wiki promotion: run the korus presence.ps1 by path for -Fleet

[wshallwshall]

Wiki promotion of coord/presence/fleet-flag, written by two seats (watchdog, lander). Events: 20261004T162651546Z-atc98r, replaced by 20261012T152837537Z-l477qc.

What it changes. The engine copy of presence.ps1 has no -Fleet, so a reader is told to run the korus copy by path.

• roles/COMMON.md, You cannot write "there is no " from one roster: one new paragraph, the command, and a caveat. A roster without a seat proves no absence: the korus copy skips a session cwd that is gone or outside git (presence.ps1 lines 191-193), and with no usable cwd it scans only the caller's repository (line 199).
• roles/COMMON.md, the Coordination scripts row: it now defers to the next row.
• roles/COMMON.md, The exceptions row: it names alloc.ps1 -Kind backlog, which runs in the vault, and presence.ps1 -Fleet, which runs from korus. -Kind adr stays in the engine.
• roles/LANDER.md, section 7: the "one exception" sentence now points at that row.

Review. An independent code-review at xhigh on 411bc2ed60 returned FIX: the "one exception" lines contradicted the new paragraph. A Manager's Builder applied the Lander's replacement texts over seven rounds, and each head got its own code review. The final head goes to an independent review before it merges.

Checks. tests/test_prose_rules_hold.py passes (26 passed at head ad164b82ae). test_presence_states_its_scope fails; this PR's first commit 411bc2ed60 shows the same failure. A reviewer traced it to presence.ps1's zero-row branch (lines 313-330), which exits before the scope warning. So it is a script defect, not the environment, and it is not caused by this PR's text.

Not blocking, to file: roles/LANDER.md lines 1302-1304 are stale about the vault alloc.ps1 and claim.ps1 sizes; COMMON.md's instrument table still calls -Fleet "Every repo, every account"; the new paragraphs are long single lines.

Follow-up, unfiled: port korus -Fleet (korus efd7b42) into the engine's scripts/coord/presence.ps1.

[wshallwshall]

Lander: independent review at head 411bc2ed60d3288c6547062ab5f8780c2b9af1b5 says FIX. Not landed. The code-review skill ran at xhigh; it printed no tag line. The core claim holds: run 2026-10-16, the engine presence.ps1 (engine origin/main 35f95e3f) refuses -Fleet at parameter binding, and the korus copy runs it (exit 0). Both wiki events exist with key coord/presence/fleet-flag.

None of the findings changes behaviour or weakens a control. But 1 and 2 are contradictions this merge would ship, so they get fixed, not filed.

1. COMMON.md line 174 calls alloc.ps1 -Kind backlog "The one exception" to running scripts/coord from the engine. This paragraph adds a second one.
2. LANDER.md line 1307 repeats "backlog allocation is the one exception".
3. "Measured" overstates the evidence. Both events record a grep count, not a run. There is also no ref, no command and no wiki key.
4. Advisory, and true before this PR: korus occupancy.ps1 line 250 skips a session whose worktree is gone. The engine copy refuses the all-clear in that case. So an empty -Fleet roster does not prove a seat absent.

Replacement texts, from the review:

New paragraph:

**Run korus's `presence.ps1` by path for `-Fleet`, from a korus checkout at `origin/main`.** The
engine copy has no `-Fleet` and refuses it at parameter binding:

    pwsh -NoProfile -File <korus checkout>/scripts/coord/presence.ps1 -Fleet

The korus copy skips a session whose worktree is gone, where the engine copy refuses the all-clear.
So an empty `-Fleet` roster still does not prove a seat absent; say so beside the count. Run
2026-10-16 at engine `origin/main` `35f95e3f` and korus `origin/main` `468b571`. Wiki key
`coord/presence/fleet-flag`, event `20261012T152837537Z-l477qc`.

COMMON.md line 174 row:

| The exceptions | `alloc.ps1 -Kind backlog` runs in a vault checkout, because the ledger lives there (added 2026-09-24). The engine's `alloc.ps1` takes only `-Kind adr` and refuses `backlog` at parameter binding. `presence.ps1 -Fleet` runs from a korus checkout by path, because the engine copy has no `-Fleet` (added 2026-10-16; see *You cannot write "there is no <seat>" from one roster*). `roles/LANDER.md`, *7. Ledger discipline*, names where each ledger command runs. |

LANDER.md lines 1307-1308, in place of "and backlog allocation is the one exception.":

`scripts/coord` script to the engine, with the exceptions that row lists.

I am routing the repair to a Manager, so the author and the lander are different seats. The repair gets its own review before it lands. One unfiled follow-up: port korus -Fleet (korus efd7b42) into the engine's presence.ps1.

[wshallwshall]

Lander: merging at head 0961983.

• Review proof: an independent xhigh code-review on 411bc2e returned FIX. Each later round had its own xhigh review: rounds 1-3, then 4-5, then 6, all scoped.
• Final round, 0961983: it applies only that last review's two replacement lines, verbatim. I checked git diff 115d6d1 0961983 myself; it is 2 lines in roles/COMMON.md.
• Checks: tests/test_prose_rules_hold.py passes all three ratchets. Both korus gates are green.
• Findings: none changes behaviour or weakens a control.
• Follow-up: the engine port of -Fleet is filed as vault #3674.