Resolve High-Impact Unknowns — 4.0

Mission

Do not merely list high-impact unknowns. Close them where the repository can answer them.

Inputs

Read:

00_open_questions.md
workflow_state.yaml  # runner-owned authoritative state for the active workflow
00_investigation_coverage.md
00_current_understanding.md
00_evidence_ledger.md

Load only relevant detailed artifacts per question.

Prioritization

Process:

CRITICAL first
then HIGH

Within equal impact, prefer questions that block multiple downstream claims or tasks.

For Each Question

1. Restate the Missing Fact

Make it specific and falsifiable.

Bad:

Checkout timeout source is unknown.

Better:

Which configuration source, key, default, and override chain determines the checkout timeout used by the Python order worker?

Also make provenance/exposure questions falsifiable.

Examples:

Under which checkout request conditions is the promotion code retrieved, normalized, and returned in the order response?

Which REST endpoints, message consumers, or scheduled jobs can reach the C++ pricing-engine boundary after enumerating every source-visible invocation site?

2. Determine Investigation Domain

Dispatch one or more skills:

configuration source / mode
→ /configuration-source-trace

runtime implementation / DI / factory binding
→ /runtime-binding-verification

persistence / DB schema / DAO
→ /persistence-schema-reverse-engineering

end-to-end operation behavior
→ /runtime-flow-analysis

integration routing
→ /integration-selection-analysis

field/data provenance
→ /runtime-flow-analysis plus targeted source trace through retrieval → mapping → output

external exposure from a known downstream boundary
→ boundary → exposure closure using caller/reference tracing

other
→ targeted source trace using system-reverse-engineer rules

3. Search From Multiple Anchors

Do not rely on the original search that produced UNKNOWN.

Use alternate anchors:

• consumer backward
• source/config forward
• symbol references
• implementations
• startup/bootstrap
• tests
• migrations/scripts
• all direct call/construction sites
• caller fan-out from impacted downstream boundaries
• serializers/mappers/parsers
• related constants/enums

4. Record Search Coverage

Before unresolved closure, record what was actually searched.

4A. Provenance / Exposure Closure Requirements

When the unknown concerns externally meaningful data, do not resolve it until the material chain is closed:

trigger
→ source/downstream retrieval
→ source data
→ parse/transform
→ mapping/state
→ serialization/output
→ observable result

When the unknown concerns external reachability of a known downstream boundary:

boundary
← ALL direct invocation/construction sites
← ALL caller chains
← external entry points / proven internal roots

Do not resolve reachability based on one representative endpoint.

If one direct invocation site remains unresolved, the question remains:

SOURCE_SEARCH_PENDING

or SOURCE_SEARCH_IN_PROGRESS.

5. Resolve

Use one:

RESOLVED_VERIFIED
RESOLVED_REJECTED
NOT_APPLICABLE
EXTERNALLY_BLOCKED
SOURCE_SEARCH_PENDING

EXTERNALLY_BLOCKED requires:

• repository search sufficiently exhausted
• alternate anchors used
• provenance chain traced as far as source allows when data/output-related
• all source-visible invocation sites classified when reachability-related
• code-side boundary/injection point identified when possible
• exact missing external artifact/value/evidence named

6. Promote If Appropriate

Resolved architectural facts still pass /claim-verification-and-promotion before canonical promotion.

Required Closure Matrix

Loop

Continue until:

HIGH/CRITICAL SOURCE_SEARCH_PENDING = 0

or context/tool limits require PARTIAL_RESUMABLE.

If limits are hit, persist exact next Q-ID, unresolved provenance/exposure link, search anchor, and skill to run.

Output

Create/update:

docs/reverse-engineering/09_high_impact_unknown_closure.md

Also update:

00_open_questions.md
workflow_state.yaml  # runner-owned authoritative state for the active workflow
00_investigation_coverage.md
00_evidence_ledger.md

Do not print the full report to console.

4.0 Source-Fact Boundary

RESOLVED_VERIFIED applies to the source question, not to downstream design recommendations.

NO CURRENT MECHANISM FOUND
!=
SPECIFIC FUTURE MECHANISM VERIFIED

If implementation guidance is useful, label it PROPOSED and do not promote it into the evidence ledger as source truth.

4.0 Run-Scoped State Isolation

When this skill is invoked by a workflow whose state_path is under docs/reverse-engineering/runs/<run_id>/, do not create or update ``workflow_state.md` or any second state file. The runner owns the single authoritative workflow-declared YAML state file.