sources: add the MIC column, conform Code on IDs 4 and 5, and fix the version table

[ben-malbeclabs]

Registry 1.4.0. A new column, and two corrections — one of which reverses a decision in #46.

Summary of Changes

A MIC column, recording each engine's ISO 10383 market identifier code. Raised by @armcconnell in review of malbeclabs/miax#2: a distinct MIC is a good test for a tradfi venue needing separate Source IDs. Recorded as a column rather than prose, so it is a value a requester fills in rather than an argument they have to remember to make.

Stated as sufficient, never necessary. The converse fails badly — one MIC routinely covers several matching engines, which is the ordinary case wherever a venue runs spot and derivatives under one code, and describes two of the venues already in this table.

Three states, each meaning something distinct. An em dash: the venue has none, normal for a decentralised exchange or anything outside the standard's scope. A code: 3 is KLSH. (withheld): one exists, but naming it would identify a venue trading under a codename, which is the state of IDs 4 and 5 — their MICs are public and would defeat the codename here.

Code on IDs 4 and 5 now reads SETAI. #46 made Code name the venue and left those two rows carrying SETAI_FINANCIALS and SETAI_COMMODITIES as an exception, on the grounds that a Code already in use downstream cannot be changed without coordinating with whoever carries it.

That reasoning holds for the publishing venues and not for this one. Nothing carries these two values: the venue has no publisher, no metric series and no registered multicast group. A grep finds them only in the two table cells and the sentence explaining why they could not change — which this PR also removes, rather than rewording. The window closes at first deploy, since dz-publisher-metrics stamps venue and source_id on every series.

The VERSIONING.md row is corrected. It still read 1.2.0 after #46 moved sources/spec.md to 1.3.0. That table is the fleet view and was the only place stating a wrong version.

MINOR rather than PATCH because adding a column to the assignment table is additive under this document's own rule — the same classification 1.1.0 took for Code and Venue.

Testing Verification

• go test ./... -count=1 in tools/conformance — 7 of 7 packages.
• SETAI_ no longer appears anywhere in sources/spec.md.
• 1.4.0 consistent across the version line, the changelog entry and the VERSIONING.md row.
• Every row has a MIC cell; the three states are defined in the column description. KLSH verified against the ISO 10383 listing for KalshiEX LLC.
• IDs 1, 2, 3 and 6 otherwise untouched. No ID renumbered, reordered or removed, and no assignment changed.

[juan-malbeclabs]

The MIC criterion over-splits: ISO 10383 segment MICs routinely share one matching engine.

Line 9 says a distinct MIC "settles the test above without further argument". But ISO 10383 has operating MICs and segment MICs, and segment MICs commonly name market segments of a single matching platform rather than separate matching. Euronext Paris carries XPAR plus XMLI, ALXP, XMAT and others, all matched on one Optiq platform under one rulebook; London and Deutsche Börse are the same shape.

Under this paragraph as written, that venue is entitled to a separate Source ID per segment MIC — which is exactly the case GLOSSARY.md excludes ("a capacity shard the venue can rebalance is a channel rather than an engine"). The registry would grant four IDs where the authoritative test grants one, and IDs are never renumbered, so the split is unrecoverable.

Worth narrowing to operating MICs, or restating the criterion at the strength the bold lead sentence uses. The lead says "strong evidence" and the closing sentence says "sufficient" — those are different rules, and a requester will read whichever helps them.

[juan-malbeclabs]

This widens the engine-distinctness test in the document that disclaims authority over it.

Line 7, unchanged just above: "GLOSSARY.md carries the full test and is the authority for it." GLOSSARY.md Matching engine (1.3.0) states a two-limb test — rules differ, or independent matching over disjoint sets — and is not touched by this PR.

So after this merges, a requester following the stated authority applies two limbs and a reviewer applying sources/spec.md 1.4.0 applies three, and the same request resolves differently depending on which document was consulted.

The repo's precedent runs the other way: the last widening landed in GLOSSARY.md 1.3.0 first, and this registry's own 1.2.0 entry records itself as "conformed the Source ID definition above to GLOSSARY.md 1.3.0". Same move here — add the MIC limb to the glossary entry, bump it, and let this document cite it.

[juan-malbeclabs]

Both Setai cells read (withheld), so the only multi-row venue with MICs cannot demonstrate the new rule.

Line 39 says "Where two rows at one venue carry different MICs, that alone establishes them as distinct engines." IDs 4 and 5 are the one place in this table where that could be shown, and both cells are withheld — and the state is defined in the singular ("one exists"), which does not say whether these two rows share a MIC or hold different ones.

There is a second-order effect: if they do hold different MICs — which the withholding rationale implies, since naming them would identify the venue — then line 67's "the first assignment resting on the second half of the rule rather than the first" is superseded by the new sufficient condition and was not updated. Either the withheld state needs to encode same/different, or that sentence needs a clause acknowledging the MIC ground now also applies.

[juan-malbeclabs]

The em dash on ID 6 contradicts this document's own account of the venue.

Line 39 defines the em dash as "the venue has none, which is the normal case for a decentralised exchange and for any venue outside the standard's scope". Line 61 says of this same row: "The operator of record is Nest Exchange Limited, licensed in the Abu Dhabi Global Market as a Recognised Investment Exchange for derivatives since 5 January 2026."

A licensed RIE is squarely inside ISO 10383's scope, so the em dash asserts something the ID 6 section denies. The column has three states and none of them means "not researched", so an unfilled cell is indistinguishable from an authoritative "none exists" — and a registry mirror built from this table will carry the assertion.

[juan-malbeclabs]

This PR deletes the only statement constraining a change to an existing Code, rather than narrowing it.

The removed sentence read: "a Code already in use downstream cannot be changed without coordinating with whoever carries it." Only its application to IDs 4 and 5 is obsolete; the rule itself is not. After this hunk, nothing in the registry requires coordination before changing any Code, and line 82 still classifies a Code edit on an existing row as a PATCH.

Concretely: a later PR renames KALSHI or BINANCE — values this document itself says downstream carries as metric label values — as a patch release. GLOSSARY.md line 32 records that this failure is silent: "a code that stops matching its live group fails silently rather than loudly." The compatibility promise at line 84 only covers what an ID means; it says nothing about the Code column.

Suggest keeping the general rule and dropping just the IDs 4/5 clause.

[juan-malbeclabs]

MIC is defined here as the engine's code, which line 9 denies and row 3 disproves.

Line 9, added in this same PR: "one MIC routinely covers several engines". Line 74 repeats the framing — "records the engine's MIC where it has one". Engines do not hold MICs; markets and market operators do.

Row 3 is the counterexample already in the table: it carries one MIC KLSH while its own section says it "carries two matching engines and is due to split". When that split lands, a requester following step 2 must record "the engine's MIC" for each new row, finds only one venue-level code exists, and has no rule saying whether duplicating KLSH across both rows is correct.

"the market identifier code under which the engine trades" (or "the venue's MIC for this market") would match how the rest of the paragraph actually behaves.