Say Blocked about a refusal the audit page had been calling Allowed

[kevin9327]

What this changes

The audit page asks "what kind of row is this" twice, from two hand-written lists:

• app/src/routes/_authed/admin/audit.tsx:146-158 and :168 — the refused / failed predicate,
  behind the label at :276-278 and the colour at :269-274.
• app/src/routes/_authed/admin/audit.tsx:36-59 — the eventType query behind the Blocked and
  Did not happen saved views.

Anything on neither list falls through to Allowed. Six event types are on neither, and all six
are written by reachable code today:

event type	written at	drawn as	in Blocked
agent.handoff_refused	server/src/agents/handoff.ts:108	Allowed	no
agent.dial_refused	server/src/index.ts:585	Allowed	no
credential.rotation_refused	server/src/credentials.ts:567	Allowed	no
session.refused	server/src/auth/index.ts:227, :269	Allowed	no
agent.handoff_failed	server/src/agents/handoff-runner.ts:403	Allowed	(should be Did not happen)
agent.escalation_failed	server/src/agents/escalation.ts:123	Allowed	(should be Did not happen)

The comment beside the predicate already states the rule this breaks:

the fallback below calls anything it does not recognise "Allowed", which for a refusal is the one
wrong answer. A trail that is confidently wrong is worse than a silent one.

And server/src/audit.ts:341-347 says of the first row in that table:

BOTH OUTCOMES, and the refused one is the more important of the pair. A hop that happened is
visible in the transcript anyway; a hop that was refused is invisible everywhere else.

It was invisible on the one screen built to show it, and affirmatively labelled Allowed. Same shape
for session.refused, which audit.ts:306-308 calls "the one somebody investigating actually
reaches for", and credential.rotation_refused, of which audit.ts:41-46 says "the refusals are
the interesting ones".

routines.dispatch_refused — the most recently added refusal — was wired into both places, which
is the pattern this follows and the reason I read the omission as an oversight rather than a choice.

The fix

The two lists become one module, app/src/lib/audit/outcome.ts, and the page derives both the row
and the saved view from it. That is the part worth more than the six additions: a refusal on one list
and not the other is either a row drawn as Allowed or a view that is not empty, just short — and the
second is the harder one to notice.

Deliberately unchanged: mcp.call_failed ("The server did not answer"), component.function_failed
("Could not be read") and bot.declined ("The Bot declined") already carry accurate labels through
DECISIONS, so only their colour is muted; that is a judgement about emphasis rather than a wrong
sentence, and I left it to you. The fallback also stays open, so a row type a build has never heard
of is still not called a refusal.

Where it runs

• New state that outlives a request? None. One pure module with two frozen lists and a
  lookup; no component state, no cache, no store.
• What happens on the second replica? The same. This is browser code deciding how to draw a
  row it was handed; the classification is a constant compiled into the bundle.
• Anything serialised? Nothing. The saved views are still a ?eventType= query string
  against the existing read endpoint, now generated from the list instead of retyped.
• Anything fanned out to a browser? No.
• New listener, port, or schedule? None.

Boundary and audit

• Every acting call still goes through the gateway. Nothing on the acting path is touched — this
  is the read surface for rows the server already writes.
• New refusals and new failures each write a row. No new events; six existing ones are now read
  correctly.
• Nothing new is trusted from the client.

Changelog

• CHANGELOG.md, under Unreleased.

Proof

Before. main's predicate and Blocked list, lifted verbatim and run over the six types:

agent.handoff_refused          -> Allowed         in Blocked view: false
agent.dial_refused             -> Allowed         in Blocked view: false
credential.rotation_refused    -> Allowed         in Blocked view: false
session.refused                -> Allowed         in Blocked view: false
agent.handoff_failed           -> Allowed         in Blocked view: false
agent.escalation_failed        -> Allowed         in Blocked view: false

After. app/tests/audit-outcome.test.ts, following the silenceOf / audit-silence.test.ts
pattern — the pure part extracted and tested, no renderHook. It pins each of the six, keeps the
six that were already right, keeps the ordinary rows "allowed", holds the fallback open for an
unknown type, and asserts that each saved view filters by exactly the family it draws (which is the
drift the module exists to stop).

bun run format:check   -> Checked 489 files. No fixes applied.
bun run lint           -> Checked 492 files. No fixes applied.
bun run typecheck      -> app / server / worker all exit 0
bun test app/tests     -> 187 pass, 0 fail (30 files)

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[davidmckayv]

Verified + CI verify green; template-appropriate correctness fix.

[davidmckayv]

CI verify green pre-rebase; CHANGELOG-only rebase; template-appropriate.