| name | verdict-report |
| plugin | agent-finance |
| description | Produces a decision-support verdict from the filled Excel model + full research cache. Reads out_* DCF outputs, computes a probability-weighted fair-value range (Bull/Base/Bear), derives a 5-level signal (ACCUMULATE / ADD / HOLD / REDUCE / AVOID), and emits a structured action plan across three time horizons. Use last in the pipeline, after excel-financial-model has been validated and opened in Excel (formulas must be calculated). Never labels output as investment advice.
|
| allowed-tools | Read, Write, Grep, Glob, Bash |
verdict-report
Purpose
Bridge the gap between the research note (report-composer) and a decision.
Produce a single verdict document that:
- Audits the Excel model — reads every computed output cell, flags anomalies
(TV% > 75%, checks != OK, WACC < g).
- Derives a fair-value range — re-runs the model in-memory for each scenario
to get Bull / Base / Bear value-per-share.
- Emits a probability-weighted signal — uses the working-view weights from
bull-bear-thesis.md to compute an expected value and compare to current price.
- Produces a structured action plan — three horizons: 0–3 months, 3–12
months, 1–3 years — with specific milestones, tripwires, and kill conditions
sourced from the research cache.
Hard rules
- Never label output as investment advice. The disclaimer line is fixed and
verbatim: "This report is decision-support research only. It is not
investment advice."
- Every figure cites source. Format:
source: <filename or URL> as-of=<date>.
- Refuse delivery if model not recalculated. If
out_value_per_share reads
None (formulas not evaluated), stop and instruct the user to open the file
in Microsoft Excel once.
- Refuse delivery if bull-bear-thesis missing. The working-view weights are
required to compute the expected value.
- Never invent scenario values. If a scenario VPS cannot be computed (e.g.
bull/bear model inputs not available), mark it
null with a gap: reason.
- Reporting currency stated on every monetary figure.
Inputs
| Input | Required | Source |
|---|
ticker | ✅ | CLI --ticker |
| Filled Excel model (recalculated) | ✅ | ${PROJECT_DIR}/output/agent-finance/models/<TICKER>_<DATE>.xlsx |
bull-bear-thesis.md | ✅ | ${PLUGIN_DATA}/companies/<TICKER>/bull-bear-thesis.md |
valuation-multiples.json | ✅ | ${PLUGIN_DATA}/companies/<TICKER>/valuation-multiples.json |
model_inputs/values.json | ✅ | Needed to re-run bull/bear scenarios |
model_inputs/sources.json | ✅ | Source annotations for re-run |
model_inputs/peer_comps.json | optional | Needed for comps-implied VPS |
risk-assessment.json | optional | Enriches tripwires section |
earnings-analysis.json | optional | Next earnings date / catalyst calendar |
Methodology
Step 1 — Read Excel outputs
Open the model with openpyxl data_only=True. Via cell_map.json read:
| Logical name | Meaning |
|---|
out_value_per_share | DCF Base value per share |
out_upside | (VPS / current price) − 1 |
out_tv_pct_ev | Terminal value as % of EV — flag if > 0.75 |
out_wacc | WACC |
out_enterprise_value | Enterprise value |
out_equity_value | Equity value |
out_comps_avg_vps | Comps-implied average value per share |
If any of the above is None, abort with:
"Excel formulas not evaluated. Open the model in Microsoft Excel once to
recalculate, then re-run verdict-report."
Step 2 — Derive Bull / Bear VPS
Re-invoke fill_model.py twice in-memory (no file save):
- Bull: override
in_scenario = 1, re-read out_value_per_share
- Bear: override
in_scenario = 3, re-read out_value_per_share
Fair-value range = [vps_bear, vps_base, vps_bull]
Step 3 — Parse working-view weights
Regex-parse bull-bear-thesis.md for the working-view line:
bull XX% / bear YY%
Extract bull_weight and bear_weight. Base weight = 1 - bull_weight - bear_weight.
Step 4 — Expected value & signal
ev = bull_weight × vps_bull + base_weight × vps_base + bear_weight × vps_bear
Signal thresholds (vs current price p):
| Condition | Signal |
|---|
ev > p × 1.20 | ACCUMULATE |
ev > p × 1.05 | ADD |
p × 0.95 ≤ ev ≤ p × 1.05 | HOLD |
ev < p × 0.95 | REDUCE |
ev < p × 0.80 | AVOID |
Step 5 — Catalyst calendar
Parse milestones from bull-bear-thesis.md (lines starting with "Milestone:").
Tag each as Q1/Q2/Q3/Q4 or year-based. Sort by proximity to today.
Step 6 — Action plan (3 horizons)
Combine milestones, tripwires from risk-assessment.json, and guidance check
points from earnings-analysis.json into three horizon buckets.
Step 7 — Write outputs
${PLUGIN_DATA}/companies/<TICKER>/model_inputs/verdict.json — machine-readable${PROJECT_DIR}/output/agent-finance/reports/<TICKER>_verdict_<DATE>.md — human report
Output schema
verdict.json
{
"ticker": "DOL",
"as_of": "2026-05-28",
"reporting_currency": "CAD",
"current_price": 174.95,
"dcf_vps": {
"bull": null,
"base": 60.94,
"bear": null
},
"comps_implied_vps": 26.78,
"bull_weight": 0.60,
"base_weight": 0.0,
"bear_weight": 0.40,
"expected_value_weighted": 0.0,
"signal": "HOLD",
"signal_conviction_pct": 60,
"upside_base_pct": -65.1,
"tv_pct_ev": 0.889,
"tv_flag": true,
"wacc": 0.0581,
"gaps": [],
"sources": [
"cell_map.json",
"bull-bear-thesis.md",
"valuation-multiples.json"
]
}
verdict.md — fixed section order
# <TICKER> — <Company> — Decision-Support Verdict
> This report is decision-support research only. It is not investment advice.
> Reporting currency: <CCY>. As-of: <DATE>.
## 1. Model Audit
### 1.1 Validity
### 1.2 Computed outputs
### 1.3 Flags & anomalies
## 2. Fundamental Verdict
### 2.1 Quality of business
### 2.2 Growth trajectory
### 2.3 Financial health
### 2.4 Valuation (DCF range + comps)
### 2.5 SIGNAL
**[ACCUMULATE | ADD | HOLD | REDUCE | AVOID]**
Conviction: XX% | Assumption: [working-view label]
## 3. Action Plan
### 3.1 Immediate (0–3 months)
### 3.2 Medium-term (3–12 months)
### 3.3 Long-term (1–3 years)
## 4. Key Assumptions to Monitor (Tripwires)
[If X changes → signal flips to Y]
## Sources
Invocation
python "$CLAUDE_PLUGIN_ROOT/skills/verdict-report/build_verdict.py" \
--ticker DOL \
--model "$CLAUDE_PROJECT_DIR/output/agent-finance/models/DOL_2026-05-28.xlsx" \
--cache "$CLAUDE_PLUGIN_DATA/companies/DOL" \
--output "$CLAUDE_PROJECT_DIR/output/agent-finance/reports/DOL_verdict_2026-05-28.md"
Or via the orchestrator when the user asks for a "verdict", "moves",
"recommendation", "buy or sell", or "que faire sur".
Pipeline position
company-profile → ... → bull-bear-thesis → model-input-builder →
excel-financial-model → report-composer → [verdict-report]
verdict-report is optional and on-demand. The pipeline delivers without
it; it is triggered explicitly by user intent.
Dependencies
excel-financial-model must have run and the xlsx must have been opened in
Excel once (formulas evaluated).bull-bear-thesis must have produced bull-bear-thesis.md with a working-view
section containing explicit bull/bear weights.fill_model.py is imported directly for in-memory scenario re-runs.
