diff --git a/docs/experiments/2026-08-09-tradenation-feasibility.md b/docs/experiments/2026-08-09-tradenation-feasibility.md
new file mode 100644
index 00000000..177a4d22
--- /dev/null
+++ b/docs/experiments/2026-08-09-tradenation-feasibility.md
@@ -0,0 +1,654 @@
+# Trade Nation — can keel trade through it?
+
+**Date:** 2026-08-09
+**KB basis:** §28.1's third prohibition names the exclusion at the root — gharar covers *"all
+financial derivative instruments, forwarding contracts, and future agreements"*, and the section
+spells out the consequence itself: this *"grounds exclusion of **CFDs**, futures, forwards, options,
+all derivatives"*. §28.1's second prohibition, maisir, is the one that catches the *other* half of
+Trade Nation's product range: *"the speculative trade or exchange of money for debt **without an
+underlying asset transfer**"* — a financial spread bet is the paradigm case, not an edge one.
+§28.2 supplies the positive statement the refusal is measured against: *"Spot crypto = you OWN the
+asset … **CFD/future/option = you own nothing**, you hold a bet on price with a counterparty →
+gharar + no ownership → excluded"*, and derives long-only from the same root. §65.11 sharpens
+maisir into an operative criterion: contracts *"settled through price differences only are covered
+under gambling"*, with the note that *"the operative criterion is **cash settlement / no delivery**,
+not futurity as such"* — which is what makes it bite on a CFD that has no expiry at all; the same
+section excludes *"Swaps, CFDs, hedging products"* by name and records that *"short-selling has been
+prohibited by almost all scholars"*. §56.1, as corrected by §66.3, governs the FX half: retail
+"spot" FX is *"automatically rolled over to the next business day … necessary to avoid the actual
+delivery of the currency"* with the
+*"interest differential"* charged for the deferral, and §66.3 sharpens the defect from "T+2 ⇒ riba"
+(too blunt) to **non-delivery by design**, supplying the litmus test this document applies
+throughout — the Malaysian Perlis fatwa's condemnation of *"no transfer of property by both
+parties, either substantively or constructively"*. §65.4/§67.1/§71.5 triangulate `qabd` across
+three independent sources onto one operative test — possession is the ability to dispose, *"nothing
+to prevent the buyer from taking physical possession … whenever he desires"* — which is the direct
+grounding for rail 17. §71.5 also records the one attribute Trade Nation genuinely satisfies:
+*"registered/licensed with a recognised regulator"* as a per-broker compliance property, re-checked
+per venue. §65.5 governs their gold and silver: *"futures trading in commodities like gold and
+silver that serve as Thaman is forbidden"*, with the 72-hour settlement bound as the citable
+outer limit. §71.6 grounds the stock-CFD point: a claim-bearing instrument requires share-style
+business *and* financial screening keel does not have, so it is rejected **by capability**, not
+merely by sector. §66.7's crypto gap — since **partially** closed by §71/§72/§85 — is not
+load-bearing here in either direction: their crypto products are refused as CFDs, not as crypto.
+**Status:** feasibility only. **No code changed, no adapter added, no dependency introduced.**
+**Evidence:** public documentation review, not a live probe, and this document does not dress it up
+as one. **No Trade Nation account was opened**, no demo registered, no MT4 terminal connected, no
+order placed or previewed. Every claim about the product range, the platforms and the regulators is
+sourced to a `tradenation.com` / `support.tradenation.com` page or a named third party, read on
+2026-08-09 and cited inline. The negative claim — that no public trading API exists — is
+established by **enumeration of their own published sitemaps**, not by a failed search; the method
+and the counts are given in full below so the claim can be re-run and falsified rather than taken
+on trust. `api.tradenation.com` exists and resolves, but it is a private backend for their own
+apps: it was checked for existence and **deliberately not probed, enumerated or reverse-engineered**
+beyond an unauthenticated request to `/`, per issue #204's own instruction. Where the public record
+is silent, this document says so rather than guessing.
+
+## The question
+
+Issue #204 asks the same question of Trade Nation that #201 asked of cTrader, and it deserves an
+answer that does not simply reuse the previous one. The two cases look alike — a retail
+CFD/spread-betting house surfaced as a candidate execution venue — but they fail in structurally
+different places, and conflating them would teach the wrong lesson to whoever reads this next.
+
+cTrader's assessment had two things to lean on: a fully documented protocol whose own message model
+made leverage, swap and short-symmetry unavoidable, and a Python SDK that had not shipped in two
+years. Both are absent here, in opposite directions. Trade Nation publishes **no** protocol
+documentation at all, so there is no message model to read — and yet a **technically usable,
+self-serve, broker-advertised Python order-placement path nevertheless exists**, because Trade
+Nation offers MetaTrader 4 and explicitly promotes Expert Advisors as its automation route. That
+inversion is what makes this document worth writing separately: the refusal here cannot rest on
+"the tooling is unmaintained" or "the transport is awkward". It has to rest entirely on **what is
+being traded**, and it does.
+
+## Verdict
+
+**No. Trade Nation is refused, and the refusal is structural — it rests on the instrument, not on
+the transport.** Nothing in Trade Nation's product range is unleveraged spot ownership. Every
+tradeable thing they offer is a contract for difference, or in some jurisdictions a financial
+spread bet on the same underlying: forex, indices, commodities, individual stocks, bonds and
+cryptocurrencies, all as CFDs, with no share ownership and no spot crypto anywhere
+(https://tradenation.com/markets, https://tradenation.com/our-platforms/) — TN Trader carries named
+bond futures (`BUND - Future (Sep)`, `BOBL - Future (Sep)`) and differential spreads (`Gold / Silver
+Diff`, `Germany 40 / UK 100 - Rolling Future Diff`) alongside the more familiar range, which only
+widens how much of it is CFD-only. keel's charter is spot-only, and several independent gates already
+enforce that charter against any venue without needing to know Trade Nation exists.
+
+| Gate | What it checks | Trade Nation's shape | Result |
+|---|---|---|---|
+| The product itself | is anything here unleveraged spot ownership? | CFDs and spread bets only — no share custody, no spot crypto, no delivery of anything | **the decisive failure.** Nothing to build an adapter *for* |
+| Rail 19 (`spot_instrument`) | product id must be `BASE-QUOTE`, uppercase, exactly one hyphen (`guards.py:750-756`) | `EUR/USD`, `XAUUSD`, `US500` (MT4), `Germany 40`, `UK 100` (TN Trader) — a slash, spaces or nothing, but never the single hyphen the grammar requires | veto on every intent; and the rail's comment says there is **no config field to widen** it (`guards.py:741-742`) |
+| Rail 18 (`settlement_currency`) | last segment vs `config.settlement_currencies` (`guards.py:696-708`) | `quote_currency_of("EUR/USD")` is `None` — no separator (`products.py:57-62`) | fails **closed**, independently of rail 19 |
+| Rail 17 (`withdrawal_capability`) | `qabd`: is the account in a state where the asset could be withdrawn on demand? (`guards.py:644-666`) | a CFD has **no underlying to withdraw, ever** — the question has no true answer | fails closed on every LIVE BUY; but see the honest caveat below — this is an *operator*-honesty gate |
+| Rail 13 (USDC-funding) | settled quote balance ≥ full order notional (`guards.py:512-541`) | margin: the whole point is to post a fraction of notional | structurally anti-margin — 100% cash cover or no BUY |
+| Rail 1 (halal allowlist) | base asset ∈ `config.allowlist` (`guards.py:368-373`) | `EUR`, `US500`, `XAUUSD`, `AAPL` are not on it, and could not be added honestly | veto, both sides, every mode |
+| Curation screen (`screen.py`) | `spot_instrument` grammar, then `backing` ∈ {ayn, dayn, native}, then `pays_yield` | a CFD is a bilateral contract — none of the three backings; the id fails the grammar one gate earlier | refused at admission, *provided the operator attests the contract rather than its underlying* — see #202 |
+| `BrokerCapabilities.asset_classes` | vocabulary: `{"spot", "futures", "equity"}` (`capabilities.py:21`) | no `cfd` / `margin` / `derivative` exists to declare | cannot be honestly declared — but **inert on the live path today**, by the class's own admission (`capabilities.py:28-40`) |
+| `Side` / strategy layer | `Side` is BUY/SELL only (`types.py:25-29`); `Setup.direction: Literal["long"]` (`base.py:39`) | CFD SELL opens a short, not a disposal | no vocabulary exists to express it |
+
+Instrument naming is not even consistent across Trade Nation's own two platforms — TN Trader spells
+these `EUR/USD`, `Gold (per 0.1)`, `US 500 (Per 1.0)`, `UK 100`, `Germany 40`; MT4 spells them
+`EUR/USD`, `XAUUSD`, `US500`, `UK100`, `DE40` — but every variant on both platforms fails rail 19's
+grammar the same way, for the same reason: none of them is `BASE-QUOTE` with exactly one hyphen.
+
+**One-line answer: Trade Nation is not a spot venue with a CFD wrapper bolted on — the CFD *is* the
+product, there is no unleveraged, delivery-settled instrument anywhere in the range, and the
+existence of a working automation path through MetaTrader 4 does not change what that path would be
+automating.**
+
+## What Trade Nation actually is
+
+Trade Nation is a retail CFD and financial-spread-betting broker with a proprietary web/mobile
+platform (TN Trader) and a MetaTrader 4 offering alongside it. Its published range is
+*"1,000+ local and global markets, such as Forex, indices, commodities, and stocks"*
+(https://tradenation.com/markets), plus cryptocurrencies and, on TN Trader specifically, bond
+futures (`BUND - Future (Sep)`, `BOBL - Future (Sep)`) and cross-instrument differential spreads
+(`Gold / Silver Diff`, `Germany 40 / UK 100 - Rolling Future Diff`) — every one of them as a contract
+for difference, and in the jurisdictions where the tax treatment applies, as a financial spread bet
+on the same underlying. Bonds are named explicitly, alongside forex, among the design spec's own
+exclusions —
+*"Bonds, forex carry, derivatives/futures/perps"*
+(`docs/superpowers/specs/2026-07-15-keel-autotrade-design.md:39`) — so Trade Nation's bond products
+hit a hard exclusion this document has already invoked for its forex products, not a new one. There
+is no share dealing, no custody, no spot crypto, and no delivery
+mechanism of any kind — their own UK Client Agreement states the no-ownership fact as a contractual
+term, not merely a marketing description: *"You will not have any rights of ownership or otherwise
+in any Instrument as a result of a Transaction with us"* (clause 3.5), quoted in full where it does
+the decisive work below. Their headline commercial proposition is **fixed spreads**, which is itself
+a tell about the product: a fixed spread is something a dealer quotes, not something an order book
+produces.
+
+The entity is regulated in six jurisdictions, per the footer of https://tradenation.com/our-platforms/
+— Seychelles FSA (SD150), UK FCA (525164), Australia ASIC (AFSL 422661), Bahamas SCB (SIA-F216),
+South Africa FSCA (49846), and Portugal/EU CMVM (601) — and does not accept US residents. This is
+worth stating plainly and without hedging, because it is the one compliance attribute in §71.5's
+per-broker list that Trade Nation genuinely satisfies: they are *"registered/licensed"* with
+recognised regulators, several of them tier-one. **The refusal below is not a claim that Trade
+Nation is disreputable, unregulated or shady.** It is a claim about what a CFD *is*, and it would
+apply identically to a flawlessly run venue. Regulatory standing answers §71.5's venue question and
+leaves §28.1's instrument question completely untouched.
+
+They also operate a "Pro Accounts Programme" — FCA elective-professional status, applied for via
+`support.uk@tradenation.com`. It is worth naming only to close it: it concerns higher leverage and
+the surrender of retail client protections. It has nothing to do with connectivity, and nothing in
+it converts a CFD into an owned asset. It is the opposite of what would be needed.
+
+## The transport question — established by enumeration, not by a failed search
+
+A negative claim about a public API is easy to make carelessly and hard to make well. "I searched
+and found nothing" is not evidence; it is the absence of evidence, and it is exactly how a
+feasibility document acquires a claim that a later reader quietly discovers is wrong. So the claim
+here is made by enumerating what Trade Nation itself publishes, with counts, so it can be re-run.
+
+Their two sitemaps — `tradenation.com/sitemap.xml` (675 URLs) and `tradenation.com/en-gb/sitemap.xml`
+(748 URLs) — enumerate **1,423 URLs** between them (counts as of 2026-08-09; they drift as articles
+are published, which is why the matched-set result rather than the total is the finding). Matching
+that set against `api|institutional|professional|developer|liquidity|white-label|b2b` returns
+**zero** connectivity pages. The only URLs containing the substring "api" (case-insensitively) are a
+news article about the American Petroleum Institute and a page whose slug contains the word
+"capital" (`https://tradenation.com/zest-capital-traders/`); neither is a connectivity page. The
+support knowledge base, `support.tradenation.com/sitemap.xml` (160 URLs), matched against
+`api|automat|algo|expert|robot|institution`, likewise returns zero. Direct requests to
+`/institutional/`, `/professional/`, `/developers/`, `/liquidity/`, `/white-label/` and `/b2b/` all
+return 404; `/api/` returns a 302 trailing-slash redirect to `/api`, which is also 404. The word
+"API" appears **zero times** in the rendered text of the home page, `/partners/`, `/about-us/`,
+`/partner-faq/`, `/tradingview/` and `/metatrader-4/`.
+
+The partner programme is affiliate and Introducing-Broker only (https://tradenation.com/partners/) —
+revenue share for referred clients. There is no connectivity tier, no liquidity tier, no technology
+tier, and no institutional desk surfaced by that sweep. Beyond their own site,
+there is no PyPI package, no GitHub SDK, and no public teardown: searching GitHub for
+`api.tradenation.com`, `tradenation trading api`, and `TradeNation-LiveBravo` returns zero results
+for all three. `api.tradenation.com` does resolve — to CloudFront, returning 403 at the root — and
+that is the entire extent to which it was touched. It is a private backend for their own
+applications. Per #204's explicit instruction it was **not** enumerated, authenticated against, or
+reverse-engineered, and this document draws no inference from it beyond "it exists and is closed".
+
+### The white-label hypothesis is dead, and that is a real finding
+
+The most promising lead was that Trade Nation might be running someone else's platform, in which
+case the vendor's API documentation would be the thing to read and the transport question would
+have a different answer. Their Android package id is `com.finsa.tradenation`, which reads like a
+vendor namespace. It is not.
+
+`FINSA EUROPE LTD` **is** `TRADE NATION FINANCIAL UK LTD` — the same company, Companies House
+number **07073413**, renamed on 23 February 2022 (and before that, `THE TRADER MANAGEMENT COMPANY
+LTD` from 2009 to 2014). `com.finsa.` is a legacy in-group namespace surviving a rename, not
+evidence of a third-party platform vendor. A deliberate sweep for the usual suppliers —
+Devexperts/DXtrade, TradeLocker, Match-Trader, Spotware/cTrader, oneZero, PrimeXM, Gain/StoneX,
+Gold-i, Your Bourse, B2Broker, Leverate — found no connection to any of them.
+
+The positive evidence points the other way: they build it themselves. Their own recruitment copy
+describes *"the next generation of **TN Trader — a proprietary, React-based trading platform**
+designed for the needs of global CFD and spread betting traders"*
+(https://apply.workable.com/j/EFF9D2D56F), and a second listing enumerates the engineering surface
+as *"Execution & Trading - pricing engines, liquidity-provider integrations, risk monitoring and
+trade-execution logic"* (https://apply.workable.com/j/DA8D46C30B). The only bought-in components
+identifiable are ChartIQ for charting and TradingView. **So there is no vendor API document to go
+and read.** That route is closed, and closing it properly is what turns "we could not find an API"
+into "there is not one".
+
+### Competitor context — so the absence reads as normal, not as a red flag
+
+Retail CFD houses split cleanly on this question, and the split tracks size and platform strategy
+rather than integrity. Self-serve public retail APIs exist at IG (`labs.ig.com`), Capital.com
+(`open-api.capital.com`), OANDA (v20), Saxo (`developer.saxo`) and Interactive Brokers.
+Application-gated ones exist at City Index / StoneX (`ciapi.cityindex.com` is live but not
+self-serve). And there are none at all at CMC Markets, Spreadex, Plus500 — or Trade Nation.
+Proprietary-platform CFD houses of Trade Nation's size push retail automation into MetaTrader
+instead of building and supporting a public API, and Trade Nation matches that pattern exactly.
+
+One distinction is worth drawing, because it would be unfair to blur it: Trade Nation is **not**
+automation-hostile in the way Plus500 is, whose terms prohibit use of any *"automated data entry
+system"*. Trade Nation actively advertises Expert Advisors. The absence of an API here is a product
+decision, not a policy against algorithmic clients. Third-party review coverage reaches the same
+reading: *"the broker does not currently provide direct API access such as FIX or REST connections,
+[but] its MT4 integration still accommodates most retail-level automation needs"*
+(https://www.tradomatix.com/broker-reviews/trade-nation/).
+
+## The MT4 / Expert Advisor route — a real path, closed on purpose
+
+This is the intellectual centre of the assessment, and the section that most distinguishes it from
+the cTrader one. It would be easy, and dishonest, to write "no API, therefore no" and stop. A
+working programmatic order-placement path to Trade Nation **does** exist, it is self-serve, and the
+broker advertises it. Any future reader who does thirty minutes of research will find it, and if
+this document has not already dealt with it, the whole refusal will look like it rested on a
+transport gap that turned out not to be one.
+
+So, stated at full strength before it is answered. Trade Nation's platforms page lists
+*"MetaTrader 4 — … **Automate using Expert Advisors (EAs)**"* alongside *"CFD trading on forex,
+indices, and more"* (https://tradenation.com/our-platforms/), and their MT4 page is explicit about
+the coding route: *"Trading robots, or Expert Advisors (EA), automate the trading process. Have
+coding experience? You can create expert advisors using **MetaQuotes Language 4**"*
+(https://tradenation.com/metatrader-4/). The live server name is published in their own help centre
+— `TradeNation-LiveBravo` (https://support.tradenation.com/connecting-mt4-accounts). An MT4 account
+is self-serve: no application, no professional gate, no minimum deposit, no manual review of the
+kind Spotware imposes on a cTrader app. And MQL4↔Python bridges are a solved, open-source problem —
+Darwinex's DWX ZeroMQ Connector (https://github.com/darwinex/dwx-zeromq-connector) is the canonical
+one. A determined engineer could have Python placing live orders at Trade Nation inside a weekend.
+
+There is a second broker-advertised automation route, and it should be named rather than left for a
+later reader to discover: Trade Nation runs its own copy-trading app, TradeCopier
+(https://tradenation.com/tradecopier/, https://tradenation.com/tradecopier-faq/), which links to an
+MT4 account and mirrors trades onto it. It is worth flagging precisely because the enumeration above
+would not have surfaced it on its own — the `api|institutional|professional|developer|liquidity|
+white-label|b2b` regex run against the two main-site sitemaps has no term that matches "tradecopier",
+and the broader `api|automat|algo|expert|robot|institution` regex was run only against the support
+sitemap, not the main site. It changes nothing about the verdict: TradeCopier is a mobile
+copy-trading client, not a Python path, and whatever it mirrors trades onto is the same MT4 CFD
+account already addressed above.
+
+**And it changes nothing, because the thing at the far end of that bridge is still a CFD.** This is
+the whole point. Every gate in the verdict table above is a gate on *what is traded*, not on *how
+the order got there*. Rail 19 asks what shape the instrument id is. Rail 18 asks what it settles in.
+Rail 17 asks whether the acquired thing can be withdrawn. Rail 1 asks whether the base asset is on
+the halal allowlist. The curation screen asks what backs it. Not one of them takes a transport as
+an argument, and none of them would notice or care that the fill arrived via ZeroMQ from a Windows
+terminal rather than over HTTPS from a REST endpoint. §28.1's gharar exclusion and §65.11's
+cash-settlement criterion are properties of the contract; a bridge does not launder them.
+
+Because the route is closed on the instrument, the practical objections to it are *secondary* — and
+they are recorded here only so that no one mistakes them for the reason, and so that no one later
+proposes "MT4 but better" as though the reason had been ergonomic:
+
+- **MT4 is CFD-only, not spread betting**, per Trade Nation's own bullet — so the route does not even
+  reach their full range, and the part it reaches is the part §28.1 excludes as gharar rather than
+  the part it excludes as maisir. Cold comfort.
+- **Fixed spreads do not apply on MT4**; it is variable-spread, which discards the one pricing
+  proposition that made Trade Nation distinctive in the first place.
+- **No equities on MT4** — those are TN Trader-only — so even the instrument set narrows.
+- **No MT5.** Their help centre's own article slug is `do-you-offer-mt5`, and the answer is no. That
+  matters concretely: the official `MetaTrader5` PyPI package, the only first-party Python
+  integration MetaQuotes ships, is therefore unavailable. Every Python path here is third-party.
+- **MT4 is a stateful Windows GUI terminal** requiring an always-on VPS plus a bridge EA holding a
+  socket open — a transport shape wholly unlike every broker keel has: a stateless, request/response
+  HTTPS client behind the `Broker` port. Adopting it would mean an availability model and a failure
+  model the engine has no vocabulary for.
+- **Hedging is advertised, not restricted.** A deal-ticket hedging toggle is a named term in Trade
+  Nation's Client Agreement and a bullet on their MT4 page — this is not a route Trade Nation is
+  trying to close off. The one formal restriction on the books is narrower than "no scalping": a UK
+  Scalping Policy aimed specifically at latency and stale-price arbitrage, not at fast trading as
+  such, and the UK Client Agreement's clause 12(k) constrains automated devices only by a good-faith
+  test. That supports, rather than complicates, the point made above: Trade Nation is not
+  automation-hostile the way Plus500 is.
+
+**The MT4/EA route is therefore refused for the same reason the venue is: it places CFD orders.
+Re-proposing it with a better bridge, a managed VPS, or an MT5 migration does not move any of the
+gates, because none of the gates is about the bridge.**
+
+## The decisive section — instrument semantics
+
+Everything above is context. This is the finding.
+
+A Trade Nation position is a bilateral contract with Trade Nation, settled in cash against the
+movement of a reference price. Buying their `XAUUSD` product does not put gold anywhere; buying
+their `AAPL` product confers no share and no vote, and the dividend adjustment it credits
+(https://tradenation.com/dividend-projections/) is a cash payment from the dealer, not a
+distribution from the issuer — which strengthens the point rather than weakening it: it is exactly
+§28.2's *"you own nothing"* fact pattern, restated in the one place a naive reader might expect
+ownership to leak through. Buying their `Bitcoin` product leaves no coin in any wallet. There is no
+custody leg, no delivery leg, no settlement of the underlying, and no mechanism anywhere in the
+product range by which one could come to exist. That is not a defect in their implementation — it is
+the definition of the instrument they sell, and their own UK Client Agreement says so in terms that
+leave nothing to marketing-copy interpretation: *"You will not have any rights of ownership or
+otherwise in any Instrument as a result of a Transaction with us. We will not transfer any Instrument
+or the rights in such Instrument (such as voting rights) to you"* (clause 3.5).
+
+Applied to the KB's own tests, the answers are unanimous and they are not close:
+
+**§66.3's litmus test — the one the Coinbase study adopted as operative — asks whether the trade
+results in transfer of property *"substantively or constructively"*.** For a CFD the answer is never,
+by construction, on the venue's own description of its product. The Perlis fatwa §66.3 reproduces
+condemns online retail forex on precisely four grounds, and a Trade Nation forex CFD satisfies all
+four rather than some of them: no transfer of property either substantively or constructively; riba
+in the financing of the margin facility; gharar; and maisir, because *"traders face profit/loss
+purely from predicting price direction, no underlying economic activity"*. §66.3's closing
+instruction is written as though for this case — *"if ever the answer becomes the latter (e.g., a
+derivative-wrapped 'spot-tracking' product), it fails this test regardless of its marketing label."*
+
+**§65.11 supplies the maisir grounding independently of the gharar one**, and its operative criterion
+is the sharp one here: contracts *"settled through price differences only are covered under
+gambling"*, where *"the operative criterion is cash settlement / no delivery, not futurity as such"*.
+A CFD has no expiry at all, so an argument that leans on futurity would miss it entirely; an
+argument that leans on cash settlement lands squarely. The same section excludes *"Swaps, CFDs,
+hedging products"* by name.
+
+**Trade Nation's spread-betting product is worse, not better.** It is the jurisdictional variant
+sold on its tax treatment, and its literal form is a wager staked per point of price movement.
+§28.1's second prohibition is maisir, defined as *"gambling/speculation/betting"*; there is no
+interpretive distance to travel to reach a product whose own name is "spread **betting**".
+
+**Short symmetry.** A CFD SELL opens a short position; it is not the disposal of a holding, because
+there is no holding. §65.11 records that *"short-selling has been prohibited by almost all
+scholars"*, and §28.2 derives long-only from the ownership argument independently — *"short = sell
+what you don't possess"*. keel's design spec names the same exclusion at
+`docs/superpowers/specs/2026-07-15-keel-autotrade-design.md:37`.
+
+**The commodity products are the worst case, not the best one.** The instinct that gold "feels
+closer to spot" than an index is exactly backwards. §65.5 makes gold and silver *stricter* than
+crypto, not looser: they are named *ribawi* commodities whose *thamaniyyah* is *"not arguable — it
+is the paradigm case"*, deferment is prohibited *"no exceptions"*, and the section states outright
+that *"futures trading in commodities like gold and silver that serve as Thaman is forbidden"*. The
+citable outer bound on settlement latency is 72 hours — reproduced in keel's own code at
+`keel/compliance/screen.py:244-248`, which warns that for gold or silver backing *"§65.5's stricter
+bay' al-sarf regime applies -- no deferment, 72h settlement bound"*. A CFD does not settle the
+underlying in 72 hours, or in 72 years. It never settles the underlying at all.
+
+**The forex products** hit the exclusion the design spec names directly — *"Bonds, forex carry,
+derivatives/futures/perps"*
+(`docs/superpowers/specs/2026-07-15-keel-autotrade-design.md:39`) — and §56.1's negative exemplar
+describes the mechanism from the inside: rolled nightly *"to avoid the actual delivery of the
+currency"*, with the *"interest differential"* charged for the privilege.
+
+**The index products** are what the Coinbase study called *"index basket + no delivery + no
+constituent screening. Worst case in the set."*
+(`docs/experiments/2026-08-05-coinbase-asset-class-feasibility.md:54`). Trade Nation's `US500`,
+`UK 100` and `Germany 40` are the same object with the futures wrapper swapped for a CFD one, which
+removes the expiry and changes nothing else that matters.
+
+**The stock products fail by capability, which is a different and stronger failure.** §71.6's rule
+is that any claim-bearing instrument requires share-style screening of *"core business activity AND
+financials"* — a two-part test keel performs neither half of, because its allowlist deliberately
+contains no issuer-claim instruments. The KB's recorded conclusion is that *"admitting any
+equity/revenue/claim-bearing token would require a screening capability we do not have"*. A stock
+CFD is worse still: it is a claim on Trade Nation about the price of a claim on the issuer. Even if
+keel had the share screen, the CFD wrapper would remain.
+
+**Crypto, honestly.** §66.7 recorded that none of its four sources addressed crypto directly. That
+gap is now **partially** closed — §71.1's IIFA Resolution 237 is a deliberate non-ruling, §72 maps
+a split and thin literature, and §85 is the first in-depth treatment by a qualified jurist. None of
+it bears on this question in either direction, because Trade Nation's crypto products are refused
+**as CFDs**, on §28.1 and §65.11, exactly as their gold and index products are. The refusal is on
+the wrapper, not the underlying, so no crypto-specific ruling is extended or relied on here.
+
+## The keel-side gates — verified against the working tree
+
+The instrument section above is sufficient on its own. What follows maps the refusal onto keel's
+actual code, in the order an adapter attempt would hit it, so that the verdict is traceable rather
+than asserted. Every line reference below was read in the working tree on 2026-08-09.
+
+**1. Rail 19, spot instrument shape.** `keel/execution/guards.py:750-756` vetoes any
+`OrderIntent.product_id` for which `parse_spot_product_id` returns `None` — *"not a well-formed spot
+product id (BASE-QUOTE, uppercase, exactly one hyphen)"*. Here, unlike in the cTrader case, the
+grammar **does** fail the venue on its real symbols, and the naming is platform-specific to boot:
+TN Trader has `EUR/USD`, `Gold (per 0.1)`, `US 500 (Per 1.0)`, `Germany 40`, `UK 100`; MT4 has
+`EUR/USD`, `XAUUSD`, `US500`, `DE40`, `UK100` — a slash, spaces or nothing on either platform, but
+never the single hyphen `_SPOT_PRODUCT_ID_RE` requires (`packages/keel-core/keel_core/products.py:46`),
+and `parse_spot_product_id` (`products.py:86-117`), which applies that regex, rejects every variant
+of both. The rail
+is un-skippable by construction: its own comment at `guards.py:737-739` reads *"BOTH SIDES, EVERY
+MODE, DCA INCLUDED — deliberately not in `LIVE_STATE_RAILS`"*, so paper trading runs it too, and
+*"a rehearsal cannot build a track record on trades live trading would veto."* The load-bearing
+sentence, though, is the charter statement two lines further down (`guards.py:741-742`):
+
+> Spot-only is this agent's CHARTER, not an operator preference, so there is no config field here
+> to widen (unlike rail 18's `settlement_currencies`).
+
+The module docstring says the same thing twice more, at `guards.py:1` — *"THE HARD RAILS (§14) —
+enforced before every order, un-overridable"* — and at `guards.py:93-94`: *"no config field to widen --
+spot-only is this agent's charter, not an operator preference."*
+
+**2. Rail 18, settlement currency.** `guards.py:696-708` resolves the settlement leg with
+`quote_currency_of` and compares it to `config.settlement_currencies`. Because `quote_currency_of`
+returns `None` for any id without a separator (`products.py:57-62`), `EUR/USD` produces
+*"cannot resolve a settlement currency … failing closed"* — a second, independent veto reached by a
+different route than rail 19's. This rail *is* operator-configurable — the field defaults to
+`{USD, USDC}` (`packages/keel-core/keel_core/config.py:319`, `config.yaml:100-102`) and
+`guards.py:689-690` calls it *"the escape hatch, which is why the set is not hardcoded here"* — but
+widening a currency set does nothing for an id that has no currency leg to widen *to*.
+
+**3. Rail 17, withdrawal capability (`qabd`).** `guards.py:644-666`. The rail's comment states the
+juristic basis: possession holds only while *"there is nothing to prevent the buyer from taking
+physical possession whenever he desires"*, and *"an asset we cannot withdraw is an asset we may not
+validly POSSESS — so acquiring more of it is the thing to stop."* It fails closed on `None`
+(`guards.py:652-659`), and the attestation feeding it is read live on every intent with a 7-day TTL
+(`keel/execution/executor.py:231-257`, whose docstring is explicit that it is *"never cached"* and
+that an expired attestation is treated as UNKNOWN rather than as `False`).
+
+For a CFD the input to this rail has no honest value at all: there is no underlying, so
+"withdrawable" is not false, it is undefined. But the honest caveat from the cTrader assessment
+applies unchanged and must not be quietly dropped here: `withdrawals_enabled` is the **operator's**
+attestation, not an adapter-reported fact, and the rail inspects no venue. It is entries-only, it is
+one of exactly two rails `offline=True` skips (`guards.py:167`), so paper never runs it — and an
+operator who ran `keel withdrawals attest --enabled` against a CFD venue would clear it. Rail 17 is
+an operator-honesty gate. It is a real gate; it is not a gate the adapter itself could not get past.
+
+**4. Rail 13, USDC funding.** `guards.py:512-541` vetoes a BUY unless the settled quote-currency
+balance covers the **entire** order notional; an unknown balance fails closed, and SELL is exempt
+because *"it produces quote currency, it doesn't consume it"* (`guards.py:513-514`). This rail is
+worth naming separately because it is *structurally* anti-margin rather than incidentally so: the
+entire commercial proposition of a CFD account is posting a fraction of notional. A venue whose
+product only makes sense at less than 100% cash cover meets a rail that only passes at 100% or more.
+The same principle is stated at the design level — *"all sizing uses actual cash only"*
+(`docs/superpowers/specs/2026-07-15-keel-autotrade-design.md:35`) — and implemented that way:
+`keel/execution/sizing.py:22-34` computes size purely as risk-capital over stop distance, with no
+margin, leverage or multiplier term anywhere in it.
+
+**5. Rail 1, the halal allowlist.** `guards.py:368-373` vetoes any intent whose base asset is not in
+`config.allowlist`, in every mode and on both sides. `EUR`, `US500`, `XAUUSD` and `AAPL` are not on it
+— and the path by which they *could* be added runs through the curation screen below, which is the
+point.
+
+**6. The curation screen.** `keel/compliance/screen.py` gates admission to the allowlist, and it
+refuses a Trade Nation instrument twice. First on shape: `screen.py:205-212` runs **rail 19's own
+grammar**, imported rather than restated so the two cannot drift, and the comment at
+`screen.py:196-200` explains that there is deliberately no `ScreenPolicy` knob for it because
+*"spot-only is this agent's charter. A knob whose only safe value is its default is a liability."*
+Second on substance: `screen.py:232-243` requires `backing` ∈ `{ayn, dayn, native}`
+(`screen.py:48-51`) and rejects `dayn` as *"a debt claim on an issuer, not an owned thing"*. A CFD
+is neither an owned thing nor a debt claim — it is a bilateral price-difference contract — so an
+honest attestation lands on the unknown-backing branch at `screen.py:233-237` and is refused for not
+being classifiable at all. `riba_yield` is separately a listed haram sector (`screen.py:42`) and
+`pays_yield` fires at `screen.py:226-230`, which is what an honestly attested overnight-financing
+instrument would trip.
+
+None of this is waivable. `WAIVABLE_CRITERIA = frozenset({"history"})` (`screen.py:60`), and the
+comment above it states that the shariah criteria and `settlement` *"can NEVER be waived"* and that
+expanding the set is *"a deliberate future decision, not a default -- do not add to it to make a
+test pass."* That is enforced three ways: filtering at `screen.py:158` before any branch runs, the
+CLI's `click.Choice(sorted(screen_mod.WAIVABLE_CRITERIA))` on both the grant and revoke commands
+(`keel/cli.py:974`, `keel/cli.py:1023`), and the repository's own note that *"only criteria in
+`WAIVABLE_CRITERIA` are ever honoured, so a row here can never bypass the shariah core"*
+(`keel/data/repository.py:746-748`).
+
+**7. `BrokerCapabilities.asset_classes` — real, but inert, and this document will not overclaim it.**
+`packages/keel-broker-api/keel_broker_api/capabilities.py:21` fixes the vocabulary at
+`frozenset({"spot", "futures", "equity"})`; `__post_init__` raises on anything outside it
+(`capabilities.py:55-57`), and the conformance suite forbids the empty-set dodge
+(`conformance/suite.py:89-98`: *"An adapter that declares nothing declares nothing checkable."*)
+There is no `cfd`, `margin` or `derivative` word to declare, so an honest Trade Nation declaration
+cannot be constructed. But the class's own docstring is unambiguous that this is not a live defence
+(`capabilities.py:28-40`): *"`asset_classes` is **not** what keeps keel spot-only today, and no
+engine code reads it … a gate built on this field would be dead code on every real path while
+reading as a defence."* It is a declaration-integrity check. It belongs in the table above because it
+is the first wall an adapter author hits, not because it is the wall that holds.
+
+**8. The type system, one level up.** Even before the rails, the strategy layer has no words for
+this product. `keel/strategy/rules/base.py:7-9` states the contract in the module docstring —
+*"Long-only spot, no leverage: `Setup.direction` is pinned to `"long"` for v1; bearish setups are
+exit/don't-buy filters, not shorts"* — and `base.py:39` pins it in the type:
+`direction: Literal["long"]`. `Side` is BUY/SELL only
+(`packages/keel-core/keel_core/types.py:25-29`), and SELL is understood everywhere as reducing a
+long: `guards._open_exposure_by_asset` (`guards.py:281-285`) adds BUY notional and subtracts SELL
+notional from a per-asset exposure that is then filtered to strictly positive values. A CFD SELL —
+which opens a short — has no representation in that model. The README says the same to a human
+reader: *"An offline-first, halal (long-only, no-leverage) auto-trading agent"* (`README.md:3`) and
+*"Long-only spot only — no leverage, shorting, or derivatives; sizing uses actual cash, so no riba"*
+(`README.md:58-59`), with *"Nothing overrides a rail — not even autonomy"* at `README.md:44-45`.
+
+### Where the rails' limits actually are — stated plainly
+
+There is a sharp point here that this document would be dishonest to leave out, because a reader who
+finds it later will reasonably conclude the rest was oversold.
+
+Rails 18 and 19 are **string-shape checks**. They are defeated by renaming — but not symmetrically
+across the range, and the asymmetry is worth stating precisely rather than glossed over.
+
+For gold: if an adapter presented Trade Nation's `XAUUSD` CFD as `GOLD-USD`, `parse_spot_product_id`
+would return `("GOLD", "USD")` and rail 19 would pass; `quote_currency_of` would return `"USD"`,
+which is in the default `settlement_currencies`, so rail 18 would pass too. This is the same residual
+`guards.py:725-735` already documents against itself in its own words — `BTC-PERP` *"PASSES this
+grammar"* because the grammar *"cannot know which four-letter tokens are currencies without carrying
+a currency table it deliberately does not carry."* But `GOLD-USD` is a fabrication: no Trade Nation
+platform names the instrument `GOLD` anywhere. Getting past rails 18 and 19 for gold costs a false
+ticker on top of everything else.
+
+For forex it costs less, and this is the sharper case. `EUR/USD` → `EUR-USD` is not a fabrication —
+it is a faithful transliteration. EUR really is the base leg and USD really is the quote leg; nothing
+is invented, only the separator is normalised. `EUR-USD` passes rail 19 (well-formed `BASE-QUOTE`)
+and rail 18 (`USD` is in the default `settlement_currencies`) on that one honest rename alone. So for
+the FX half of the range, the string-shape gates are defeated by **one** lie, not two, and only rail 1
+(`EUR` is not on the allowlist) plus the curation screen (which cannot classify a CFD's backing)
+stand between the renamed adapter and an order — see #202 below for why the screen's refusal is
+itself conditional. Gold needed a fabricated symbol before those gates even engaged; forex does not.
+
+What stops the renamed CFD, on both instruments, is the remaining gates: rail 1, because neither
+`GOLD`/`XAUUSD` nor `EUR` is on the allowlist; the curation screen, which would have to admit it first
+and cannot classify its backing; rail 13, because a margin product cannot post full cash cover; and
+rail 17, which fails closed absent an attestation the operator would have to make falsely. So the
+defence holds — but note precisely *how* it holds, and that it holds by a different margin depending
+on the instrument. **The gold path only "works" through a false product name and a false
+attestation; the forex path needs only the false attestation, since the renamed id is not itself a
+lie.** Rails 18 and 19 stop an accident and an ordinary mistake, not a determined operator who has
+decided to misrepresent the venue to their own agent — and for forex specifically, it is rail 1 and
+the curation screen doing that work alone, not the string-shape rails.
+
+The one genuinely missing piece is already recorded as an open issue rather than papered over here:
+**#202 — "`AssetAttestation` cannot express an instrument wrapper, so the curation screen cannot see
+a CFD."** `AssetAttestation` (`screen.py:76-86`) is keyed on `asset`, a base-leg symbol, with no
+field for the wrapper the asset is traded through. An operator attesting the *underlying* of a Trade
+Nation Bitcoin CFD would record `sector=crypto, backing=native, pays_yield=False` — BTC's existing,
+already-admitted spot attestation — and the screen would admit it, because swap cost, leverage and
+counterparty exposure are properties of the contract and the schema has no way to name the contract.
+The screen's refusal described in gate 6 above is therefore conditional on the operator attesting
+the *instrument* rather than its underlying, and nothing in the schema forces that. That is exactly
+#202's subject, and it is the reason this document does not present the curation screen as an
+unconditional third defence.
+
+## What would have to be true for this to change
+
+Not effort, and not the broker-port migration. Two things could change the verdict, and only two:
+
+**(a) Trade Nation offering genuine unleveraged spot ownership with delivery and withdrawal** —
+share dealing with real custody, or spot crypto in a wallet, or allocated metal — reachable
+programmatically. Nothing in their published range is close to this. They are a CFD and spread-betting
+house by licence, by platform design, by pricing model and by their own recruitment copy; their
+engineering roadmap describes *"pricing engines, liquidity-provider integrations, risk monitoring"*,
+which is the architecture of a dealer, not a custodian. This is not a roadmap item anyone on keel's
+side can influence.
+
+**(b) keel abandoning spot-only.** That is the charter, not a default. Rail 19's comment says there
+is no config field for it because there should not be one (`guards.py:741-742`), and the design spec
+heads the section *"Governing constraint — halal (non-negotiable)"*
+(`docs/superpowers/specs/2026-07-15-keel-autotrade-design.md:30`), hard-excluding leverage/margin/
+borrowing, carry/rollover/funding, short selling and derivatives/futures/perps among the bullets at
+`:35-39`, against a purpose line that reads *"riba-free (interest-free), long-only,
+spot-crypto"* (`:13`).
+
+Note specifically what does **not** get to (a). An elective-professional "Pro" account does not — it
+raises leverage. An MT4 account does not — it trades the same CFDs through a different terminal. A
+future Trade Nation REST API would not — it would place the same orders over a nicer transport. Each
+of these answers a question this document is not asking.
+
+Nor does a **swap-free ("Islamic") account**, and this is worth pre-empting explicitly, because it
+is the most predictable challenge to this document and the one most likely to be raised as though it
+were a fix. Swap-free accounts are near-universal in retail CFD — they exist precisely so a broker
+can offer the product to clients who object to riba on the overnight rollover. What a swap-free
+account removes is the interest differential §56.1 and §66.3 condemn on the *financing* of the
+position. What it does not touch is §28.1's gharar or §65.11's cash-settlement-no-delivery criterion,
+because both are properties of the **contract** — no underlying transfer, settlement through price
+differences only — not properties of how the position is financed overnight. A swap-free CFD is
+still a CFD: no ownership, no delivery, still refused on the same instrument-level grounds this
+document rests on. We could not confirm Trade Nation offers one — no live page surfaced in either
+sitemap, and guessed slugs for one 404 — so this paragraph pre-empts the challenge rather than
+reports a fact about Trade Nation's actual offering.
+
+### (c) A market-data-only integration — answered directly, because it will be asked
+
+The cTrader assessment took this variant seriously and concluded "not barred, but not worth it".
+Trade Nation's case is **weaker on both halves**, and it is worth being clear about why rather than
+copying the previous conclusion across.
+
+On the "not barred" half, the architecture is the same: every gate the refusal rests on is
+execution-side. Rail 17 gates `is_buy`; rails 18 and 19 gate `OrderIntent.product_id` at order-intent
+time; rail 13 gates a BUY's funding; the screen gates admission of a *tradeable* asset. A read-only
+price feed touches none of them, and would not go through the `Broker` port at all —
+`keel/data/market_feed.py` types its entry points on `CoinbaseClient` directly rather than on any
+port interface (`market_feed.py:20` the import, `:74` and `:120` the two signatures), so a
+market-data source is already architecturally a different kind of object in this codebase than a
+`Broker` is. One gate does reach it: `MarketFacts.product_id` runs rail 19's grammar
+at admission (`screen.py:205-212`), and Trade Nation's un-hyphenated symbols fail it — so even a pure
+data feed's candidate ids would have to be renamed before they were screenable, which is the same
+cosmetic-rename problem as above wearing different clothes.
+
+On the "worth it" half, the case is materially worse than cTrader's on two counts.
+
+First, **the marks are worse**. A CFD price is a dealer's quote, not a settlement print — and Trade
+Nation's headline proposition is *fixed* spreads, meaning the quote is a commercial construction
+rather than a passive reflection of an order book. cTrader at least surfaced Level II depth from
+liquidity providers. And the instruments quoted are ones keel does not and cannot trade: `US500`,
+`Germany 40`, `EUR/USD`, `XAUUSD`. keel's live universe is Coinbase USD spot, so this is not even
+cross-venue basis on an asset keel holds — it is reference data for a universe keel has no exposure
+to.
+
+Second, and decisively, **there is nothing to read it from**. cTrader's market-data case rested on a
+documented protocol with published message types, granularities and rate limits; the argument against
+it was cost/benefit. Trade Nation has no public data API at all. The only ways to obtain their marks
+are a private backend #204 forbids probing, or scraping a React application, or an MT4 terminal
+bridge — i.e. the same VPS-plus-EA machinery described above, stood up permanently to harvest
+dealer quotes for instruments keel does not trade.
+
+**Conclusion: no. Unlike cTrader, where declining the data was a judgment call worth revisiting if a
+cross-venue FX reference need ever arose, here there is no viable read path to decline in the first
+place.** If keel ever needs index or FX reference data, the venues to look at are the ones that
+publish a documented data API — several are named in the competitor section above — not this one.
+
+## The alternative, recommended honestly
+
+The question behind #204 is presumably not "Trade Nation specifically" but "should keel have a
+second execution venue". That is a reasonable thing to want, and refusing this candidate should not
+be read as refusing the goal.
+
+The shape that fits is another **spot** venue: real ownership of the asset, a withdrawal path that
+gives rail 17 a truthful input, `BASE-QUOTE` instrument ids that rails 18 and 19 can parse without a
+rename, and an asset class that can be declared in the port's existing vocabulary without inventing a
+word. That work is already underway in this tree —
+`packages/keel-broker-robinhood/keel_broker_robinhood/adapter.py:89` declares
+`asset_classes=frozenset({"spot"})`, with `quote_currencies=frozenset({"USD"})` at `:88` and a
+`supported_orders` set that deliberately omits `market_ioc_quote` rather than declaring a capability
+the venue does not honestly have (`:68-78`). That adapter is the right template for a second venue,
+and the useful next step for anyone who arrived here wanting one is to finish the broker-port
+migration behind it — `keel/commands/_common.py:137-156` still constructs `CoinbaseClient`
+directly — rather than to look for a way to make a CFD house fit.
+
+## What was deliberately NOT built
+
+No `packages/keel-broker-tradenation/` package. No dependency added anywhere in the workspace — no
+MQL bridge, no ZeroMQ client, no MT4 tooling. No entry point registered under `keel.brokers`. No
+conformance test written against a Trade Nation fake. No `pyproject.toml` touched. This document is
+the only file this commit adds.
+
+The reasoning is the one the cTrader assessment set out, and it applies here with more force rather
+than less. This codebase already deleted one piece of speculative machinery for a closely related
+reason: R1's guard rail consulting `BrokerCapabilities.asset_classes`, built and then removed before
+merge because nothing on the live path calls `capabilities()`, so it would have been dead code
+reading as a **defence** (`docs/experiments/2026-08-05-coinbase-asset-class-feasibility.md`; the
+pointer survives in `guards.py:743-747`). An adapter is the mirror image: dead code reading as a
+**capability**.
+
+keel holds no general rule against unwired adapters — `packages/keel-broker-robinhood/` is a live
+counter-example in this same tree, built and tested deliberately ahead of the broker-port migration
+that will use it. The rule is against dead code that misrepresents itself, and the distinction is
+exactly the one this document has been drawing throughout. An unwired Robinhood adapter is a spot
+venue waiting on plumbing; it will become tradeable when the migration lands, and its presence in the
+tree tells the truth about that. An unwired Trade Nation adapter would sit beside it implying the
+same thing — a live option pending only wiring — when the actual state is that every instrument it
+could ever place an order for is refused on charter grounds that no amount of wiring, and no
+migration on any roadmap, will move.
+
+The MT4 route makes this trap sharper here than it was for cTrader, and that is the note to end on.
+Because a working Python path genuinely exists, a Trade Nation adapter would not merely *look*
+plausible in the tree — it would actually run, and place real orders, against a live self-serve
+account, today. That is precisely why it must not exist. The only thing standing between this
+codebase and executing leveraged CFDs is that nobody wrote the adapter, and the right way to keep it
+that way is to leave the reasoning here in prose and the package directory empty.