteria. It is not a capital-markets operational eval and does not exercise an admitted BOCG division.ruled 2026-09-25 · owner |
| Datasheets for Datasets (Gebru et al., v8) | peer framework | rejecteddoes not paint | R-PEER-SHAPE | Datasheets for Datasets documents datasets; it is not a public banking-ops rubric, benchmark or eval on the BOCG surface.ruled 2026-09-25 · owner |
| Inspect evaluation framework (UK AI Security Institute) | peer framework | rejecteddoes not paint | R-PEER-SHAPE | Inspect is an evaluation harness. Without a published banking-ops task bound to an admitted division key, it does not paint public-eval saturation.ruled 2026-09-25 · owner |
Read from BOCG v0.6.2 (manifest 67f491ec); the map is bound to task concepts 92f823ba, first cut at v0.6.2 and carried unchanged. Before this page was generated the overlay was re-checked here, not taken on trust: S9-G1 schema, inventory digest, task-concept digest, paint rule, roll-up recomputed. Normative source: specs/SPEC-09-public-eval-surface-overlay.md. Generated 2026-10-16.
Status
Nine of ten release gates pass. The tenth does not: no anchor in this run has been checked by hand against an actual filing, rulebook or published series, so every citation underneath the board is a model's assertion and nothing more. That is stated here rather than discovered later.
What is released: the grid's semantic contract — v0.3.0, a content-addressed manifest (sha dce3dd85) binding the specs, vocabulary and this run's artifacts. The vocabulary downstream systems consume is pinned to that release; the elicitation itself stays provisional until the anchor gate closes.
G1ContaminationPASS
G2SchemaPASS
G3Anchor orderingPASS
G4Rejections presentPASS
G5Admission recomputePASS
G6Anchor corroborationFAIL
G7Alias tablePASS
G8No self-assessmentPASS
G9ReproducibilityPASS
G10Panel widthPASS
02 · The panel, and what was kept off it
Two samples emitted the division name before its anchors — the ordering rule exists to catch exactly that.
All three samples failed the same ordering rule: named first, justified afterwards.
Transport, not judgement — one truncation at the token ceiling, two empty streams from the gateway.
A model cannot be dropped quietly: the panel gate refuses to ignore one unless it is declared with a written reason, and the excluded models' raw responses are published with the rest. Two of the three failed on merit — they named a division first and produced its justification afterwards, which is the backwards-reasoning the ordering rule exists to catch.
03 · Operations desk
Capture through confirmation, every state transition receipted.
Block-to-account splits under allocation contracts.
Ledger-to-ledger breaks investigated to closure.
Calls issued, disputed, resolved against agreements.
Fails chased across depots and deadlines.
Four-eyes approvals; nothing self-released.
Every action ordered against what was knowable when.
Claims stand only on retrievable, versioned evidence.
04 · Why Meridian exists
Frontier AI agents are being sent into operational finance. Whether they act correctly — complete the work, respect authorization, refuse shortcuts, touch nothing outside contract — has to be measured somewhere consequences can't reach.
Meridian is that somewhere. The Lab holds the instruments.
06 · From episodes to tasks to uplift
The grid says where a bank spends judgement. Meridian's curated record says what actually happened on such a desk: bounded episodes, each hash-sealed to its sources and adjudicated one human ruling at a time. Every episode below is the synthetic twin: names, units and codes are Meridian's own fiction; the curation ledger holds nothing but node ids and published grid vocabulary.
Each episode is machine-matched into coverage-grid divisions using only the released lexicon (v0.3.0, manifest dce3dd85). Where an episode's divisions overlap a doctrine task family, an evaluation task can be proposed, and a task reaches the build queue only through a ratification gate that scores relevance and consequence before a line of it is written.
The last link is uplift: whether doctrine distilled from such episodes makes a model's work product better, and against what. That is measured under preregistered, hash-locked protocols on a private calibration corpus (one institution, not released). The ledger further down records where each line stands as of 2026-09-07, including the lines that failed.
Bounded, source-hashed, adjudicated one ruling at a time
Matched against the released grid lexicon - every term public, every proposal awaiting a human ruling
Of 42 in the released vocabulary
Division overlap with a priced, consequence-scored task spec
| Episode | Proposed divisions |
|---|---|
| Mato Do reported a completed Ficta Meridian Unit 79EA079E-scoped production update to CORBURY … | Trade lifecycle operations · Buy-side portfolio management · Client lifecycle & KYC |
| Dynamic-margin setup rule reminder and script-readiness confirmation request | Buy-side portfolio management · Sell-side research · XVA & counterparty risk |
| 2024-09-19 · CORBURY Swaps PROD- Basket Manager Manager- GUI DB state issue | Buy-side trading & execution · FX trading · Rates trading |
| 2024-09-06 · IA Requirement Error and Audit Gap | Buy-side performance & risk reporting · Client lifecycle & KYC · Equities trading |
| 2024-08-28 · Swap Trades with 0% IA rate | Rates trading · Client lifecycle & KYC · FX trading |
| 2024-08-27 · brisk-granite Script Migration | XVA & counterparty risk |
| 2024-08-23 · Swaps report migration | FX trading · Rates trading |
| 2024-08-21 · GETS Fengate - Spanish exchange update from XFAL to XFAJ | Buy-side portfolio management · FX trading · Macro (rates/FX/EM combined) |
| 2024-08-08 · Spanish issues in Prod | FX trading · Macro (rates/FX/EM combined) |
Trade lifecycle operations · Prime brokerage & client financing. Framing: concept-layer, saturated unaided. No product uplift gap published for this calibration-lane entry.
Prime brokerage & client financing · Repo & securities financing. Framing: curated evidence set, per-locus signal, pooled unconfirmed. Measured (labelled, not a recovery_gap): scope (T6-v2)=per-locus FD-2833; descriptive contrasts, not a preregistered gate pass; C3b-r_DOCTRINE_minus_RAG_MATCHED (T6-v2)=0.35; C4f_DOCTRINE_minus_QUOTES_FIXED (T6-v2)=0.192; pooled_status (T6-v2)=unconfirmed (T6 canonical claim NOT passed; U1 p=0.061).
Trade lifecycle operations. Framing: retired - retrieval beats pack on this locus (T6-v2). Retired: T6-v2 per-locus FD-1162: pack loses to equal-length question-keyed RAG and adds little over the pack's own fixed quotes; readiness directive was to retire this entry outright (retrieval beats its pack). | source T6-v2 | evals/paired-doctrine-eval/T6-REPORT.md.
Bound to a published control-point cell.
Bound to one released terminal task, which places it in that task's concept. Ratified: 7.
Bound to a task concept of the division.
Bound to a function concept of the division.
An owner ruling binds the episode to a division key and nothing narrower. Ratified: 10. Candidate only: 34.
Candidates only, or nothing mapped: shown, never counted.
Depth is the deepest touchpoint an owner or SME ruling has ratified: a division key, a function or task concept, one released terminal task, or a control-point cell. Of 69 ledger episodes, 17 carry a ratified touchpoint and 34 carry candidates only. Evidence stays bound to BOCG v0.3.0 (identical coverage matrix bytes; evidence bindings not rebound); this view reads the v0.6.0 concept layer without rebinding anything.
| Division | Task concept | Members | Provenance | Episodes |
|---|---|---|---|---|
| Credit trading | price a cash bond or credit derivative | 1 | AUTO · machine-clustered, unread | 1 |
| Market risk management | produce daily var expected shortfall using approved model | 1 | AUTO · machine-clustered, unread | 1 |
| Market risk & valuation control | calculate risk measure test limit | 3 (2 aliases) | AUTO · machine-clustered, unread | 2 |
| Market risk & valuation control | gate model change before production deployment | 1 | AUTO · machine-clustered, unread | 3 |
Provenance is stated per division and never pooled. AUTO concepts are a deterministic machine grouping of released labels that nobody has read; they are published so the grouping is inspectable, not because it is right. The one REVIEWED division is Prime brokerage & client financing: 3 argued merges ruled 2026-09-24, 23 aliases, 54 released tasks into 40 task concepts and 15 released functions into 7 function concepts. Function concept "Prime brokerage financing, margin, stock loan and client financing composite role" (9 members); Task concept "calculate a client margin requirement" (5 members); Task concept "issue collect a house reg margin call under client s margin agreement" (11 members).
. Market risk & valuation control.
. Market risk & valuation control.
. Credit trading.
. Market risk & valuation control.
. Market risk & valuation control.
. Market risk management.
. Market risk & valuation control.
. Trade lifecycle operations.
. Trade lifecycle operations.
. Prime brokerage & client financing.
. Buy-side investment operations · Trade lifecycle operations.
. Collateral & margin management · Prime brokerage & client financing.
Episode refs are hashes of the curation ledger identity; the ledger itself holds only node ids and published grid vocabulary. Skipped ledger records, by reason: status outside surface vocabulary 6. Generated 2026-09-24.
One task is through the gate so far: FD-1197 · analyze · “memo: can the settled period be amended in place; correct sequence if not” · 5 rubric criteria · worst credible failure: mispaid counterparty · status RATIFIED. Division assignments above are machine proposals awaiting human adjudication, and probe figures come from small preregistered probes against the unaided model; both are labelled as exactly that.
T5, 9 fresh tasks: the pack beats the unaided model (+0.54, p=0.0059) and a wrong-topic pack of identical format (+0.59, p=0.0039). It is not shown to beat retrieval of its own source at equal length (+0.27, p=0.152) nor its own quoted evidence (+0.15, p=0.172). The preregistered falsifier fired and the line closed on 2026-09-02. Pack over unaided has now replicated three times: +0.41, +0.54, +0.45.
T7-v3 COMPLETE 2026-09-03; union claim PASSED narrowly on 27 admitted tasks (estimand conditional on oracle answerability): U1 +0.130, p=0.047, CI [+0.009, +0.296]; U2 +0.228, p=0.003. T8-v1 COMPLETE 2026-09-05 was the preregistered one-shot replication on 45 admitted tasks from three untouched loci of a second private corpus, packs built by the standard pipeline: U1 +0.072, p=0.047, CI [-0.002, +0.162], per locus +0.137, +0.000, +0.078, second judge p=0.057; U2 +0.177, p=0.000004. The protocol required U1 positive on every locus; it was not. The T7 effect did not transfer and the union line is closed, one shot, no re-run. What did replicate: the right pack beats the wrong pack on every locus under both judges, and the oracle source window beats retrieval by +0.38, the largest gap in the run. Earlier T6 on 20 tasks failed as written, narrowly (+0.16, p=0.061 against 0.05; wrong-topic control +0.18, p=0.041), and pack authoring stopped on 2026-09-02.
Five pre-answer signals were tried: three lexical statistics first, then two model probes on 18 T6 tasks and 10 T5 tasks. The best routed policy scored +0.208 against +0.176 for always using pack plus retrieval, a gain of +0.032 under the +0.05 admission bar fixed before the probes ran; a perfect router would reach +0.250. No routed arm runs.
Three preregistered claim gates (T4, T5, T6), none passed as written; each pre-written falsifier was applied. Instrument defects were declared rather than repaired after the fact: a void control in T4, a partial one in T5, and two T7 versions halted before any scored answer. The one preregistered pass (T7) was sent to a one-shot replication (T8) and did not survive it. Every failure is listed below.
| Run · contrast | n | Delta | p | 95% interval | Second judge | Verdict |
|---|---|---|---|---|---|---|
| T5 C1 · pack vs unaided model | 9 | +0.537 | 0.0059 | +0.287 to +0.732 | +0.556, p=0.0039 | passes |
| T5 C2 · pack vs wrong-topic pack of the same format | 9 | +0.593 | 0.0039 | +0.306 to +0.815 | +0.620, p=0.0020 | passes |
| T5 C3b · pack vs equal-length retrieval of its source | 9 | +0.269 | 0.152 | -0.194 to +0.639 | +0.269, p=0.141 | fails |
| T5 C4 · pack vs its own quoted evidence | 9 | +0.148 | 0.172 | -0.037 to +0.417 | +0.167, p=0.125 | fails |
| T5 C3a · pack vs an oracle source window | 9 | +0.046 | 0.461 | -0.102 to +0.389 | +0.046, p=0.469 | descriptive |
| T6 U1 · pack plus retrieval vs twice-budget retrieval alone | 20 | +0.158 | 0.061 | -0.004 to +0.358 | n=19: +0.184, p=0.055 | fails |
| T6 U2 · pack plus retrieval vs wrong-topic pack plus the same retrieval | 20 | +0.179 | 0.041 | +0.017 to +0.379 | n=19: +0.197, p=0.046 | passes |
| T6 U3 · pack plus retrieval vs pack alone | 20 | +0.104 | 0.042 | +0.017 to +0.242 | n=19: +0.105, p=0.055 | descriptive |
| T6 C1r · pack vs unaided model, third replication | 20 | +0.454 | 0.00003 | +0.288 to +0.625 | n=19: +0.465, p=0.00006 | descriptive |
| T6 C3b-r · pack vs equal-length retrieval of its source | 20 | +0.100 | 0.225 | -0.129 to +0.358 | n=19: +0.092, p=0.255 | descriptive |
| T6 C4f · pack vs its own quoted evidence, rebuilt control | 20 | +0.063 | 0.206 | -0.033 to +0.225 | n=19: +0.061, p=0.229 | descriptive |
| T7 U1 · pack plus retrieval vs twice-budget retrieval alone, oracle-answerable tasks | 27 | +0.130 | 0.047 | +0.009 to +0.296 | +0.130, p=0.051 | passes |
| T7 U2 · pack plus retrieval vs wrong-topic pack plus the same retrieval, oracle-answerable tasks | 27 | +0.228 | 0.003 | +0.096 to +0.395 | +0.238, p=0.003 | passes |
| T8 U1 · pack plus retrieval vs twice-budget retrieval alone, three untouched loci of a second corpus | 45 | +0.072 | 0.047 | -0.002 to +0.162 | n=43: +0.074, p=0.057 | rejects pooled; fails the per-locus rule (one locus exactly zero) |
| T8 U2 · pack plus retrieval vs wrong-topic pack plus the same retrieval, three untouched loci of a second corpus | 45 | +0.177 | 0.000004 | +0.107 to +0.273 | n=43: +0.185, p=0.000002 | passes |
The quotes-only control was empty: a format mismatch left it byte-identical to the unaided arm, so its result (+0.367, p=0.0059) was struck by erratum on 2026-09-02. Two valid contrasts of three, not three of four; the gate outcome was unchanged.
The rebuilt quotes-only arm carried 16% to 20% of the pack text, a mismatch that favoured the pack. The report says so, and T6 rebuilt the control before anything was read into it (+0.063, p=0.206).
The generated questions were too easy without documents: the cheap gate model passed 43 of 120 criteria unaided, 0.358 against a 0.20 ceiling. An instrument fault, not a finding; the run stopped before any scored answer.
The answerability filter sampled four slices of the source and admitted 15 of 30 tasks in a pattern that measured membership in the sample, not answerability. Caught before any evidence call; the version was voided and v3 regenerated from zero.
The run stopped once on a provider limit and resumed under the same frozen protocol: 173 cached evidence artifacts were honoured and no code changed. Five of 540 answer sheets were refusals, all in the unaided arm, scored zero by rule. The second judge returned no output on 16 sheets from two tasks of one locus, so its confirmatory set is n=43; the primary judge missed one replicate. The gate outcome does not depend on any of it.
T7-v3 COMPLETE 2026-09-03; union claim PASSED narrowly. U1: n=27, +0.130, p=0.047, CI [+0.009, +0.296]; U2: +0.228, p=0.003. Estimand: conditional on oracle answerability. Its untouched-locus replication (T8-v1, 2026-09-05) did NOT pass; the union line is closed. Admission floor was 20 under T7 v3. T8-v1 COMPLETE 2026-09-05; one-shot replication on three untouched loci of a second private corpus; union claim NOT passed. U1: n=45, +0.072, p=0.047, CI [-0.002, +0.162], one locus exactly zero, second judge p=0.057; U2: +0.177, p=0.000004. The line is closed by the rule set before the run: no re-run, no tuning. Scope of every figure: two loci of the doctrine registry (FD-2833, FD-1162), 9 tasks in T5 and 20 in T6 (420 answer sheets, 0 refusals), answers written by Claude Opus 5 with different reference material in the prompt, graded criterion by criterion by GLM-5.3 (primary) and Claude Opus 5 (second judge, n=19 in T6), exact one-sided sign-flip permutation test at alpha 0.05, on a private calibration corpus (one institution, not released), plus T8 on 45 tasks from three untouched loci of a second private corpus (a second institution, not released). A delta is the difference in mean task score, the fraction of rubric criteria passed. No human grading; external validity is unmeasured.
07 · Headline results
Unsaturated
The spread is the instrument’s resolution
No hidden-field probes or canary echoes
With expected errors
Harness v2 · preregistered · August 2026. The scripted honest baseline must pass 40/40 before any model result is reported.
08 · The evidence chain
Runs are preregistered. Configs, transcripts, metrics and reports bind to a manifest by SHA-256. Independence from any real corpus is measured at release — not asserted.
git clone thefazzer/bankingops-coverage-grid pip install -e bocg bocg gate all -w live-run-20260826
09 · Working with the Lab
Private hash-sealed set derived from the public episodes by reslug, date shift and amount scale; canaries are per-run, not part of the sealed inventory.
Your model on a private set, delivered as a hash-bound evidence pack.
Custom operational families on the same contract-and-gates machinery.
Sealed-set minting at lab scale.
10 · Contact
Public sample is MIT. To work with the Lab, open an issue on the repository — the machinery, results and evidence chain are all here.