A citation in the wrong namespace: file the second cause on the class that owns the invariant

[gunbai-bot]

Appends a receipt to unlanded_citation_indistinguishable_at_the_citing_end. One file, no new class, no repairs to other authorities.

Why a receipt and not a new row

Checked the row directory whole at origin/main first. The invariant is already owned: that row rules "a canonical row may cite only what resolves on main at the moment the row lands." A Rust item spelled in .dag dotted syntax violates exactly that rule, so this is a second cause for one invariant, not a new class.

positional_citation is the other near neighbour and is distinct — it owns file:line and text-anchored edits, where the handle is positional by construction.

The specimen

An annotation cited v1.compiler.cli_run run_dag_parse_sweep and v1.compiler.bin.v1_src_dag_parse. Neither module exists or ever will: both are Rust items — a crate module and a cargo bin — wearing .dag dotted-module syntax.

The discriminator, which is why the repairs differ

The timing cause cites a real identity early: it resolves eventually, and all three admissible forms that row ranks work by managing that interval — name the PR and its openness, or describe by shape until it lands.

The namespace cause never resolves. There is no PR to name and no landing to wait for, so none of the three forms applies; the repair is to cite the symbol in its own namespace. Same invariant, different cause, different repair.

It is a rule gap, not four sloppy authors

DESIGN §3 says cite the symbol rather than the position and says nothing about namespace, so a Rust item in dotted form violates no stated rule — and the dotted form is the only citation idiom the prose has, making it the default spelling for a cross-language symbol. That is a defect of the available vocabulary, exactly as this row's filed receipt is a defect of the available word.

The proposed lens trigger is refuted, by the owning row

The proposal was a lens refusing a dotted prefix matching no declared module. That row's own ceiling paragraph already rules it out: deciding whether an identity-shaped token in a paragraph is a citation or a mention is undecidable, so a prose scanner is a heuristic and §5 forbids dressing one as a wall.

The demonstration is self-applying — such a lens would red on the receipt's own paragraph, which names every non-resolving spelling in order to record it. A check whose first false positive is its own class's ledger entry is not a wall.

The trigger is instead a typed cross-language symbol reference. std.decl_ref DeclarationRef carries module_path + decl_name and resolves against .dag modules, so a Rust citation has no typed form to climb into and stays prose by necessity, not by choice. Until that exists the cause sits where the row already puts the class: outside the modeled guarantee, observed and repaired by reading.

Population named, not counted, and not repaired

The members are enumerated by symbol rather than counted — a count of a live population is the transcribed-measurement decay surface §6 forbids, and an identity list is what a later reader can re-join. The instrument (a corpus grep against origin/main, not the working tree) is named so it can be re-derived.

None is repaired. Three separate authorities, not this filing's subject, and a row that silently repairs its own evidence leaves a reader unable to check it.

Honesty

Rung unchanged at mitigatable. The receipt was written by the author who introduced the specimen, two hours after fixing the identical spelling in another PR where a reviewer flagged it — so the census after a correction has to run forward over what is written next, not only backward. The same pass swept the PR body, a second asserting site no diff-scoped check reads.

Verification

• v1_src_dag_parse: 5235 files parse-clean.
• gunbc compile with this row as entry: 0 blocking errors, no diagnostic on the edited module.
• Only pre-existing text change is the closing paren moving to the new final receipt; the sentence it was attached to is preserved verbatim.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MuB8wHjijaFwNoyzm78stF

[briansrls]

Exact-head review of 28e15cc: REQUEST_CHANGES.

The fold into unlanded_citation_indistinguishable_at_the_citing_end is sound. The new specimen shares the row's invariant (a canonical citation that cannot resolve at the citing end), while the cause differs from the timing case; rejecting the proposed prose lens is also correct, because citation-vs-mention is not decidable from an identity-shaped token in prose and the proposed scanner demonstrably reds on the receipt that describes it.

One load-bearing trigger claim is false on the current tree. The new receipt says a typed Rust-symbol reference can resolve against the seed's own declaration index, which the parse sweep already builds. That index does not index Rust items. run_dag_parse_sweep parses every .dag file under DAG_PARSE_SWEEP_ROOTS and constructs one ModuleDeclarationRecord per declared .dag module; its cited-symbol wall resolves authored DeclarationRef literals against those .dag declaration records. A Rust crate module, Rust function, or Cargo-bin identity is outside that index's population by construction.

Keep the capability-grain trigger — a typed cross-language/Rust-symbol reference is a defensible next rung — but do not claim the existing .dag declaration index already supplies its resolver. The trigger must name the missing Rust-symbol authority/producer (or a future cross-language index that includes it) without pre-solving that design. Otherwise the row records another trigger that appears buildable from an existing capability but cannot actually be retired by the construction it names.

I would also narrow or remove the claim that dotted form is the ONLY citation idiom the prose has; that exclusivity is not needed to establish the vocabulary gap. The load-bearing point is that DESIGN/std.decl_ref provides no typed Rust-symbol carrier, not that authors have no textual way to write a Rust path.

Freshness is separate: current main advanced after this head, so any corrected/integrated head is a new subject. No merge action taken.

[gunbai-bot]

Both findings addressed in 0f313544f7, pushed 17:30:09Z — after this review was submitted at 17:19:33Z against 28e15ccde7, so they crossed.

The declaration-index claim. Verified in source before rewriting rather than from the review text: ModuleDeclarationRecord carries module_path, rel_path, declared, variants — .dag module paths and the items authored in them, built by parsing .dag sources. Rust crate modules, functions and cargo bins are outside that population by construction, exactly as stated. The row now records the retraction rather than quietly repointing, and says the Rust-side producer is genuinely unbuilt:

ITS PRODUCER IS GENUINELY UNBUILT AND THIS ROW MUST NOT BORROW ONE, which an earlier revision did and is corrected here rather than quietly: it named the seed's own declaration index as the resolver. THAT INDEX CANNOT SERVE.

The exclusivity claim is dropped — now "the idiom every in-corpus citation uses" rather than "the ONLY citation idiom the prose has." You're right that the vocabulary gap does not need it; it was the stronger of the two available claims and the one I could not have established.

The finding is worth more than the fix, and I'd rather name it than let it pass as a wording nit: this is the third variant of the same trigger-grain defect I wrote today — one trigger naming an already-built capability, one naming a surface the corpus forbids, and this one naming a producer that exists but cannot serve. All three look reachable while nothing they name can retire them. That the third landed inside a row about citations that cannot resolve is the part I should have caught.

On freshness: agreed that an integrated head is a new subject. Holding integration until there is a merge to catch, rather than re-integrating against a base that keeps advancing while nothing can land.

— sent from smart-ant-528