output-format.md

  1# Output format
  2
  3Crosscheck uses semi-structured Markdown: stable headings and labeled bullets,
  4but readable prose over rigid JSON.
  5
  6## Runner capability report
  7
  8```markdown
  9# Runner capabilities
 10
 11## Prose Complete harnesses
 12
 13### <harness name>
 14
 15- Status: available | unavailable | uncertain
 16- How discovered:
 17- Help inspected:
 18- Model-list command inspected:
 19- Notes:
 20
 21## Runners
 22
 23### <runner name>
 24
 25- Status: available | unavailable | skipped | uncertain
 26- Prose Complete harness:
 27- Command:
 28- Invocation profile:
 29- Help inspected:
 30- Model-list command inspected:
 31- Safety classification:
 32- Gaps:
 33
 34## Candidate model roster
 35
 36### <exact candidate model ID>
 37
 38- Prose Complete harness:
 39- Runner:
 40- Invocation profile:
 41- Provider/gateway:
 42- Model family:
 43- Generation:
 44- Variant:
 45- Availability: listed | user-pinned | uncertain
 46- Evidence:
 47```
 48
 49## Lane report
 50
 51```markdown
 52# lane-1
 53
 54- Status: active | skipped | failed
 55- Runner:
 56- Invocation profile:
 57- Model target:
 58- Model family:
 59- Focus:
 60
 61## Findings
 62
 63### F001
 64
 65- Claim:
 66- Location:
 67- Evidence:
 68- Impact:
 69- Confidence:
 70- Suggested fix direction:
 71- Uncertainties:
 72- Assumptions:
 73
 74## Coverage Notes
 75
 76## Non-findings / rejected concerns
 77```
 78
 79## Candidate ledger
 80
 81```markdown
 82# Candidate Ledger
 83
 84## C001
 85
 86- Source lanes:
 87- Claim:
 88- Location:
 89- Evidence:
 90- Proposed severity:
 91- Proposed confidence:
 92- Related candidates:
 93- Uncertainties:
 94- Assumptions:
 95```
 96
 97## Lane plan
 98
 99```markdown
100# Lane plan
101
102## Reviewer lanes
103
104### lane-1
105
106- Status: active | skipped
107- Runner:
108- Invocation profile:
109- Model target:
110- Model family:
111- Focus:
112- Safety class:
113- Validator invocations:
114
115## Validator invocations
116
117### validator-1
118
119- Covers reviewer lane:
120- Runner:
121- Invocation profile:
122- Model target:
123- Model family:
124- Workspace copy:
125- Reused validator profile: yes | no
126```
127
128## Validation report
129
130Allowed verdicts: `accept`, `reject`, `weaken`, `needs-human-check`,
131`duplicate-of`, and `out-of-scope`.
132
133```markdown
134# Validation report
135
136## C001
137
138- Verdict:
139- Validator evidence:
140- Reasoning summary:
141- Confidence adjustment:
142- Notes:
143```
144
145Novel validator concerns go under `Possible follow-up`, not into findings.
146
147## Final review packet
148
149```markdown
150# Crosscheck
151
152## Summary
153
154## Findings
155
156### F001: concise title
157
158- Severity:
159- Confidence:
160- Candidate IDs:
161- Source lanes:
162- Validator status:
163- Evidence:
164- Suggested fix direction:
165
166## Needs human judgment
167
168## Rejected or downgraded candidates
169
170## Coverage
171
172- Reviewed workspace:
173- Runners discovered:
174- Lanes active/skipped:
175- Validator invocations planned/run:
176- Skipped runner reasons:
177- Model families represented:
178- Execution mode:
179- Guardrail caveats:
180- Validator mode/degradation:
181- Safety violations:
182
183## Limitations
184
185## Suggested next steps
186```