[Senate] Resolve open PRs #913 + #984 given v1 freeze
Task
- ID: e32ca3cd-f4a4-4aab-901e-cf92fe9bca2e
- Type: one_shot
- Layer: Senate
Effort: deep
Goal
Two substantial PRs from the 2026-04-27/28 session are still open on
2026-05-18 — both work-complete on branch, both blocked from merging.
Decide their disposition.
**SciDEX PR #913 — Pantheon (rounds + LLM metadata + synthesizer
artifacts + @-mention picker + convene UX)**:
- Branch:
fix-pantheon-metadata-rounds
- 5 distinct feature commits all on branch.
- The **backfill + Migration 146 + frontend synthesizer behavior
ran on prod** during the session (live
pan_b6c942aa page renders
with
unresolved=true|false flags and corrected round numbers).
- The code for those features is not on main. Some scripts
(
scripts/abandoned_run_audit.py,
scripts/orphan_checker.py
--once) leaked through via other PRs, but the Pantheon-specific
pieces (LLM-metadata plumbing, synthesizer resolver, @-mention
picker, convene-UX changes) didn't.
- Conclusion: prod is running on either a) the unmerged branch
directly, or b) a separate Pantheon-backend deploy. Either way the
v1 main checkout doesn't reflect what's live.
SciDEX PR #984 — MCP unified registry:
- Branch:
mcp-unified-registry.
- All 8 migration steps complete on branch.
- 57 tests pass on branch.
- Deletes 4 legacy files (
scidex_mcp_server.py,
scidex_pantheon_mcp_server.py,
scidex/agora/mcp_tools.py,
tests/test_pantheon_mcp_server.py).
- Those files still exist on main — so anyone reading main
currently sees the pre-refactor MCP layout.
The 2026-05-18 audit noted: "v1 may be frozen as of 2026-05-13;
active development is in v2 substrate/prism repos." If that's
correct, then chasing v1 merges is moot — the work either applies
to v2 or is dead.
What this task does
Confirm the freeze state of v1. Source: AGENTS.md (post the
re-trim in spec
445ecead-...),
.orchestra/config.yaml, or
direct operator query. Document the answer.
If v1 is NOT frozen:
- Rebase both branches onto current
origin/main.
- For PR #913: run the pantheon test suite + smoke-test the
synthesizer + manually verify the @-mention picker in a fresh
debate. Merge.
- For PR #984: re-run the MCP test suite; verify
python3 -m scidex.mcp boots; update
.claude/mcp.json to
use the unified binary. Merge.
- Decommission
scidex-orphan-check.service if 6+ successful CI
cycles of the orphan-check recurring task (
bb654176-...) are
visible in
task_runs.
If v1 IS frozen:
- Inventory the v2 substrate's pantheon module. Determine which
of the 5 PR #913 features (rounds, metadata, synthesizer,
@-picker, convene UX) need to port across.
- Same for v2's MCP equivalent — does it have an analogous
registry? If yes, decide whether to port the alias-cleanup or
not.
-
Close PRs #913 + #984 with a comment linking to the v2 PRs
that supersede them, OR with a "obsolete due to v1 freeze"
reason that the next reviewer can trust.
Either way: prevent the runtime drift between main and the
live Pantheon backend by writing a deployment-audit script that
verifies the deployed Pantheon backend's git SHA matches a
reachable on-main commit. Drift > 7 days should open a Senate
triage task.
Acceptance criteria
☐ v1 freeze status documented in a short note under
docs/planning/.
☐ Both PRs either merged or closed with a clear disposition
comment.
☐ If closed: the work isn't lost — either ported to v2 or filed
as v2-side specs.
☐ Deployment-audit script (or equivalent inventory) lands so the
next "live behavior doesn't match main code" surprise is caught
fleet-side, not by manual operator audit.
Why priority 75
Below the watchdog-message-clarity work (which was 95) but well
above typical research tasks. Unmerged but-shipped-in-prod PRs are
a credit-risk: anyone reading main for "what does Pantheon do?"
gets a wrong answer.
Dependencies
445ecead-... AGENTS.md re-trim — should land first so the
freeze status can be recorded there cleanly.
Lineage
- Closed-unmerged: SciDEX #1014, #1225.
- Open: SciDEX #913, #984.
- Originating session:
docs/planning/2026-04-27-orchestra-session-summary.md.