docs(roadmap): make ROADMAP forward-looking; drop stale present-state sections

[igerber]

Summary

• Make ROADMAP.md forward-looking only: delete the drift-prone "Current State" and "Recently Shipped" sections; the intro now points to the README estimator catalog / RTD / CHANGELOG, which maintain "what exists today" authoritatively. (The roadmap had frozen at ~v3.3.0 while the library shipped v3.5.3 — ~952 commits of drift.)
• Prune forward sections of work that has shipped: HeterogeneousAdoptionDiD (now a shipped estimator), the triple-difference covariate-identification audit (Complete on the methodology tracker for both TripleDifference and StaggeredTripleDifference), and Butts spillovers (shipped as SpilloverDiD).
• Slim the AI-Agent Track "Building blocks already in place" enumeration to a one-line pointer.
• Fix the stale SpilloverDiD survey_design= flag in llms-full.txt (Wave E lifted the rejection — HC1/CR1/Conley analytical-TSL supported); leave the still-accurate covariates= NotImplementedError flag intact.
• Update docs/business-strategy.md's structural description (internal doc) to match the forward-looking layout.

Methodology references (required if estimator / math changes)

• N/A — no estimator/math changes. Docs-only. The only file under diff_diff/ is the bundled agent-guide text diff_diff/guides/llms-full.txt, edited solely to document an already-shipped SpilloverDiD survey capability (Wave E); no code, no methodology, no behavior change.

Validation

• Tests added/updated: none (docs-only).
• Backtest / simulation / notebook evidence: N/A (no notebooks changed).
• Verified post-edit: no dangling intra-doc references, no inbound links to the deleted ROADMAP anchors, and no shipped estimator left in a "future" section.

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

✅ Overall Assessment: Looks good

Executive Summary

• No P0/P1 methodology or correctness issues found.
• The SpilloverDiD survey_design= guide update matches the Methodology Registry and implementation gates.
• Replicate-weight and no-effective-PSU lagged-Conley limitations are documented/tracked in TODO.md, so they are non-blocking.
• No code behavior, estimator math, defaults, or inference computation changed.
• One minor documentation drift remains in docs/business-strategy.md.

Methodology

Finding: P3 informational — SpilloverDiD survey support is correctly documented

• Location: diff_diff/guides/llms-full.txt:L508-L515
• Impact: The updated guide now aligns with the registry: HC1/CR1 survey TSL and Conley survey paths are shipped; replicate-weight variance remains unsupported.
• Concrete fix: No action required. This matches docs/methodology/REGISTRY.md:L3700-L3706, L3708-L3755, and L3801-L3802.

Finding: P3 informational — documented/tracked limitations remain non-blocking

• Location: diff_diff/guides/llms-full.txt:L515; TODO.md:L98-L99
• Impact: Replicate-weight SpilloverDiD(survey_design=...) and lagged Conley with no effective PSU are explicitly tracked deferred work, not silent correctness bugs.
• Concrete fix: No action required.

Code Quality

Finding: No issues

• Severity: N/A
• Impact: PR is documentation-only; no code paths, APIs, or inference helpers changed.
• Concrete fix: None.

Performance

Finding: No issues

• Severity: N/A
• Impact: No runtime code changed.
• Concrete fix: None.

Maintainability

Finding: P3 — minor cross-doc wording drift

• Location: docs/business-strategy.md:L335-L337
• Impact: The section says ROADMAP “Shipping Next” includes BusinessReport / DiagnosticReport, while the edited ROADMAP.md:L11-L13 now frames only context-aware practitioner_next_steps() as next work on top of the shipped BR/DR layer.
• Concrete fix: Reword to: Context-aware practitioner_next_steps() follow-up to the shipped BusinessReport / DiagnosticReport layer.

Tech Debt

Finding: No blocking issues

• Severity: N/A
• Impact: Deferred SpilloverDiD survey limitations are already tracked in TODO.md:L98-L99.
• Concrete fix: None for this PR.

Security

Finding: No issues

• Severity: N/A
• Impact: No secrets, credentials, network behavior, or executable code changes observed.
• Concrete fix: None.

Documentation/Tests

Finding: P3 — no tests added, acceptable for docs-only PR

• Location: changed files only
• Impact: No estimator behavior changed. I checked doc consistency around the changed SpilloverDiD claim and git diff --check passed.
• Concrete fix: Optional: update the minor docs/business-strategy.md:L335-L337 wording noted above.

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: d4456d13ba797021fe7dea5e230633d88fc25e43

---

✅ Overall Assessment: Looks good

Executive Summary

• No P0/P1 methodology, correctness, or inference issues found.
• No estimator code, defaults, math, weighting, variance, or SE paths changed.
• The SpilloverDiD survey_design= guide update matches the Methodology Registry and implementation gates.
• Prior P3 wording drift in docs/business-strategy.md section 7 is resolved.
• One residual P3 planning-doc drift remains in section 8, but it is informational and not approval-blocking.

Methodology

Finding: P3 informational - SpilloverDiD survey support is documented consistently
Impact: The guide now correctly says survey_design= is supported for analytical/TSL HC1, CR1, and Conley paths, while replicate-weight variance remains unsupported. This matches the registry and implementation fail-closed gates.
Locations: diff_diff/guides/llms-full.txt:L508-L515, docs/methodology/REGISTRY.md:L3687-L3706, docs/methodology/REGISTRY.md:L3708-L3757, docs/methodology/REGISTRY.md:L3801-L3802, diff_diff/spillover.py:L2363-L2407, diff_diff/spillover.py:L3219-L3261, TODO.md:L98-L99
Concrete fix: None required.

Code Quality

Finding: No issues
Severity: N/A
Impact: No runtime code changed.
Concrete fix: None.

Performance

Finding: No issues
Severity: N/A
Impact: No executable paths changed.
Concrete fix: None.

Maintainability

Finding: P3 informational - residual internal planning-doc drift
Impact: Section 7 now matches the forward-looking ROADMAP framing, but section 8 still lists BusinessReport and DiagnosticReport as queued “Shipping Next” items even though they are shipped and the ROADMAP now frames only context-aware practitioner_next_steps() as the follow-up. This is internal-doc drift, not a methodology or correctness issue.
Locations: docs/business-strategy.md:L362-L374, ROADMAP.md:L11-L13, README.md:L82, diff_diff/business_report.py:L89, diff_diff/diagnostic_report.py:L259
Concrete fix: Optionally update docs/business-strategy.md:L366-L374 to remove BusinessReport as queued work and rephrase the first queued item as the context-aware practitioner_next_steps() follow-up.

Tech Debt

Finding: No blocking issues
Severity: N/A
Impact: Deferred SpilloverDiD replicate-weight variance and no-effective-PSU lagged Conley limitations are tracked in TODO.md.
Concrete fix: None.

Security

Finding: No issues
Severity: N/A
Impact: No secrets, credentials, network behavior, or executable changes observed.
Concrete fix: None.

Documentation/Tests

Finding: P3 informational - no tests added, acceptable for docs-only PR
Impact: The PR changes documentation and bundled guide text only. git diff --check passed.
Concrete fix: None required for approval.