rene.fichtmueller/transceiver-db
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

docs: changelog entries for equivalence-pipeline audit + switch drift-guard

Browse Source
This commit is contained in:
root 2026-07-17 20:22:33 +00:00
parent bc53c5ebf3
commit 71d828caea
1 changed files with 2 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CHANGELOG_PENDING.md
View File

@ -1,6 +1,8 @@
# TIP Changelog
Format: `{"d":"YYYY-MM-DD","t":"TYPE","m":"Description"}`
{"d":"2026-07-17","t":"FIX","m":"Equivalence pipeline audit: found 2 more sources writing straight to auto_approved without the recent-price/standard_name discrimination catalog-reconcile.ts uses (spec-matcher.ts flat-0.85, no gate) vs one that already had it right (maintenance:find-equivalences enhanced mode implicitly requires standard_name to hit its 0.85 bar and pre-filters recent price). Fixed spec-matcher.ts to insert pending with re_research_due_at=NOW() instead of bypassing review; one-time reclassification of the 1,455 pre-existing spec-basis auto_approved rows (sql/140): 972 held up, 483 demoted (some permanent mismatch, some re-research in 21d for missing evidence)."}
{"d":"2026-07-17","t":"FEAT","m":"Switch-side drift-guard (sql/141): closed the third direction of the physically-impossible-compat invariant (compatibility writes=126, transceiver speed changes=138, now switches.max_speed_gbps/ports_config changes=141). Currently a no-op live (nothing updates those columns today outside the one-time sql/125 backfill) -- preventive, ahead of the planned switch datasheet-enrichment automation. Verified with a synthetic in-transaction test (1250 rows correctly flagged, rolled back)."}
{"d":"2026-07-12","t":"FIX","m":"Switch-DB compat_speed_guard drift gap closed (sql/138): the sql/126 trigger only validated compatibility writes, so a later transceivers.speed_gbps update (e.g. the daily enrich-modulation-fec.sh recovery) never re-checked existing compatible spec_match rows against the switch speed ceiling. Found live: 189 rows had drifted physically-impossible again. One-time remediation (189 rows -> incompatible, backed up to compat_speed_drift_backup_20260712) plus a new AFTER UPDATE OF speed_gbps trigger on transceivers closes the gap for both write directions. check_switch_quality() passes again."}
{"d":"2026-07-11","t":"FEAT","m":"Modulation/FEC enrichment (sql/135): IMMUTABLE helpers derive modulation (NRZ/PAM4/DP-16QAM/DP-QPSK/Coherent), fec_type and inbuilt_fec from speed_gbps + form_factor + part_number/standard_name regex, incl. coherent detection (>=100G-floored ZR/DCO/OpenZR/QAM + CFP2-DCO/ACO via form_factor). Fill-only + idempotent via fn_tip_enrich_modulation_fec(); daily self-healing cron 03:30 (scripts/enrich-modulation-fec.sh). Modulation coverage >=100G rose from 3-8 percent to 46 percent (400G 59, 800G 56, 1600G 100); 4224 rows enriched. Rules built by a 4-lens optical-standards workflow and adversarially verified — the 10G-ZR mislabel, CFP2-DCO miss, AOC/DAC scope leak and bare-QSFP28 over-guess were all caught and fixed before apply. vendor_verified rows never overwritten; enriched rows tagged data_confidence=enriched_estimated + enriched_fields."}
{"d":"2026-07-11","t":"FIX","m":"speed_gbps plausibility guard + cleanup (sql/136): ~22 percent of rows carried junk speed (2129 at <=0, 167 astronomical >1600 values, ~2400 sub-0.09 fragments) from cable/MTP-trunk part numbers parsed into speed. speed_gbps made nullable (unknown speed is honestly NULL); tip_speed_is_implausible() + BEFORE trigger trg_speed_plausibility sanitize junk to NULL on write and never abort ingestion; one-time cleanup nulled 4708 rows (id-keyed reversible snapshot in tip_speed_gbps_rollback). The 0.09..1600 gray band (~355 legit FC/SDI/CPRI/OTN line rates) is intentionally preserved. 87 percent of nulled rows are dac/aoc/accessory where speed is meaningless."}