[blarghmatey]

What are the relevant tickets?

witan task tk-an-omnigraph-exit-75-503-raises-storeunavailable-dd7ce2, found by the independent review of #573.

Description (What does it do?)

When a remote call exhausts the unavailable retry budget, StoreUnavailable always said witan "could not connect", including when the server answered every attempt with a 503. That sends an operator to the network when the problem is server state (a restart, a graph still loading). This words the message by whether the last attempt reached a server: "<url> answered but kept refusing it as unavailable" vs. "could not connect to <url>".

Implementation details

• _server_answered in witan_core/omnigraph.py treats an attempt as answered when the pooled HTTP transport recorded a status, when the CLI exited 75, or when the error carries one of the 0.13 not-yet-serving wordings (_GRAPH_NOT_YET_SERVING, which 0.13 reports with exit 1).
• Exit 75 as proof the server answered is from upstream v0.14.0: exit_for returns 75 only for effects: none, action: retry, which a data command gets from a server-served disposition or a server admission refusal. A connect or discovery failure carries no code or api_contract_mismatch and exits 1.
• A read cut off mid-flight over HTTP has no status and keeps the "could not connect" wording. That is outside this change.
• Failure classification is unchanged in both _classify_cli_error and omnigraph_http.py, so this does not overlap the 0.14 dispositions rework.

How can this be tested?

• just test-witan-core -k "'exhausted'" runs the four new tests: CLI exit 75 with a 503, a 0.13 loading-graph refusal, an HTTP 503 through _retry_loop, and a CLI connect failure.
• just test-witan-core locally: 910 passed, 7 skipped, 1 failed. The failure is test_the_binary_on_path_is_the_declared_format, because the omnigraph on my PATH reads storage format 14.

https://claude.ai/code/session_01FGZebE9nnZy6bBL3eprDdQ

[Copilot]

🟢 Approval recommended

No active review issues block approval of the diagnostic-wording change.

No open issues 🎉

What changed in this PR

This PR improves witan-core’s StoreUnavailable diagnostics by distinguishing server refusals from connection failures when the retry budget is exhausted. Retry behavior and failure classification remain unchanged.

Changes:

• Selects error wording based on whether the last attempt received a server response.
• Adds four regression tests covering CLI and HTTP refusals, plus connection failures.
• Documents the diagnostic change in a changelog fragment.

File	Description
packages/​witan-core/​witan_core/​omnigraph.py	Detects server responses and updates the exhausted-retry message.
packages/​witan-core/​tests/​test_omnigraph.py	Tests refusal and connection-failure wording across CLI and HTTP paths.
packages/​witan-core/​changelog.d/​20261105_090207_blarghmatey_omnigraph_503_unavailable_wording.md	Records the improved unavailable-store diagnostics.

_{Review effort: Balanced}