oregon-stories · semantic index × authority citations
Rules farthest in meaning from their cited authority — a screening queue, not a verdict list
For 24,049 rules, the semantic index scored how much the rule's text resembles the statute it cites as its legal authority. Most sit comfortably (median 0.68; the closest pairs, like a disability-discrimination rule against ORS chapter 659A, reach 0.95). The tail is the interesting part — and the confound is stated before the tail is: a broad enabling statute naturally reads nothing like the specific rules it authorizes. That is lawful and common. So the bottom of this distribution is a REVIEW QUEUE, ordered for human attention, not a list of problems.
How similar are rules to their cited authority? (5th/25th/50th/75th/95th percentiles)
The thirty farthest pairs — the front of the review queue
| rule | its cited authority (ORS) | cosine | authorities cited |
|---|---|---|---|
| 635-049-0195 | 498.002 | 0.426 | 8 |
| 582-115-0060 | 344.710 | 0.439 | 4 |
| 582-080-0040 | 344.710 | 0.443 | 4 |
| 808-002-0390 | 671.525 | 0.446 | 3 |
| 820-001-0035 | 672.325 | 0.457 | 2 |
| 635-049-0073 | 498.002 | 0.457 | 8 |
| 695-047-0080 | 541.890 | 0.459 | 2 |
| 736-080-0070 | 390.770 | 0.460 | 4 |
| 461-125-0150 | 411.116 | 0.460 | 5 |
| 695-015-0080 | 541.956 | 0.463 | 1 |
| 291-133-0025 | 137.285 | 0.464 | 5 |
| 943-012-0025 | 654.010 | 0.464 | 2 |
| 635-049-0205 | 498.002 | 0.467 | 8 |
| 859-050-0105 | 161.295 | 0.467 | 2 |
| 839-014-0185 | 658.730 | 0.468 | 1 |
| 291-133-0015 | 137.285 | 0.469 | 5 |
| 695-047-0070 | 541.890 | 0.470 | 2 |
| 333-029-0045 | 446.350 | 0.471 | 3 |
| 950-060-0090 | 414.665 | 0.471 | 1 |
| 411-069-0150 | 409.750 | 0.471 | 1 |
| 635-049-0255 | 498.002 | 0.471 | 8 |
| 736-002-0042 | 390.300 | 0.472 | 2 |
| 808-003-0850 | 671.570 | 0.473 | 1 |
| 411-069-0120 | 409.750 | 0.474 | 1 |
| 950-060-0120 | 414.665 | 0.475 | 1 |
| 635-049-0235 | 498.002 | 0.477 | 8 |
| 333-275-0180 | 670.280 | 0.479 | 2 |
| 695-047-0100 | 541.890 | 0.479 | 2 |
| 813-310-0095 | 456.506 | 0.480 | 2 |
| 695-047-0110 | 541.890 | 0.480 | 2 |
Read this before quoting the chart
Low similarity is not wrong authority. The dominant, innocent explanation — general delegations ('the department may adopt rules…') — probably accounts for most of the tail, and only reading the pair tells you which is which. That reading has a home: the conflict-candidate triage workflow in executive-regulatory-frameworks, where a human verdict is recorded per pair and carried forward. Scores are cosine over per-document mean embeddings (BAAI/bge-m3); each rule is scored against the FIRST statute in its authority line (rules citing many authorities are marked). Nothing on this page is served by the platform's resolve/authority tools — it is a screening artifact, cached with its fingerprint.