E.19:4.3.1 - Tiny golden case bank for regression and worked examples
Select a case that exercises the changed entry risk. Cases 1–4 specialize I.2.4, I.2.2, I.2.6, and I.2.3 respectively; E.11 governs the entry-distribution use, and the direct subject patterns govern the recovered claims. Cases 5–6 add search and retrieval stress under PCP-ENTRY-E1 and PCP-ENTRY-E4. Another relevant E.11/I.2 case may be used. Select empirical evidence under PCP-ENTRY; unselected cases need no run or absence note.
The tempting_wrong_pattern_or_wrong_relation column is conditional on an observed or independently plausible reader mistake that changes selection or use. Leave it unused when that condition is absent; keep the case’s positive recognition and admissible stop.
| Case | case_signal | expected_first_entry_pattern_comparison_set | candidate_patterns | tempting_wrong_pattern_or_wrong_relation | admissible_entry_stop | companion_or_projection_functions_that_help | projections_that_do_not_define_semantics |
|---|---|---|---|---|---|---|---|
| 1 | “we need a shortlist, not one winner” | pattern-comparison set for comparison, pool treatment, and selected-set result declaration | A.19.CN, A.17-A.19, C.18, C.19, G.0, and G.5 when selected-set result declaration is claimed | treating C.11 as one one-off choice when the real entry-recognition function is selected-set result declaration or candidate-set stabilization | admissible candidate-pattern set stabilised or selected-set result declaration opened | README scenario or E.11 entry-distribution cue, one pattern Problem frame, one expanded entry-disambiguation case if compact cues still fail | one README blurb, one thin echo, one lexical-query row alone |
| 2 | “we have a vague cue, not yet a claim” | pre-articulation cue pattern-comparison set | C.2.LS, A.16, A.16.1, B.4.1, B.5.2.0 | forcing the cue into one endpoint-claim, quality, or assurance pattern too early | entry-recognition-reclassified or cue preserved for the admissible next entry-recognition function | README scenario or E.11 entry-distribution cue, one pattern Problem frame, one case-linked I.2 expanded entry-disambiguation case when needed | one coarse public entry projection alone |
| 3 | “this is the same EntityOfConcern re-expressed for another audience” | same-EntityOfConcern rewrite pattern-comparison set | A.6.3.CR, A.6.3.RT, E.17.EFP, E.17.ID.CR | changing the EntityOfConcern or creating a competing semantic rule track merely to serve another audience | wrong-pattern-rejected or same-EntityOfConcern rewrite opened | one expanded entry-disambiguation case, one pattern Problem frame, governing-entry pointer | one parallel explanatory blurb treated as one second pattern body |
| 4 | “the API says X” | boundary-claim unpacking pattern-comparison set | A.6, A.6.B, A.6.C, A.6.P, C.16.Q, A.6.A, E.17 | treating one boundary phrase as one agent duty, promise, quality verdict, or generic agreement paragraph without atomic claim assignment or quality-term repair with recovered characteristic and scale | boundary-claim-pattern-opened, quality-term-repair-exited, or atomic claim set opened | one boundary-focused E.11 entry-distribution cue, one pattern Problem frame, one expanded entry-disambiguation case where interface/access/confused-quality wording is common | one query cue or public entry projection treated as the governing entry |
| 5 | “I found a pattern by search, but I am not sure it is the right one” | one pattern-local recognition-signature case under the selected pattern-comparison set | one candidate applicable direct pattern body plus one case-near related pattern when needed | one lexical near-match or same-family pattern without governing-entry fit | non-use-confirmed or pattern-selected | one pattern Problem frame, one E.11 entry-distribution cue, one lexical-query hook | one search-query row alone |
| 6 | “the LLM retrieved a helpful-looking paragraph but not the pattern” | one retrieval-facing first-entry pattern-comparison case | one applicable direct pattern body plus one applicable projection function | one stale thin echo or one projection-only companion function answered as if it were the governing entry | governing-entry-opened or expanded-entry-disambiguation-case-needed | one governing-entry reference, one projection-only status marker, one retrieval-facing pointer to the applicable direct pattern body | one thin echo chunk without governing-entry reference or projection-only cue |
In case 3, evaluate episteme identity separately under C.2.1: its discriminators are claim content, EntityOfConcern, and effective reference scheme. Changed wording or audience alone leaves that identity unchanged; a changed discriminator identifies another episteme. A.6.3.CR permits claim-preserving or explicitly loss-declared same-EntityOfConcern re-expression under its own use conditions.
Selected cases test:
- entry-recognition consistency;
- wrong-pattern or wrong-entry rejection;
- admissible entry-stop honesty;
- lexical-query discipline;
- thin-echo retrieval hygiene;
- and governing-entry and projection separation in the changed entry text.
When one empirical or retrieval evidence run is selected, keep recoverable the facts needed to understand its question and result. A structured result may use the following field names, including only those needed by that run:
viewpoint_class
task_prompt_or_query
expected_governing_entry_or_admissible_candidate_set
near_miss_patterns_or_projection_functions_if_any
time_budget_if_relevant
success_criterion_if_relevant
success_or_failure_note
observed_failure_mode_if_any
rationale_or_repair_action
When retrieval evidence is selected, keep retrieval result, answer
faithfulness, and stale-echo result distinct without forcing benchmark-style
reporting on ordinary edits.
Use the retrieval questions in PCP-ENTRY-E4, including its conditional
thin-echo reference check.
Ordinary local guidance stays prose-only rather than minting one stable
governing-entry reference by default.