Gate the hardware capability archive

README.md

@@ -113,7 +113,7 @@ The reference scripts in this private workspace are research inputs, not product
 - Step 5 is complete with an allowlisted dry-run cart, exact rollback preview, bounded audit history, deterministic transactions, failure injection, and a strict typed elevated-worker protocol.
 - Step 6 is complete: the first two low-disruption MMCSS settings passed direct and published-worker apply/read/rollback/read gates on disposable Windows 10 22H2 and Windows 11 VMs. NIC/driver writes and the normal UI remain locked while Step 7 starts.
 - Step 7a added a read-only `--probe` mode: it writes a redacted capability report (`socktuner-probe-<timestamp>.json` on the Desktop) that collaborators with real Intel/Realtek NICs can share. Personal data is masked; driver identity, NDIS keywords, defaults, and valid ranges/enums are preserved. The probe changes nothing on the machine.
-- Step 7b unlocks live writes. The **Tuning plan** tab lists the properties the selected driver actually advertises, filtered by intent preset — latency, bandwidth, power and wake, Wi-Fi radio, VLAN and identity — with preview, apply, and exact rollback from the audit history. Capability reports so far cover Intel I226-V and Wireless-AC 3168, Realtek RTL8125 and 8852CE, and MediaTek MT7925.
+- Step 7b unlocks live writes. The **Tuning plan** tab lists the properties the selected driver actually advertises, filtered by intent preset — latency, bandwidth, power and wake, Wi-Fi radio, VLAN and identity — with preview, apply, and exact rollback from the audit history. Capability reports so far cover Intel I226-V and Wireless-AC 3168, Realtek RTL8125, 8852CE and 8922AE, and MediaTek MT7925; the generated [capability archive index](alpha-tester-output/INDEX.md) is the authoritative list.
 
 ## How a change is applied
 

alpha-tester-output/Hyper-V-Virtual-Ethernet-unknown.json

@@ -348,17 +348,17 @@
       "step": 1,
       "registryDataType": 1,
       "canRemove": false,
-      "areas": "Other",
-      "risk": "High",
-      "tradeOff": "Advertised by the driver but not yet characterised by SockTuner. Treat as experimental: change one value at a time, measure, and roll back if behaviour changes.",
+      "areas": "Latency, Throughput",
+      "risk": "Medium",
+      "tradeOff": "RSS spreads receive processing across CPUs and queues. Disabling it can bottleneck one CPU; changing the queue count can add scheduling and interrupt overhead.",
       "settingId": "nic.*NumRSSQueues",
       "isStandardKeyword": true,
       "evidence": "DriverAdvertised",
       "isEnumerated": false,
       "isNumericRange": true,
       "currentDisplay": "8",
       "defaultDisplay": "8",
-      "areasDisplay": "Other",
+      "areasDisplay": "Latency, Throughput",
       "constraintDisplay": "1–64",
       "isModifiedFromDefault": false
     },
@@ -454,17 +454,17 @@
       "step": null,
       "registryDataType": 1,
       "canRemove": false,
-      "areas": "Other",
-      "risk": "High",
-      "tradeOff": "Advertised by the driver but not yet characterised by SockTuner. Treat as experimental: change one value at a time, measure, and roll back if behaviour changes.",
+      "areas": "Latency, Throughput",
+      "risk": "Medium",
+      "tradeOff": "RSS spreads receive processing across CPUs and queues. Disabling it can bottleneck one CPU; changing the queue count can add scheduling and interrupt overhead.",
       "settingId": "nic.*RSS",
       "isStandardKeyword": true,
       "evidence": "DriverAdvertised",
       "isEnumerated": true,
       "isNumericRange": false,
       "currentDisplay": "Enabled",
       "defaultDisplay": "Enabled",
-      "areasDisplay": "Other",
+      "areasDisplay": "Latency, Throughput",
       "constraintDisplay": "Disabled (0), Enabled (1)",
       "isModifiedFromDefault": false
     },

alpha-tester-output/INDEX.md

@@ -20,8 +20,11 @@ default) that the tuning surface uses. Reports captured before schema 12 have on
 | Intel | Intel(R) Ethernet Controller I226-V | 2.1.5.7 | 21 | 21 | 10.0.26200.0 |
 | MediaTek | MediaTek Wi-Fi 7 MT7925 Wireless LAN Card | 5.7.0.4669 | 20 | — (pre-schema-12) | 10.0.26200.0 |
 | Realtek | Realtek 8852CE WiFi 6E PCI-E NIC | 6001.16.172.0 | 13 | — (pre-schema-12) | 10.0.26200.0 |
+| Realtek | Realtek 8922AE WiFi 7 PCI-E NIC | 6102.24.146.0 | 12 | 12 | 10.0.26200.0 |
 | Realtek | Realtek Gaming 2.5GbE Family Controller | 1125.28.20.1224 | 28 | — (pre-schema-12) | 10.0.26200.0 |
 | Realtek | Realtek PCIe 2.5GbE Family Controller | 1125.21.903.2024 | 30 | — (pre-schema-12) | 10.0.26200.0 |
+| Realtek | Realtek PCIe 2.5GbE Family Controller | 1125.31.50.603 | 30 | 30 | 10.0.26200.0 |
+| Realtek | Realtek PCIe 2.5GbE Family Controller | 10.79.50.1003 | 33 | 33 | 10.0.26200.0 |
 
 ## Virtual and filter adapters
 
@@ -34,8 +37,8 @@ Kept for completeness; these are not tuning targets.
 
 ## Coverage
 
-- 6 physical adapter model(s) across 3 vendor(s): Intel, MediaTek, Realtek.
-- 77 distinct NDIS keywords observed.
+- 9 physical adapter model(s) across 3 vendor(s): Intel, MediaTek, Realtek.
+- 82 distinct NDIS keywords observed.
 
 ### Gaps worth filling
 
@@ -52,6 +55,7 @@ Hardware with no report yet, roughly in order of how common it is among the targ
 1. Run `SockTuner.exe --probe` on the machine; it writes a redacted report to the Desktop.
 2. Drop the file into this folder, or pass its folder to the build script.
 3. Run `python tools/build-probe-archive.py` to file it, split it per model and refresh this index.
+4. Run `python tools/build-probe-archive.py --check` to verify regeneration produces no diff.
 
 Reports contain no machine name, IP addresses, routes or full MAC addresses; the vendor OUI
 prefix and driver identity are kept deliberately, since they are the point of the archive.