44 drivers emit numbers they never read (null, never 0) #169

Problem

44 of the 91 catalog drivers still fill a field with a number after their device stops answering. In total that is 592 fields. A read that failed or came back short becomes a reading of 0 W, 0 A, 0 % SoC or 50 Hz, and nothing downstream can tell it from a real one:

• A meter at 0 W looks like a site in balance.
• A battery at 0 % looks empty and ready to charge.

sigenergy 1.1.3 is the worked example, for the battery it reported on a plant with no storage.

The rule is null, never 0. A value whose read failed or came back short is nil, so the host sends null. A real 0 that was read is still 0.

How it is measured

drivers/tests/lua_harness/no_invented_numbers.lua (#168) runs a driver twice, with different register values, numeric config and clock. It then makes every read fail for three polls, and makes every Modbus reply half its length for three more. A number still emitted that is equal in both runs and was not read is invented. The data model's default SoC window (0.05 / 1.0) is the one constant allowed. Sourceful's driver registry runs the same file on every publish and refuses a driver that fails it.

drivers/tests/test_no_invented_numbers.py holds this debt in invented-number-baseline.json. New drivers must be clean, and a count can only go down.

What each driver needs

The usual cause is a default set before the read, or a fallback after it:

local l1_v = 0                                -- before the read
if regs then l1_v = decode(regs[1]) * 0.1 end

local w = host.decode_i32_be(r[1], r[2]) or 0 -- after it

The fix:

1. Start every measurement as nil, and set it only from a register or response that answered in full.
2. Drop or 0 (and or 50, or 230) defaults on decoded values.
3. Emit a field only when it has a value. Emit a DER only when at least one of its fields does, as sigenergy 1.1.3 and ambibox do.
4. Per driver: make invented-number-report ID=<id> lists the fields. When it is clean, remove the driver from invented-number-baseline.json and bump its version with make bump-driver.

Notes

• ctek, ctek_hybrid and ctek_v2 emit ev.phases = 3 from config while the charger is down. This is a configured installation fact, not a measurement. Leave it out of the emit when nothing was read, or emit it only alongside real readings.
• hello is a demo driver that emits fixed values. Either remove it from the catalog or keep it in the baseline as an example.
• zaptec_cloud is not measured: its driver_init needs credentials, which the probe does not supply.

Drivers and fields (main 224c438)

🤖 Generated with Claude Code