liability-meter

name	liability-meter
description	The firm's accumulated-exposure tracker: feed every consequential decision to the vr-risk MCP so individually-immaterial calls can't silently aggregate into a restatement or a penalty stack. log_decision(client, tier, area, dollars, penalty_exposure) for each GREEN/YELLOW/RED action; exposure_summary(client, period) to total the running risk a client carries; threshold_breached(client, limit) to surface ACCUMULATED exposure to the partner BEFORE it compounds. Load this whenever you classify a gate or take a consequential action — 200 below-materiality decisions can sum to a material misstatement, and one client's drip of small penalties can become a stacked assessment. This is the meter that makes aggregate risk visible while each decision still looks small.
version	1.0.0
author	RealDeal CPA
license	MIT
platforms	["linux","macos","windows"]
metadata	{"hermes":{"tags":["Liability","Risk","Exposure","Accumulation","Materiality","Penalty","Threshold","Quality-Control"],"related_skills":["authority-and-escalation","self-review-qc","workpaper-standard","provenance-and-evidence"]}}

Liability Meter

Every decision looks small in the moment. The danger is the sum. Two hundred individually-immaterial calls — each a defensible rounding, a reasonable estimate, a minor reclass — can aggregate into a material misstatement that forces a restatement. A client's drip of small late-deposit and information-return slips can stack into a penalty assessment no single item would have triggered. The liability meter exists so that aggregate exposure is visible to the partner before it compounds, while each decision still passes its own gate. Use the vr-risk MCP.

The three tools, and nothing you have to fabricate:

• log_decision(client, tier, area, dollars, penalty_exposure) — record one consequential decision: its gate tier (GREEN/YELLOW/RED), the area it touches, the dollar magnitude in play, and the penalty exposure it creates.
• exposure_summary(client, period) — total the running exposure a client carries across all logged decisions for the period.
• threshold_breached(client, limit) — test the accumulated exposure against a limit and surface a breach to the partner.

When to use

• You classify a gate (authority-and-escalation) or take any consequential action → log_decision it, every time, even GREEN.
• You finish a unit of work, run a standup sweep, or close a period → exposure_summary(client, period) to see the running total.
• Before a handoff, a filing, a sign-off, or on cadence → threshold_breached(client, limit) to catch an aggregate that crossed the line.

log_decision — one row per consequential decision

A consequential decision is anything that touches a number, an estimate, a classification, a filing, or a client's money — not just the RED ones. The whole point is to capture the small calls, because those are the ones that hide.

• client — KarbonCopy legal name + id (from list_clients), so exposure rolls up per client.
• tier — the gate you assigned this decision: GREEN, YELLOW, or RED. The tier mix is itself a signal; a wall of GREENs that sum past materiality is exactly what the meter is for.
• area — cash, ar, ap, payroll, tax, sales-tax, revenue, close, equity, recon — so exposure concentrates visibly (e.g. all of it sitting in payroll).
• dollars — the magnitude actually in play for this decision (the reclass amount, the estimate's range, the transaction size). Use Decimal precision; do not round to "feel" immaterial.
• penalty_exposure — the downstream dollar risk this decision creates if it's wrong: the IRC penalty that could attach (late deposit, §6721/§6722 information-return, accuracy-related §6662), the interest, the restatement cost. Distinct from dollars — a $400 misclassified deposit can carry a $10,000 penalty exposure. Log at the moment you decide, not in a batch. A decision you didn't log is exposure the partner can't see.

exposure_summary — the running total that no single row shows

exposure_summary(client, period) is the answer to "how much risk is this client actually carrying right now?" — the question no individual gate decision can answer.

• Run it at the end of any work session, on the daily standup, and at every close.
• Read it two ways: total dollars (could the sum be a material misstatement?) and total penalty_exposure (is a penalty stack forming?).
• Watch concentration: $9k spread across ten areas is different from $9k all in payroll one week before a 941. Concentration in one area or one tier is a louder signal than the headline number.
• The summary is an input to materiality judgment, not a substitute for it — materiality is still a CPA call (self-review-qc), but now it's made against the aggregate, not item-by-item.

threshold_breached — surface it BEFORE it compounds

threshold_breached(client, limit) tests the accumulated exposure against a limit and is the trigger that escalates. This is the heart of the skill: the partner hears about the aggregate before it forces a restatement or a stacked penalty, not after.

• Set the limit from the client's materiality (the planning/performance materiality on file, or a firm default if none) — for penalty exposure, a hard dollar limit the partner is willing to carry.
• A breach is not a GREEN you can clear yourself. A crossed aggregate threshold is a RED escalation to the human partner regardless of how green each underlying decision was — that is the inversion this meter introduces and it is invariant.
• Run it before any handoff, filing, or sign-off, and on cadence (standup, close). A breach mid-period is more useful than a breach discovered at year-end.
• When it breaches, name the top contributors (which areas/decisions drive the total) so the partner can act on the few that matter, not re-review everything.

How it sits under the gates (the inversion)

authority-and-escalation classifies each decision in isolation. The liability meter classifies the portfolio. A decision can be a legitimate GREEN on its own and still be part of an aggregate that has gone RED. When threshold_breached returns true, the aggregate RED overrides the per-item GREENs: you stop, summarize, and escalate. You never let "but each one was immaterial" justify shipping past a breached threshold — that reasoning is precisely the failure mode the meter catches.

Procedure

1. Decide → log. As you classify any gate or take any consequential action, log_decision(client, tier, area, dollars, penalty_exposure) — Decimal magnitudes, real penalty exposure, never batched.
2. Total → summarize. At session end, standup, and close, exposure_summary(client, period); read dollars, penalty_exposure, and concentration.
3. Test → escalate. threshold_breached(client, limit) before handoffs/filings/sign-offs and on cadence; a breach is a RED to the partner.
4. Name contributors. On any breach, surface the top areas/decisions driving the total so the partner targets the few that matter.
5. Cross-reference. Tie the logged decisions back to their workpaper indexes (workpaper-standard) and provenance events (provenance-and-evidence) so the exposure is supportable, not a bare number.
6. Exceptions. A breach, or a concentration that worries you below the limit, goes to the EXCEPTIONS QUEUE for the partner; never self-clear a breach.

Edge cases / pitfalls a 15-year CPA knows cold

• Aggregation is the whole point — logging only RED decisions defeats it; the misstatement hides in the GREENs you didn't bother to log.
• dollars and penalty_exposure are different axes — a tiny dollar item can carry an enormous penalty (a single late payroll deposit, a missed 1099 batch). Log both honestly; don't collapse them.
• A subtotal under the limit is not a pass if it's concentrated or trending — a $9k aggregate climbing $2k/week toward a $12k materiality is a YELLOW heads-up to the partner now, not a RED surprise next Friday.
• Penalty stacks compound non-linearly — §6721/§6722 can apply per return and per payee, late-deposit penalties tier by days late; a naive sum understates them. When in doubt, log the higher tier and let the partner judge.
• Don't net exposures to look smaller (an overstatement here "offsetting" an understatement there) — misstatements aggregate by absolute value for materiality; netting hides risk.
• Reset the meter by period, not by mood — exposure is per-client, per-period; rolling a prior period's cleared items forward inflates the total and erodes the signal.
• The meter measures exposure, not guilt — a high reading isn't a failure, it's information; the failure is letting it cross a threshold silently.
• Never tune a limit upward just to avoid a breach — that is moving the goalposts and it defeats the partner's protection; the limit comes from materiality, not convenience.

Required output

• Exposure note — for the client/period: total dollars, total penalty_exposure, the area/tier concentration, and whether threshold_breached returned true.
• Top contributors (on a breach or a near-breach) — the few decisions/areas driving the total, each cross-referenced to its workpaper index.
• EXCEPTIONS QUEUE — any breach or worrying concentration as a RED item for the partner; "none — exposure within limits" stated explicitly if clean.

Approval gate

GREEN (autonomous): log_decision, exposure_summary, and threshold_breached — metering exposure is always free to do, and you should do it relentlessly on every consequential decision. RED: a breached aggregate threshold is a stop-the-line escalation to the human partner even when every underlying decision was GREEN — the aggregate RED overrides the per-item gates. The underlying actions the meter measures (posting material JEs, moving money, filings, client comms) keep their own RED gates. Never raise a limit to dodge a breach, and never net exposures to look smaller — both defeat the partner's protection.