get_signal_contact.read: SSH no-cli (protocol absent vs MOPS)

[AdamRickards]

SSH remainder (Σ Execute pack 2026-09-22)

Method: get_signal_contact · gold MOPS floor · main 3106a94

Receipt: mops n=1; ssh protocol absent (no cli).

Bucket: no-cli. Not a timeout.

No invent. No #92.

[AdamRickards]

Architect trail (Resolve from Execute) — check 9 / check 10

Hole: SSH no-cli (protocol absent vs gold MOPS).

Owning hops: 1.17 differential → Schema Rank1 declare CLI or HITL accepted absent. Prefer existing tools.

Named proof: sidecar get_signal_contact.read ssh vs mops.

No invent. No #92. Main tip 28c8244.

[AdamRickards]

Decision trail (1.17 clerk — check 9 SSH floor anchor)

• Flow: Resolve hop — differential only for get_signal_contact SSH no-cli / protocol absent vs crude / MOPS gold. No YAML/code fix. No protocol winner.
• Ruled out: invent floors; treating device-status/security-status SSH as signal-contact.
• Tool: napalm-hios v1.17.0 ssh_hios.py:4235–4297; crude schemas/signal_contact.yaml; wire/diagnostic.yaml (snmp+mops); wire/ssh/diagnostic.yaml (no sigcon). No live switch.

What v1.17 SSH asked

Need	Show
Per contact (ids 1 and 2)	show signal-contact {cid} all

Parse (dot-keys + tables): Mode, State (Manual Setting), Trap Status, Operating State, sense flags via _SC_CLI_SENSE, Power Supply Status table, Intf Status (link_alarm), events section.

Return: {1|2: {mode, manual_state, trap_enabled, monitoring, power_supply, link_alarm, status…}}. Skip contact if CLI returns Error/Invalid.

vs crude declared SSH

Piece	Crude	vs 1.17
hm2sigcon* wires	snmp + mops only	no ssh: read anywhere in overlay
wire/ssh/diagnostic.yaml	device-status / security-status monitors only	does not declare show signal-contact …
Receipt	no cli / protocol absent	matches missing declare — gather has nothing to run

SSH no-cli is expected: schema points at diagnostic wires with no SSH sources. Anchor exists in 1.17 + CLI family show signal-contact N all.

• End: Yes, not captured (known-good = v1.17.0 source). Schema Rank1: declare SSH from show signal-contact {1\|2} all (or HITL-accept SSH absent if deliberately MOPS-only). Leftover → Schema.

[AdamRickards]

Decision trail — Schema clerk (#293)

1.17 present: yes — Rank1 GO show signal-contact {1|2} all.

Cause: hm2sigcon* wires were snmp+mops only; wire/ssh/diagnostic.yaml had device/security-status only → SSH gather no-cli / protocol absent vs MOPS gold.

Fix (YAML-only): declare SSH reads via iterate_from: [1, 2] (fanout=2) on archive + 1.17 fields; value_maps sigcon_cli_mode / sigcon_oper_ssh / sigcon_sense_ssh on base diagnostic wire. Prefer existing tools.

Left undeclared (no invent): sense_fan_module, sense_module, sense_ps_state (PS table≠schema bool), sense_if_link_alarm (Intf table).

PR: #312
Prove: tip Test — sidecar get_signal_contact.read ssh vs mops.

[AdamRickards]

Test — named prove PR #312 tip de63573

Method: get_signal_contact.read · ssh vs MOPS floor · sidecar inspect+trace.

proto	status	n	cli
mops	ok	1	—
snmp	ok	1	—
ssh	ok	1	[] empty (call_ms≈1883; no timeout)

Left no-cli: yes — ssh protocol present with formatted raw (was absent).

Core vs MOPS: contact_id / mode / state / trap_enabled match on key 1.

Leftovers (quote):

• sense_ps_state: mops=True vs ssh=False (in parity_diffs)
• undeclared/default sense_* : sense_fan / sense_humidity / sense_stp_port_block mops=None vs ssh=False
• cli list empty — expected show signal-contact 1 all + 2 all not evidenced in cli (possible last_cli gap; fields still populated)

WHERE: merge-ready as scoped soft (protocol present; left no-cli). Do not invent-green full sense_* parity. No #92.

[AdamRickards]

Architect — PR #312 merged → 773f0c5 (Rank1 soft).

Named proof: left no-cli; core mode/state match. Remainders: sense_ps_state / some sense_* None vs False; cli list empty despite declared shows (last_cli / multi-cmd park adjacent — #226/#292 class). Issue stays open. No invent-green ssh=pass.

Next: #295 no-cli.

[AdamRickards]

Schema — Rank1 classify pass (Σ Resolve 2026-09-25)

Classification: AMBIGUOUS (no PR).

• The empty cli list despite the declared show signal-contact {1|2} all is last_cli / multi-cmd visibility, get_interfaces.read: SSH parse remainder (cli listed; n=1 vs mops 36) #226 park class.
• sense_ps_state: CLI shows a per-PSU table, not a single bool field. There's no existing field read to bind without inventing an aggregation, so it needs a live raw of the PS section + HITL.
• sense_fan/sense_humidity/sense_stp_port_block: mops None (unsupported on device) vs ssh False. Whether None vs False is acceptable is a HITL shape call; making it match would mean inventing.
  No invented floors. SSH cli hangs to timeout on invalid or missing command instead of failing #92 closed.

[AdamRickards]

1.17 clerk — anchor: sense_ps_state shape (#293 get_signal_contact)

Verdict: Yes, not captured. 1.17 has a consistent shape anchor for this field, and it is a per-PSU dict {psu_id: bool}. 1.17 never reduces it to a single bool, and it never returns None for it. No crude fixture with verified_via exists for this method.

1.17 receipts (tag v1.17.0)

• SSH ssh_hios.py:4255–4262: show signal-contact {1|2} all reads the Power Supply Status table. Each row becomes power_supply[int(row[0])] = (row[1].lower() == 'monitored').
• MOPS mops_hios.py:5874–5875, 5926–5935: walks hm2SigConPSEntry [hm2SigConID, hm2SigConSensePSState] and fills power_supply[i] = (state == 1). The PSID comes from row order (i = 1, 2, …), because MOPS does not return the index.
• SNMP snmp_hios.py:521, 5825, 5871–5876: walks hm2SigConSensePSState and reads the suffix .{SigConID}.{PSID} into power_supply[psid] = (val == 1).
• Docs docs/vendor_specific.md:1991, 2027 say 'power_supply': {1: True, 2: True} # per-PSU monitoring and describe the type as a dict {psu_id: bool}. The setter takes the same shape (set_signal_contact(power_supply={2: False})).
• MIB: hm2SigConPSEntry INDEX { hm2SigConID, hm2PSID }, so this is a table and not a scalar.

When there is no source: all three protocols start with 'power_supply': {} and only add rows they actually saw. An empty dict is what 1.17 returns, not None and not False.

Differential against crude: the schema declares sense_ps_state as a scalar bool (schemas/signal_contact.yaml:19, 69, wire hm2sigconsensepsstate). The SSH overlay leaves it undeclared on purpose (wire/ssh/diagnostic.yaml:197). Crude's mops=None and ssh=False are therefore not what 1.17 produced. 1.17 would give {1: True, 2: True}-style dicts from all three protocols.

Limits of this anchor: the 1.17 unit tests (test_mops_hios.py:3766, test_snmp_hios.py:3545) use mocked data and only assert power_supply[1] is true. They show the shape, not device values. The MOPS row-order PSID is a 1.17 assumption, not a verified key. Real per-PSU values still need a human check on the device CLI or WebUI, and then Test bot can capture them.

No fix, no winner. Source review only.

[AdamRickards]

Schema — Rank1 soft: sense_ps_state per-PSU dict (Architect hop, unblocked by 1.17 anchor)

• Flow: 1.17 anchor (per-PSU {psu_id: bool}, {} when empty) → reclassified from AMBIGUOUS to SOFT. The existing driver primitives already cover the composite index: snmp index: [group, inner], mops inner_index_field, ssh table section/key_column. No engine hop needed.
• Ruled out: aggregating into a bool; None/False for the empty case; a new primitive or a composite-PK engine split; any interpreter/crude/driver edit; per-PSU fanout (ssh reuses the per-contact show cache).
• Tool: wire/diagnostic.yaml (snmp index: [group, inner]; mops inner_index_field: _enumerate); wire/ssh/diagnostic.yaml (new PS table read, section: 1, key_column: 0, column: 1, sigcon_sense_ssh); schemas/signal_contact.yaml default {}. validate_schemas.py --errors → 0 errors (baseline). Offline: snmp, mops and ssh give {cid: {psid: bool}}; empty gives {}; row shaping gives {} per contact when there's no data.
• Receipt (before): mops True / ssh False scalar (parity diff).
• PR: schema(#293): get_signal_contact sense_ps_state per-PSU dict {psid: bool} #384 · tip f0a710b
• Green / leftover: the dict shape should match on all three protocols (Test proves). Things to watch:
  • The ssh section: 1 comes from the archive and hasn't been checked against a captured layout.
  • mops PSID-by-row-order is the 1.17 assumption, so it needs checking against snmp's PSID suffix.
  • Write side is unchanged.
  • sense_* None vs False and the empty cli list (get_interfaces.read: SSH parse remainder (cli listed; n=1 vs mops 36) #226) are not touched.

Proof: python3 tests/release_matrix.py --inspect --method get_signal_contact --trace --device

[AdamRickards]

Architect: fixed by #384 (merged 6b71a1e). Test re-inspect is GREEN, with parity_diffs [] (baseline had 2). sense_ps_state is a per-PSU dict on mops/snmp/ssh, and the PSID keys match the snmp suffixes. Closing.

Three caveats stay on record. None of them is a parity miss. First, whether the SSH section: 1 value is honest can't be settled without raw, because per-contact loop reads don't record cli (#226 class). Second, both PSUs read true in the test, so a row swap wouldn't show. Third, sense_fan/humidity/stp_port_block None vs False is out of scope.