executive-comp-peer-builder

Example: Executive Compensation Peer Group Builder

The first Executive Compensation arm agent: a dark, board-ready dashboard that builds a defensible executive-comp peer group the way a Compensation Committee actually does it — a hard screen for membership, then a transparent fit-rank for ordering — and stops at a human approval gate.

It reads the shared screener over a universe of real public companies (as-disclosed public financials — a dated, illustrative snapshot; the subject is the same synthetic Acme Corp the rest of the portfolio uses), proposes a recommended core peer group plus a substitution watchlist, documents every exclusion, and carries the committee's target-percentile policy forward to the benchmarking arm. It never recommends pay.

Synthetic subject, real peers. Acme (the subject) is synthetic; the candidate peers are real public companies with as-disclosed public financials — an illustrative, dated snapshot (provenance + per-figure sources: governance/real-peer-data.md), not investment/comp advice; verify against each company's filings for current actuals. No real name carries a fabricated pay/TSR figure — the ISS pay screen and rTSR valuation run on a separate synthetic universe.

The two-step model (the way a committee works)

1. Screen — the gate. Membership is decided by a hard, transparent per-criterion pass/fail: revenue and market cap each within 0.5–2.0× of the subject, and membership in the documented software/SaaS peer group (a set of GICS sub-industries — GICS fragments SaaS across sectors). Every inclusion and exclusion defends itself on one line. Headcount is a soft factor (not a gate) — matching disclosed market practice, where revenue and market cap are the primary size anchors.
2. Fit-rank — the order. Within that in-band group, peers are ranked by a pure revenue-weighted size-closeness score over revenue, market cap, and headcount (100 = identical size; 0 = at a band edge) into a recommended core and a watchlist. The score orders the group; it never changes who is in it.

This separation is the point: a board can challenge a score ("who set those weights?"), but not a transparent screen ("in the software/SaaS peer group, within 0.5–2.0× our size").

Run it

cd examples/executive-comp-peer-builder
python3 run.py                                              # draft only
python3 run.py --publish                                    # refused: needs a named committee approver
python3 run.py --publish --approved-by "Compensation Committee Chair"

Test it

python3 evals/test_peer_builder.py

Sample output

• Committee dashboard (HTML)
• Committee digest

What it demonstrates

• Defensible by construction: membership is a transparent in/out screen; the fit score only orders the group it never gates. Both survive a board's cross-examination.
• Committee record: same-industry companies kept out on size are listed with the exact criterion they failed — no cherry-picking.
• One consistent company: the subject is the same synthetic Acme used across the whole portfolio (no second, contradictory profile).
• Presentation + governance only: every PASS/FAIL and fit score comes from the shared screener; the agent does no math, fails closed, and stops at a human approval gate.
• Bridges to benchmarking: carries the committee's target-percentile policy forward — applied only after the peer group is approved. No pay recommendation is made here.