fix: le dossier dit la verite sur le code, et la garde peut echouer

Cinq constats, aucun sur la frontiere des droits : le dossier et la garde.

LE MOTIF ENREGISTRE SUR `copyWithin` ETAIT FAUX, et c-etait raisonne, pas joue -
dans le meme lot que `rule_justify-a-return-value-by-running-it`, dont la
premiere application est a son propre voisin. Mesure : le vrai `readOnlyArray`
ne leve PAS, il execute et DETRUIT le premier element (["A","B","C"] devient
["B","C","C"]), tandis que `push`/`splice`/`sort`/`reverse`/`fill`/`shift`
levent a la lecture. `copyWithin` echappe a cette garde pour la raison exacte
qui le fait nommer ici : il est dans la meme liste mal etiquetee dont la garde
est faite. Le refuser est donc PLUS protecteur que la cible, pas equivalent :
une divergence deliberee, consignee comme telle, avec la mesure a cote.

LE FICHIER DOCUMENTAIT ENCORE LA REGLE QUE CE ROUND A SUPPRIMEE, et se
contredisait : deux en-tetes decrivaient le defaut par TYPE ("un compte, un
drapeau passent") et disaient que `isLoading` n-est pas nomme, alors que la
ligne 759 le nomme. Le contrat gardait les deux politiques a la fois. Une
doctrine qui survit a une reecriture est precisement ce qui egare le suivant.

LA GARDE NE VERIFIAIT PAS CE QU-ELLE ANNONCAIT. Prouve : une copie verbatim avec
{copyWithin, slice} en exceptions et `slice` retire de la liste servie passait
1 pass / 0 fail. Elle compare maintenant une PARTITION - l-ensemble effectif
amont egale l-union, les trois parts sont disjointes, l-exception vaut exactement
un membre nomme - et lit aussi les entrees SYMBOLE, invisibles a l-ancienne
expression. Verifie rouge dans les trois directions.

LA DEROGATION ECMASCRIPT ETAIT INFONDEE TELLE QUE RAISONNEE : la specification
ferme les CLES, pas les VALEURS. Un `Symbol.toStringTag` implemente en accesseur
rendant `Set(${size})` publiait le compte non filtre. Les deux cles sont
calculees a partir de la forme enveloppee. Et ce que la vue divulgue par
ailleurs - l-existence d-une cle sur la cible - est dit exactement, avec la
mesure qui montre qu-aucune presence de cle ne depend des objets.

`reduce` branchait sur `init === undefined` au lieu de l-arite.

.project/concepts/app-contract/polyfill-surface/contract_polyfill-surface.md

@@ -260,7 +260,7 @@ Where a call must first find out whether something already exists — a document
 
 **TWO COPIES OF THIS PACKAGE IN ONE APPLICATION MISBEHAVE SILENTLY, and nothing here can detect it.** What this package remembers it keeps in the package itself, not in any handle you hold: which documents are subscribed and who is listening for them, which identity the session settled on, which documents are open. None of that is shared between two copies. An application that ends up with two therefore runs two of everything — a document subscribed through one copy is invisible to the other, so changes to it simply never arrive; and the identity settled in one is not the identity the other acts as, so the same call writes as one user or the other depending on which copy it reached. Nothing raises, nothing warns, and no call can report it: from every entry point a second copy is indistinguishable from the first, and the symptoms surface far from the cause as missing updates and writes attributed to the wrong person. This is a packaging property, not a behaviour to code around — **declare this package once, as one dependency of the application**, and let your own code and every library built on it resolve to that copy. A library that builds on this package should declare it a *peer* rather than a dependency for exactly this reason, so that the application remains the one place it is named.
 
-**No unfiltered read through `useShape`, and no way round it through the object it hands back.** What the call returns is NextGraph's own result, filtered: `data` and `promise` carry only items in documents you hold the key for, `isLoading` passes through, the two page commands run and whatever they hand back is held to the same rule as any other member, and `subscription` - the ORM's own handle, which reads and writes with no capability filter at all - is refused out loud rather than handed on. **Any member this package has not DECIDED is refused, whatever its type** - that is how a member a later `@ng-org/orm` adds is settled before anyone has seen it. It is not decided by type: a COUNT is a plain number that describes the items, and nothing separates it from a page size, so nothing undecided passes. A member the SDK does not carry at all reads `undefined`, since there is nothing behind it. Inside `data`, every member that yields items is filtered, and an ordered result addresses and counts the items you may read: `size`, `length`, an index, `Object.keys`, `in` and `JSON.stringify` all agree, and none of them says how many you may not. **Reading a member's DESCRIPTOR answers exactly what reading the member answers**, which is not a detail: `Object.getOwnPropertyDescriptor` - and the spread, `Object.entries` and console inspection built on it - reads a property without going through the view's own read, so a refused member throws there too rather than handing its value over. That holds for an ACCESSOR as much as for a value: the getter you are handed reads through the view, and its setter is refused. **A mutation reaches the collection underneath, and neither hands back nor acts past what the view shows.** Writing is not broken - your own `add` reaches the document and the reactivity it drives. **All three ask the same question**: `add` THROWS for an object naming a document you hold no key for, and for one naming NO document (there is then nothing that owns it, and it would land in a collection every identity shares - set `@graph` to a document you hold) (it returns the collection whether or not it did anything, so it has no value meaning "that did not happen"), `delete` of an item the view does not show answers `false` - the Set contract's own word for "it was not there", and true of the view - and `clear` empties the view rather than the collection every identity shares. `add` hands back the VIEW rather than the collection, because `Set.prototype.add` returns the collection so calls can chain and chaining onto the unfiltered one gives it away entirely. **Writing to what you are handed is refused** - `data[0] = x`, `delete data[0]`, `data.length = 0`, `Object.defineProperty`, `Object.setPrototypeOf` and `Object.freeze` - on the RESULT OBJECT as much as on `data`, and on this package's own account rather than because the ordered collection upstream happens to refuse some of them too. **Every member `@ng-org/alien-deepsignals` declares non-mutating answers**, computed over the items you may read: the iterator helpers, the array reads (`at`, `slice`, `join`, `indexOf`, `toSorted`, `with`) and the ES2025 set combinators (`union`, `difference`, `isSubsetOf`). The one exception is `copyWithin`, which that list names and which WRITES; upstream's own ordered collection throws on it too, and so does this. A document reached through that view alone, read nowhere else first, does not appear.
+**No unfiltered read through `useShape`, and no way round it through the object it hands back.** What the call returns is NextGraph's own result, filtered: `data` and `promise` carry only items in documents you hold the key for, `isLoading` passes through, the two page commands run and whatever they hand back is held to the same rule as any other member, and `subscription` - the ORM's own handle, which reads and writes with no capability filter at all - is refused out loud rather than handed on. **Any member this package has not DECIDED is refused, whatever its type** - that is how a member a later `@ng-org/orm` adds is settled before anyone has seen it. It is not decided by type: a COUNT is a plain number that describes the items, and nothing separates it from a page size, so nothing undecided passes. A member the SDK does not carry at all reads `undefined`, since there is nothing behind it. Inside `data`, every member that yields items is filtered, and an ordered result addresses and counts the items you may read: `size`, `length`, an index, `Object.keys`, `in` and `JSON.stringify` all agree, and none of them says how many you may not. **Reading a member's DESCRIPTOR answers exactly what reading the member answers**, which is not a detail: `Object.getOwnPropertyDescriptor` - and the spread, `Object.entries` and console inspection built on it - reads a property without going through the view's own read, so a refused member throws there too rather than handing its value over. That holds for an ACCESSOR as much as for a value: the getter you are handed reads through the view, and its setter is refused. **A mutation reaches the collection underneath, and neither hands back nor acts past what the view shows.** Writing is not broken - your own `add` reaches the document and the reactivity it drives. **All three ask the same question**: `add` THROWS for an object naming a document you hold no key for, and for one naming NO document (there is then nothing that owns it, and it would land in a collection every identity shares - set `@graph` to a document you hold) (it returns the collection whether or not it did anything, so it has no value meaning "that did not happen"), `delete` of an item the view does not show answers `false` - the Set contract's own word for "it was not there", and true of the view - and `clear` empties the view rather than the collection every identity shares. `add` hands back the VIEW rather than the collection, because `Set.prototype.add` returns the collection so calls can chain and chaining onto the unfiltered one gives it away entirely. **Writing to what you are handed is refused** - `data[0] = x`, `delete data[0]`, `data.length = 0`, `Object.defineProperty`, `Object.setPrototypeOf` and `Object.freeze` - on the RESULT OBJECT as much as on `data`, and on this package's own account rather than because the ordered collection upstream happens to refuse some of them too. **Every member `@ng-org/alien-deepsignals` declares non-mutating answers**, computed over the items you may read: the iterator helpers, the array reads (`at`, `slice`, `join`, `indexOf`, `toSorted`, `with`) and the ES2025 set combinators (`union`, `difference`, `isSubsetOf`). The one exception is `copyWithin`, which that list names and which WRITES: it is refused here, and that is STRICTER than NextGraph's own ordered collection, which runs it and destroys an element (its guard is built from the same mislabelled list). A deliberate difference, stated so you do not discover it as a surprise. A document reached through that view alone, read nowhere else first, does not appear.
 
 **`isLoading` does not mean what its name says, and the fault is upstream's.** `@ng-org/orm`'s declaration describes it as "`true` when no data is available yet"; the hook that ships computes it as the subscription's `isReady`, which is the inverse. Both artifacts were read and they disagree. This package forwards the value it is handed rather than correcting it, because a layer that quietly improves the target teaches you a behaviour you would have to unlearn. **Branch on `data === undefined`**, which is unambiguous, and treat `isLoading` as unusable until upstream settles it.
 
@@ -290,10 +290,10 @@ What each level means here, in this package's own terms:
 
 - **Read your objects through `data`, not by iterating what the call returned.** `@ng-org/orm`'s React hook returns `{ data, promise, subscription, isLoading, nextPage?, previousPage? }` around the collection, and it has for every release this package builds against. This package went on treating the return AS the collection: `[...result]` and `result.size` threw, and `result.data`, `result.promise` and `result.subscription` each handed back **every identity's items**, with nothing raised anywhere. If you were reading `.data`, you were reading unfiltered. That is the defect this version closes, and it is why this entry is worth reading before upgrading rather than after.
 - **`subscription` is REFUSED where it used to be handed over.** It is the ORM's own `RdfOrmSubscription`, which reads and writes with no capability filter at all and which `## Non-guarantees` has always said this package does not publish. Reading it now throws, naming itself. `data`, `promise`, `isLoading` and the two page commands are what remains, and they are what a caller needs: the SDK's own example destructures `data` and `isLoading` and nothing else.
-- **A member of that object that this package does not know is refused unless it is a plain scalar.** The rule is the value's TYPE, not its name: whatever can carry an item is an object or a function, so a member the next `@ng-org/orm` adds is already decided, while a flag or a count still passes. The previous rule - "anything that is not a function is harmless" - is precisely what let three members of the new shape through, so it has been replaced rather than extended with three names.
+- **A member of that object that this package has not DECIDED is refused, whatever its type.** All six of the members `@ng-org/orm` ships are decided; a seventh that a later version adds refuses until someone decides it. Two earlier rules were tried and both let a leak through - "anything that is not a function is harmless", which handed over three members of the current shape, and "a plain scalar cannot reach an item", which handed over any count not spelled `size` or `length`. There is no test on a value's type that separates a count derived from your data from a page size, so the rule is not about types at all: it is that nothing undecided is answered.
 - **The two arguments are the SDK's own types**, where the signature took `unknown` for both. `useShape(shapeType: ShapeType<T>, conf: RdfOrmConfig<T> | string | undefined)`, which is what the surface already said the pair was. A caller passing the SDK's shape and config compiles unchanged; one passing something else now does not, which is why this counts.
 - **`size` and `length` count what you may read, and so does every other way of asking.** An ordered config (`orderBy`) makes the collection an array, which says how many items it holds in four places - `length`, `Object.keys`, `in`, and the descriptor of each index - and only the first was filtered. `data[0]` now names the first item you may read rather than the first the engine returned, and the positions past them do not exist.
-- **A member's DESCRIPTOR answers what reading it answers.** `Object.getOwnPropertyDescriptor(result, "data")` returned the collection UNFILTERED, and `...(result, "subscription")` the handle that reading it refuses: that call does not go through the view's own read. Neither does a spread, `Object.entries`, `JSON.stringify` or a console inspection, so all of them were reading round the filter. They now read through it - which also means `{...result}` and `console.log(result)` throw on `subscription`, where they used to print it. Log `data`.
+- **A member's DESCRIPTOR answers what reading it answers.** `Object.getOwnPropertyDescriptor(result, "data")` returned the collection UNFILTERED, and `...(result, "subscription")` the handle that reading it refuses: that call does not go through the view's own read. Neither does a spread, `Object.entries`, `JSON.stringify` or a console inspection, so all of them were reading round the filter. They now read through it - which also means `{...result}` and `Object.keys(result)` THROW on `subscription`, where they used to hand it over. **But do not read that as "logging is filtered".** A console is the HOST's, not this package's: `console.log(result)` in Bun and in a browser devtools inspects the object BEHIND the view and prints `data` and `subscription` in full, for every identity, and no interception can prevent it - measured. That is a property of runtime introspection, not a hole a version can close. **Do not log or inspect what a shape read hands you when more than one identity shares the page**; iterate `data` and log what you mean to show.
 - **A mutation no longer hands back the collection, nor acts past the view, and all three ask the same question.** `add` returned the collection - `Set.prototype.add` returns `this` so calls chain, measured on the plain `Set` and on the real `DeepSignalSet` alike - so `view.add(x)` gave every identity's items to a caller whose every way IN was filtered, through the first member the ORM's own example reaches for. It now hands back the view, and it now REFUSES an object naming a document you hold no key for: ungated, it let you plant an object into someone else's document, never see it again, and have the two-way binding commit it there. `delete` removed an item the view declines to confirm the existence of; `clear` emptied the collection every identity shares. **Your own writes still reach the document** - that is the one thing that must not change, and `e2e/orm-live-read.ts` measures it against the deployed broker.
 - **A write to the collection itself is refused.** `data[0] = x`, `delete data[0]` and `data.length = 0` all went through against a collection that allowed them, the last defeating the emptying rule outright; `Object.setPrototypeOf(data, ...)` replaced the real collection's prototype, after which iterating it throws for every identity, and `Object.freeze(data)` made it non-extensible. The view refuses all of them itself now, rather than relying on the ordered collection upstream refusing some of them for it.
 - **The members upstream declares non-mutating answer again.** Eighteen array reads on an ordered result, plus `includes` and the seven set combinators, were throwing for the holder whose data it was - a leak traded for an outage. They are served over the items you may read.