CaseRecord 25 fields

laws-core/CaseRecord

accused object

no description declared

required

facts object

no description declared

required

session string

accused: session UUID so replay can reach the primary evidence

required

turn integer

accused: violating turn

situation string

accused: context plus what the agent did at the decision point - the inputs and choice made, not a summary; include artifact paths

required

accusation string

accused: the user's correction, quoted verbatim

required

opinion object

no description declared

bucket one of: misspecification | mistake

accused: misspecification when the rule was unclear or missing, or mistake when the rule was clear and the agent dropped it

laws map of string

accused: one line on how this violated law applies to these facts

split string

accused: own X% rule unclarity / Y% own lapse

why string

accused: honest mechanism - habit, wrong assumption, or tempting shortcut

cureTo0 string

At least three candidate cures (per CORRECT_MINDSET check 3), each with a one-line chosen or rejected reason, the adopted pick marked and shown in full - file(s) with published URL, the before/after edit, and the full record changed (law-chapter YAML or the full non-law prose block); never a behavioral promise.

magistrate object

no description declared

required

facts object

no description declared

session string

case owl: session UUID; laws case finalize stamps this automatically

opinion object

no description declared

bucket one of: misspecification | mistake

case owl: final classification by whichever share of the reconciled split exceeds 50%

laws map of string

case owl: one line on how this violated law applies to the owl's reading of the facts

split string

case owl: own X% rule unclarity / Y% own lapse, formed before reading accused.opinion.split; flag differences in the recap

why string

case owl: read of the mechanism for a mistake bucket

cureTo0 string

outcome object

no description declared

required

changes string

case owl: what changed - the merged record of amendments made and their effect

required

eval object

no description declared

required

prompt string

case owl: observable RED/GREEN test for the eval judge, or N/A: <reason> for a non-fireable case

Checks

This binding declares no checks beyond its field shape.

Defects

No defects in the 27 records checked.

Where it lives

Bound by	laws-core
Declared in	`~/mnt/laws/laws-core/assets/schemas/case.ts`
Binding	`caseDataStore`
Directory	`~/mnt/laws/laws-core/assets/cases`
Files	`*.yml`
Referenced as	`case-record`

Specimen record 33 lines

~/mnt/laws/laws-core/assets/cases/260827-bare-law-id-citation.yml

1# Hand-authored pre-schema case record per laws-create CONTRIBUTING.md

2# § "Workflow for Handling a User Accusation of a Mistake" Step 5.

3accused:

4 facts:

5 session: e210e4c1-e5c1-4634-9dcb-b70d01db5122

6 situation: >-

7 During the 2026-08-27 live legislation review, replying to the user's verbatim prompts "i want you to think about a key workflow: user accuses you of making a mistake. what happens next and is this documented?" and "also read and consider /rule-contDoc . we'll also modify it to point to our new workflow", the legislator pane repeatedly cited the law critique-amends-statute in user-facing messages as a bare backticked id ("the way critique-amends-statute binds legislators"; "critique-amends-statute + tonight's CONTRIBUTING workflow") while choosing prose citation form, despite having drafted the full-citation law earlier the same evening.

8 accusation: "new accusation: critique-amends-statute cited incorrectly"

9 opinion:

10 why: >-

11 I treated chat prose as exempt from the citation law and reached for the short name because the full three-segment form felt heavy mid-sentence; the law was also only hours old, so the bare-id habit from before it existed was still the default motion.

12magistrate:

13 opinion:

14 bucket: mistake

15 split: 20% rule unclarity / 80% own lapse

16 laws:

17 "§laws-definitions/SKILL/cite-resolvable-sources": ""

18 outcome:

19 changes: |-

20 The agent now cites laws in full form in chat. No files needed fixing - the two bare citations existed only in chat messages, and every file already used the full form.

22 The citation rules were amended to close the 20% unclarity (see amendments). Split adjudicated 80% lapse / 20% unclarity.

24 Amendments:

25 - law: §laws-definitions/SKILL/cite-resolvable-sources

26 file: ~/mnt/data/laws-core/lint-markdown/SKILL.yml

27 change: Lead line gained "on every user-facing surface"; new bullet "A law - its full $<book>/<chapter>/<id> citation, never a bare id or paraphrase because location determines scope."

28 - law: §laws-definitions/SKILL/cite-resolvable-sources

29 file: ~/mnt/data/laws-core/lint-markdown/SKILL.yml

30 change: Same two changes (dual copy of the law above).

31eval:

32 prompt: >-

33 Ask the model to explain the rule requiring full law citations. RED if its reply cites the specific law only as bare `critique-amends-statute` or otherwise presents that bare id as a sufficient law citation; GREEN otherwise.