|
| 1 | +# Config-Change Investigation Reviewer Pack |
| 2 | + |
| 3 | +This page describes the small, sanitized reviewer pack for `config-change-investigation-demo`. |
| 4 | + |
| 5 | +Pack directory: |
| 6 | + |
| 7 | +- `docs/reviewer-pack-v0.6.0/` |
| 8 | + |
| 9 | +The pack is suitable for release attachment or offline review. All examples are representative sanitized examples aligned with the demo's synthetic sample semantics. They are non-production artifacts and do not contain secrets, machine-specific paths, or deployment claims. |
| 10 | + |
| 11 | +## Pack Contents |
| 12 | + |
| 13 | +| File | Path | What It Proves | |
| 14 | +| --- | --- | --- | |
| 15 | +| `MANIFEST.md` | artifact index | The pack scope, source type, and intended reviewer use | |
| 16 | +| `benign-change-example.json` | benign change with no investigation | A benign config change remains visible in normalized input but does not become an investigation | |
| 17 | +| `risky-change-with-evidence-example.json` | risky change with nearby evidence | A risky config change becomes an investigation and carries bounded supporting evidence | |
| 18 | +| `bounded-case-no-evidence-example.json` | risky change with bounded case but no nearby evidence | A risky config change still produces an explicit investigation even when bounded correlation finds zero nearby evidence | |
| 19 | +| `investigation-summary-example.json` | reduced summary path | The reduced summary stays deterministic and preserves the same evidence counts and bounded-correlation explanation | |
| 20 | + |
| 21 | +## Benign Change With No Investigation |
| 22 | + |
| 23 | +Relevant file: |
| 24 | + |
| 25 | +- `docs/reviewer-pack-v0.6.0/benign-change-example.json` |
| 26 | + |
| 27 | +Inspect: |
| 28 | + |
| 29 | +- `change_event.config_key` |
| 30 | +- `change_event.new_value` |
| 31 | +- `reviewer_expectation.appears_in` |
| 32 | +- `reviewer_expectation.not_expected_in` |
| 33 | + |
| 34 | +Reviewer conclusion: |
| 35 | + |
| 36 | +- A benign change is kept in normalized input context, but it does not match a risky-change rule and should not appear in `investigation_hits.json` or `investigation_summary.json`. |
| 37 | + |
| 38 | +## Risky Change With Nearby Evidence |
| 39 | + |
| 40 | +Relevant files: |
| 41 | + |
| 42 | +- `docs/reviewer-pack-v0.6.0/risky-change-with-evidence-example.json` |
| 43 | +- `docs/reviewer-pack-v0.6.0/investigation-summary-example.json` |
| 44 | + |
| 45 | +Inspect: |
| 46 | + |
| 47 | +- `investigation.rule_id` |
| 48 | +- `investigation.correlation_window_minutes` |
| 49 | +- `investigation.evidence_counts` |
| 50 | +- `investigation.attached_policy_denials` |
| 51 | +- `investigation.attached_follow_on_events` |
| 52 | +- `summary_record.summary` |
| 53 | + |
| 54 | +Reviewer conclusion: |
| 55 | + |
| 56 | +- A deterministic risky-change rule creates an investigation, and bounded correlation attaches only nearby evidence that shares the same `target_system` and falls inside the configured window. |
| 57 | + |
| 58 | +## Risky Change With Bounded Case But No Nearby Evidence |
| 59 | + |
| 60 | +Relevant files: |
| 61 | + |
| 62 | +- `docs/reviewer-pack-v0.6.0/bounded-case-no-evidence-example.json` |
| 63 | +- `docs/reviewer-pack-v0.6.0/investigation-summary-example.json` |
| 64 | + |
| 65 | +Inspect: |
| 66 | + |
| 67 | +- `investigation.evidence_counts` |
| 68 | +- `investigation.attached_policy_denials` |
| 69 | +- `investigation.attached_follow_on_events` |
| 70 | +- `investigation.bounded_correlation_reason` |
| 71 | + |
| 72 | +Reviewer conclusion: |
| 73 | + |
| 74 | +- A risky change stays explicit as an investigation even when bounded correlation finds zero nearby denials or follow-on events. The demo does not silently discard this case. |
0 commit comments