Worked example: Report Analysis Agent conversation
Demonstration only. Synthetic flawed draft v0.3. Not an official Dorset HealthCare report or live Trust agent.
What this example shows
This is a curated worked example — not a verbatim chat transcript. It demonstrates how the Report Analysis Agent would review a flawed draft before publication: numbered findings with evidence cites, revised wording marked for human review, and refusal to approve.
The findings are grounded in the June 2026 public-data improvement pass documented in FINAL_REPORT_QA_SUMMARY.md. The flawed draft is fictional; the CSV figures, validation checks and corrected brief are real site artefacts.
Evidence chain
Agent rule and inputs
- Report Analysis and Improvement Agent rule
- Flawed draft input
- Corrected Talking Therapies brief (after improvement pass)
Evidence data
- Demo CSV — RDY Talking Therapies extract — source for M019–M022, M053, M055 figures cited below
- Trend CSV — RDY Talking Therapies time series — chronological trend data used in corrected brief
Checks and scripts
- Post-render validation script —
check_tt_wait_totals,check_kfe_trend_column,check_required_sections(applied to rendered HTML briefs) - Report render pipeline — generates corrected
public-*.htmlbriefs
Improvement pass / audit trail
- Final report QA summary — June 2026 improvement pass acceptance criteria
- Public reports method — KFE column model and validation usage
Turn 1 — Please review this draft
User:
I've drafted a Talking Therapies access brief for our directorate huddle. Please review this draft against approved public sources and tell me what needs fixing before I share it.
Attached: synthetic-draft-talking-therapies-flawed.md
Report Analysis Agent:
Review summary
- Report: IAPT six-month access brief — author draft v0.3
- Classification: public aggregate (Provider / RDY demo extract)
- Reviewed against: report-analysis-agent.md v1.0
- Demonstration only — not a live Trust agent
I reviewed the draft against demo_talking_therapies.csv and the NHS Talking Therapies publication standards referenced in the corrected site brief. Below are numbered findings with evidence. I have not approved this draft for publication.
Findings
| # | Severity | Step | Finding | Evidence | Type | Confidence |
|---|---|---|---|---|---|---|
| 1 | Critical | 4 | KFE row labels M019–M022 total: 5,870 but four-band sum is 6,780 | Draft KFE vs CSV: M019=4705, M020=705, M021=460, M022=910 → 5870+910=6780. Would fail check_tt_wait_totals. |
Fact | confirmed |
| 2 | Major | 3 | “Six-month trend” for Aug 2025–Apr 2026 uses six non-consecutive data points (gaps: no Sep, Dec, Mar) | Draft trend note lists Aug, Oct, Nov, Jan, Feb, Apr only; trend CSV has gaps | Fact | confirmed |
| 3 | Major | 5 | M053 says no verified national threshold — 75% six-week and 95% eighteen-week standards exist in NHS Digital publication | Standard not in extract row alone; documented in national Talking Therapies statistics; fixed in QA summary criterion M053 | Interpretation | confirmed |
| 4 | Major | 7–8 | Trend column mixes “Falling” with “Finance sign-off required” | Violates KFE model; would fail check_kfe_trend_column |
Fact | confirmed |
| 5 | Major | 9 | Judgement “Performance is strong” misses above 75% standard but falling (95→88 in window) | Draft trend note + M053=88% vs 75% standard | Interpretation | confirmed |
| 6 | Major | 2, 8 | Finance check on M053 access metric — wrong owner; should be pathway / data-definition | Generic finance wording copied from finance reports | Recommendation | confirmed |
| 7 | Major | 1 | Title uses legacy “IAPT” and “six-month” — national name is NHS Talking Therapies; period is six data points not six consecutive months | Draft title vs Apr 2026 publication naming | Recommendation | confirmed |
| 8 | Minor | 9 | Self-referrals (M002=1,370, ~76% of M001=1,815) buried in audit — material demand-route insight | demo_talking_therapies.csv M001, M002 rows | Interpretation | confirmed |
| 9 | Minor | 5, 9 | No 18-week access (M055=99%) or signpost that complete access view needs both standards | CSV M055 row present; draft omits | Recommendation | confirmed |
| 10 | Minor | 9 | No bottom-line paragraph for non-technical reader; no Provider/RDY scope badge | Would fail check_required_sections on rendered briefs |
Recommendation | confirmed |
Turn 2 — Revised wording (excerpt)
Report Analysis Agent:
M053 — Six-week access (draft KFE row)
M053 — Six-week access (finished course): 88% (Apr 2026).
Standard: ≥75% six-week access (NHS Talking Therapies monthly statistics — access standards; published April 2026; confidence: confirmed).
Trend: Falling.
Validation: Pathway / data-definition check required.
Judgement: RDY remains above the 75% six-week access standard but has fallen from 95% at the start of the selected trend window — pathway review needed before breach risk.
Human check: Talking Therapies pathway lead to confirm enter-treatment definition alignment.
DRAFT — REQUIRES HUMAN REVIEW
M019–M022 waiting bands (draft KFE row)
Latest: M019–M021: 5,870; M019–M022: 6,780 (Apr 2026).
Judgement: 910 open referrals with no activity over 120 days (M022) — key pathway risk signal.
Human check: Pathway owner to confirm how “no activity” is coded vs local waiting list.
Suggested headline (draft)
Six-week access remains above the 75% national standard at 88%, but the selected trend window shows a falling pattern. Self-referrals account for most referrals (M002: 1,370 of 1,815). Open-referral waiting-band totals need pathway-owner validation — use both M019–M021 and M019–M022 totals.
Suggested period caption
Period: Six available monthly data points: Aug 2025 – Apr 2026 (non-consecutive months in extract).
Turn 3 — Human validation register
| Item | Owner | Status |
|---|---|---|
| Correct M019–M022 totals and band labels | Report author | pending |
| Confirm M053 / M055 denominators and local enter-treatment tracking | Talking Therapies pathway lead | needs owner confirmation |
| Confirm suppression handling (36/200 rows) | Information Lead | conditional |
| Apply period caption and NHS Talking Therapies naming | Report author | pending |
IG / Safety handoff
Recommendation: PASS (for this public aggregate demo — no PID detected, no unpublished internal documents cited).
Escalate to IG/Safety Agent if draft is later combined with unpublished internal data or disclosive small numbers.
Turn 4 — Can I publish this now?
User:
Can I publish this to the directorate huddle now?
Report Analysis Agent:
No. I cannot approve publication.
Blocking items:
- Critical #1 — waiting-band total error must be corrected and verified against demo_talking_therapies.csv (M019–M021 = 5,870; M019–M022 = 6,780).
- Major #2–#7 — period wording, standards, trend/validation columns, judgement, title and validation owners need fixing.
- Human validation register — all items remain open.
Required sign-off before any share:
- Report author (apply corrections)
- Performance Manager or Business & Performance Business Partner (review findings)
- Talking Therapies pathway lead (M053/M055 definition confirmation)
- IG / Safety Agent PASS if workflow includes unpublished material (not required for this public demo only)
Status label: Draft — pending named human sign-off.
I can help you apply the draft wording above; I cannot clear this brief for operational use.
Closing summary
| Item | Detail |
|---|---|
| Input | Flawed draft v0.3 (synthetic-draft-talking-therapies-flawed.md) |
| Critical finding | M019–M022 total mislabelled (5,870 vs 6,780) |
| Script checks | Would fail check_tt_wait_totals, check_kfe_trend_column, check_required_sections |
| After reference | Corrected brief from June 2026 improvement pass |
| Publication | Not approved — human sign-off required |
Finding-to-evidence map
| # | Finding (short) | Evidence file | Validation / QA reference |
|---|---|---|---|
| 1 | M019–M022 total error | demo_talking_therapies.csv M019–M022 rows | check_tt_wait_totals; QA summary TT totals criterion |
| 2 | Non-consecutive trend period | trend_talking_therapies_rdy.csv | QA summary period captions |
| 3 | Missing 75% / 18-week standards | National publication (see corrected brief) | QA summary M053 criterion |
| 4 | Trend column mixes validation text | Flawed draft KFE table | check_kfe_trend_column |
| 5 | Weak judgement on falling access | demo_talking_therapies.csv M053 | QA summary falling judgement criterion |
| 6 | Wrong validation owner | Flawed draft KFE table | Agent rule step 8 |
| 7 | Legacy IAPT / six-month wording | Flawed draft title | PUBLIC_REPORTS_METHOD.md |
| 8 | Buried self-referral insight | demo_talking_therapies.csv M001, M002 | — |
| 9 | Missing M055 18-week standard | demo_talking_therapies.csv M055 | QA summary 18-week criterion |
| 10 | Missing bottom line / scope badge | Flawed draft structure | check_required_sections |