Case 5/0: the dispatch that only spoke to the terminal

🩺 Show the full clinical history → 📋 Session protocol →

Clinical record · Software Pathology

The organ worked. It resolved the alias, told the two activation paths apart, applied the extensions gate and diagnosed better than the tree's average. It was implanted inside the terminal's entry point, the one place in the project no other surface passes through.

The double ledger — what it cost, and what it left

After the intervention, dispatch answers through nu -c "use reflection/modules/domain.nu *; ...", which is the exact shape the daemon already uses to call validate and describe. Parity was measured over the tree’s six domains and over one unknown token, which must be returned untouched for the fall-through to keep working.

The case is filed RESOLVED WITH DECLARED DEBT, and the word DECLARED did work the same day: of the three debts written down, two were paid on being read aloud. rg was not declared as a capability — the hole jq had already had — so it was declared, and the cell went from not being able to look to answering through its harness. And the source wrapper gained detection, so code/ontoref pm scope answers where it used to die on nu::parser::extra_positional. Paying the second opened a new hole — two dispatchers, one cell watching one — and the cell now watches both; a check-mutation also revealed that its delegation half was satisfied by a COMMENT, and it is now anchored to the call. One of three remains, the one with no mechanism: the premise imported from the agent’s memory. The capability still reaches 1 of 4 surfaces, and that is sequence rather than debt — the module is the precondition for the other three, not the other three. This case is one of four mistakes made that afternoon; the wider reading of all four — and of why they are ordinary human habits before they are machine failures — is in The mirror that does not flatter.

The point of abandonment — what the table doesn't show

What stopped being watched was not the dispatch: it was the question «where can this be loaded from?». Every capability in this project answered it the same way at some point, by putting its logic in a module. No rule required answering it, so the one capability that did not never had to declare so either.

Differential diagnosis — what was ruled out

Etiology — the cause — The siting, verbatim

# install/ontoref-global — bash
_dispatch_domain() {
  # Two activation paths coexist (ADR-012 repo_kind ⊕ ADR-073 bond; ADR-076 G-1=A).
  local bond_carrier="${ONTOREF_PROJECT_ROOT}/.domains-ontoref/${first_arg}/bonds.ncl"
  ...
  if [[ -z "$activation_path" ]]; then
    echo "  ontoref: domain '${first_arg}' is not available for this project" >&2
    exit 1
  fi
}

The project also had TWO wrappers, and only one carried this policy: grep -c _dispatch_domain code/ontoref → 0. That is why code/ontoref pm scope died on nu::parser::extra_positional while the installed binary answered. Two symptoms, one cause, and neither was legible from the other.

Treatment — The therapy

# .ontoref/reflection/modules/domain.nu
export def "domain dispatch" [domain_id: string, ...args: string] {
  let id  = (domain resolve-alias $domain_id)
  let act = (domain activation $id)
  ...
}

# the same shape the daemon already uses for validate and describe:
$ nu -c 'use reflection/modules/domain.nu *; domain dispatch "pm" "scope"'
  role-scope  /Users/Akasha/Development/ontoref
  ADRs 100 · modes 35        EXIT=0

Detection stays in bash on purpose, with its reason written into the file itself: «is argv[1] a domain?» is asked on every ontoref invocation and must not cost a nu spawn. The policy had no such excuse.

Prognosis

The uncomfortable part is that there was nothing to repair in the moved code: it moved almost intact, and parity was measured domain by domain against the shipped bash. What failed was a siting decision taken once, months ago, that nothing ever asked again — and that nobody could notice by reading the code, because the code worked. The only symptom available was a grep returning zero, and a zero reads as «not yet» rather than as «cannot».

The regimen — the session, replayed with protocol

What was asked — reconstructed from .coder/sessions/2026-08-25-172344-fix-after-coder-session-2026-08-25-0.txt:6 — the opening prompt verbatim, with its line number. The same file holds, at :352, the human refutation that ruled out the first dead end, and at :421 the instruction to move dispatch into the module.

después de la sesión .coder/sessions/2026-08-25-012742-tengo-la-sensacin-de-que-mucho-de-lo-que-hay.txt
no me quedó claro si necesitamos un subdominio en outreach tipo pragma.ontoref.dev
y un directorio propio dentro de ontoref o fuera

What should have been asked

Before opining on where a domain lives, measure which surfaces reach its dispatch today, and
why. And any premise coming from your memory: open it, cite it by file and line, or do not
use it.

The gate before delegating: The capability-reachable-beyond-the-terminal cell written and watched FAILING before a single line was moved. Without it the move is declared done and nothing distinguishes «the policy is in the module» from «the policy is in both places», which is how this defect comes back.

The ADR trigger: The moment it was decided that DETECTION stays in the wrapper and INVOCATION does not. That is not a style preference: it is a rule every future capability inherits, and it separates a per-invocation cost from a policy. It would have fired adr? there, not three steps later.

Prophylaxis — what enforces the lesson today

• ✓
  The rule, argued and with its severity split: the entry point detects, never invokesadr:101/entry-point-detects-never-invokes
• ✓
  The general claim, Soft and naming its absent verifier (no static oracle for «policy»)adr:101/capability-policy-is-not-entry-point-only
• ✓
  Domain dispatch lives in a module, not in the entry pointcontract:.ontoref/biop/capability-reachable-beyond-the-terminal.ncl#nucleus.cmd
• ✓
  The policy withdrawn from the wrapper, with its env contract declared.ontoref/reflection/modules/domain.nu — domain dispatch · domain activation · domain resolve-alias
• ✓
  The wrapper keeps DETECTION and writes down why that is the boundarycode/install/ontoref-global — header of _dispatch_domain
• ✓
  The source wrapper dispatches too: code/ontoref pm scope answerscode/ontoref — _dispatch_domain, delegating to the module, never invoking
• ✓
  The cell watches BOTH wrappers, and its oracle is the call, not the wordcontract:.ontoref/biop/capability-reachable-beyond-the-terminal.ncl#nucleus.cmd
• ✓
  rg declared as a capability: the cell can answer through its harness.ontoref/reflection/capabilities.ncl — capability rg
• ✓
  A memory index hook may not carry an expired fact, nor be missingcontract:.ontoref/biop/memory-hook-carries-no-stale-fact.ncl#nucleus.cmd
• ✓
  An absent memory store reports unknown, never an accusation.ontoref/reflection/modules/checks.nu — «no agent memory store» signature in harness-failure

⊘Declared debt: All three debts were paid, and what remains is a declared LIMIT rather than an outstanding task: the general rule that was asked for — «a hook holds a pointer, never a fact» — has no static oracle, because telling a pointer from an assertion is a reading and not an rg. What IS checked is the failure mode that cost: an EXPIRED fact in a hook, and a memory with NO hook. The limit is declared inside the cell itself, where whoever reads it will meet it.