@@ -108,3 +108,31 @@ open follow-up in `000_index.md`. This removes one route to it, not the default.
108108All four pass, ` bun run typecheck ` clean, cursor suite green on ` ssh lidge ` ,
109109pushed. Test 1 demonstrated red on the pre-fix tree.
110110
111+
112+ ## Shipped
113+
114+ Commits ` f145fd513 ` (fix + tests) and ` c9681d043 ` (review follow-up).
115+
116+ The plan's tagging design survived implementation, with one correction the audit
117+ forced and one it surfaced afterwards:
118+
119+ - ** The guard is ` !expectedClose && !emittedTerminal ` ,** not ` !expectedClose `
120+ alone. Tagging after a terminal was already queued would have added a second
121+ one, and buffered JSON processes both — flipping a completed turn to failed.
122+ - ** The typed error carries its own message** , because a raw ` NGHTTP2_CANCEL `
123+ string is re-matched by ` classifyCursorError ` and reported as an intentional
124+ "Cursor stream suspended". A turn that failed unexpectedly would have claimed a
125+ deliberate suspension, which is worse than the silent drop it replaced.
126+ - ** It also re-exposes the originating transport code.** Wrapping hid it from the
127+ ` turn-failed ` diagnostic, leaving the one summary that exists to explain this
128+ failure without an error code. The regression test pins the non-obvious
129+ consequence: carrying ` NGHTTP2_CANCEL ` back onto the wrapper must not make
130+ ` isCursorBenignCancelError ` match it again — provenance is checked first.
131+
132+ Implementation note: ` classifyTurnFailure ` is a closure beside ` summarizeFailure `
133+ rather than a method, because ` summarizeFailure ` is itself a per-run closure over
134+ the turn's state. A method could not reach it.
135+
136+ Verified on ` ssh lidge ` at ` c9681d0435 ` : typecheck clean, 630 pass / 0 fail.
137+ Red-before-green demonstrated, and independently reproduced by the reviewer as
138+ 4 pass / 1 fail under a mutation that disables the classification.
0 commit comments