Session population freeze must share the facts workspace qualification

Traceability

• Spec ID: 34-session-population-topology-binding
• Story: QoderAI/better-harness#34
• Status: Implemented

Intent

harness evidence-bundle freezes one shared Session population and then requires the Session facts lane to report exactly the frozen eligible count. The population is discovered through the sources command without the frozen topology/analysisScope, so it admits Sessions through the lenient legacy workspace match, while the facts lane re-qualifies the same frozen inventory with the strict topology-scoped rules (classifyWorkspaceCwd plus qualifyWorkspaceSessionInventory). Any Session admitted leniently but rejected strictly (for example a transcript with conflicting workspace CWDs) makes the two counts disagree deterministically, and the bundle fails closed with SESSION_POPULATION_BINDING_MISMATCH even though every Session file is valid. Both discovery and hydration must apply one identical qualification policy so the frozen binding and the facts lane always agree.

Acceptance Scenarios

• AC-1: collectSessionPopulation passes the frozen bundle topology and analysisScope to its sources discovery, so population discovery and the facts lane qualify Sessions under the same workspace-match policy.
• AC-2: A Claude evidence-bundle Session lane over a fixture home containing one clean workspace Session and one conflicting-CWD Session reports eligibleSessions and selectedSessions equal to population.binding.eligible.count instead of throwing SESSION_POPULATION_BINDING_MISMATCH; the conflicting Session is omitted from both paths.
• AC-3: The binding check itself remains fail-closed and unchanged; existing population, selection, and admission bindings keep their public shapes.

Non-goals

• Relaxing or removing the SESSION_POPULATION_BINDING_MISMATCH fail-closed contract.
• Changing provider discovery formats, workspace qualification rules, fingerprints, or the active-Session omission policy.
• Changing lead selection, scoring, report schemas, or renderer behavior.

Plan and Tasks

1. Add a focused red regression test in test/better-harness-evidence-bundle.test.mjs that drives collectSessionPopulation and collectSessionEvidence against a real Claude fixture home with one clean and one conflicting-CWD Session under a frozen repo-root topology.
2. Pass topology and analysisScope from the frozen context into the sources discovery inside collectSessionPopulation, mirroring the facts call in collectSessionEvidence.
3. Run the focused evidence-bundle and Session population suites, then the full repository gate.

Test and Review Evidence

• Red run: the new regression test fails on the unfixed code with SESSION_POPULATION_BINDING_MISMATCH (population 2 vs facts eligible 1).
• AC-1/AC-2: node --test test/better-harness-evidence-bundle.test.mjs
• AC-3: node --test test/session-population.test.mjs test/session-workspace-provider.test.mjs
• Regression gate: npm test