From 3894d2b4eb8339d98efa9358d74ab7ca794675bc Mon Sep 17 00:00:00 2001 From: bong-water-water-bong Date: Tue, 10 Nov 2026 20:11:19 -0400 Subject: [PATCH] tests/census_watch: derive "reviewed classes are documented" from significant.json The arm asserted the rule for a hand-written three-name dict, so the file it is about could grow to 37 classes with no row in docs/arch-gaps.md and the test stayed green. It now loops over the registry's reviewed classes, with a floor so an empty load cannot pass vacuously, and the failing line names the classes that are missing a row. That change found one entry already in breach: DeepseekV4FlashSuperExpertForCausalLM (added by engine#1922) was the single undocumented class out of 37, so it gains the row the contract asks for, with the reason it is not an alias of deepseek_v41 (AW-SECD-compressed weights, which nothing at the pin decodes). Both arms were checked by mutation: without the doc row the check fails naming the class, and with an empty registry it fails on the floor rather than passing. --- docs/arch-gaps.md | 12 ++++++++++++ tests/census_watch_test.py | 10 ++++++++++ 2 files changed, 22 insertions(+) diff --git a/docs/arch-gaps.md b/docs/arch-gaps.md index 8f7d5fdb5..aa1d2cb0a 100644 --- a/docs/arch-gaps.md +++ b/docs/arch-gaps.md @@ -1075,3 +1075,15 @@ which are retried on every run; and the `!! FLM` lines. Those keep the 2026-11-0 and `smolvla` as above, and `!! NPU FLM` is engine #588. This run's issue window also includes the `gpt-oss` (`GPT-OSS-20B`, `GPT-OSS-Safeguard-20b`) and `lfm2` (`LFM2-1.2B`, `LFM2-2.6B`) Q4NX packs. Their mapping is HRX-only too, so they belong to #588, not to a registry edit. + +## 2026-11-10: `DeepseekV4FlashSuperExpertForCausalLM` — reviewed, and recorded here + +Found by the check this page's contract implies and `tests/census_watch_test.py` now enforces: every class in +`registry/significant.json` has to be cited here, and this one was added there (engine#1922) without a row, so it was +the single undocumented entry out of 37. + +* `DeepseekV4FlashSuperExpertForCausalLM` (`model_type` `deepseek_v4_flash_superexpert`) — DeepSeek V4.1 Flash + SuperExpert, **published as remote code and with AW-SECD-compressed weights** (8.5 GB, claimed 0.962 retention); + MLA-style geometry (`kv_lora_rank` 512, `q_lora_rank` 1536, `qk_rope_head_dim` 64, `v_head_dim` 128, 32 heads / + 16 KV, IM 4096). **Not an alias of `deepseek_v41`**: nothing at the pin decodes AW-SECD weights, so mapping it to + that architecture would name a backend that cannot read the file. Reported, not alerted (`"watch": "report"`). diff --git a/tests/census_watch_test.py b/tests/census_watch_test.py index 04f04b906..69270cde7 100644 --- a/tests/census_watch_test.py +++ b/tests/census_watch_test.py @@ -311,6 +311,16 @@ def fetch(url): check("reviewed non-alias classes are documented", all(cls in gap_docs for cls in reviewed_gaps) and "not an alias" in gap_docs) +# engine#2014 review: the list above is hand-written and names three classes, so it keeps covering those three +# while the file it is about grows (37 entries today) -- a fourth, fifth and thirty-seventh class can be added +# with no row in docs/arch-gaps.md and this file stays green. significant.json's own `about` says its classes +# are documented there, so derive the list from the file: a missing row then fails here instead of resting on +# three names that nobody will remember to extend. The floor stops the loop passing vacuously if the file were +# ever loaded empty -- a check that iterates nothing must not report success. +undocumented = sorted(cls for cls in sig if cls not in gap_docs) +check(f"every one of the {len(sig)} reviewed classes is cited in docs/arch-gaps.md" + + (f" -- missing: {', '.join(undocumented)}" if undocumented else ""), + len(sig) >= 30 and not undocumented) # 14. A reviewed class marked `"watch": "report"` is listed but does not alert (engine#381). check("GrugMoeForCausalLM is reviewed as report-only",