An investigative roadmap for the question “who gained and who lost from this administration’s policy, and through what mechanism?”

The instrument is a Policy Winner–Loser Map (WLM). It is a map, not a score:

• Stage 1 builds the structure of a market — the value-chain taxonomy — before any company is named, because a structure built while looking at a logo set bends to fit the logos;
• Stage 2 traces enacted actions through named transmission channels onto that structure, with a stated baseline, an observable, and a refuting condition on every claim;
• both sides are mandatory — every action has beneficiaries and losers, and a one-sided map is incomplete and usually motivated.

Unit of analysis: a market, across many actions. This is what separates it from legislative-impact-scoring (one action, two tracks, scored) and from the person-scoring instruments — malice-evaluator (a person’s actions), spokesperson-misinformation-assessment (statement veracity), dangerous-rhetoric-assessment (rhetoric).

Method is administration-agnostic; the tracked instance is the sitting administration. An instrument that only works on one administration is a partisan artifact and dies with it. This one must produce an equally rigorous map of a policy our readers like.

Canonical rules live in docs/policy-winners-losers-methodology.md — this skill operationalizes them. If the skill and the charter disagree, the charter governs. The Stage 1 method is docs/value-chain-mapping-method.md.

When to Use This Skill

Run it when the question is about a market and who policy moved within it:

• “Who benefits from the tariffs on X?”
• “Which industries gained from this deregulation, and who was displaced?”
• “Who are the winners and losers of the administration’s energy policy?”
• Building the structural backbone for an investigative series on an industry
• Adding a market-structure section to a policy explainer or accountability profile

Do NOT run it for:

The Two Gates (run before anything else)

Both fail closed. They are the analogues of the DMA’s ideology-vs-malice gate and the SMA’s advocacy-vs-misinformation gate.

Gate 1 — Benefit is not corruption

A firm or sector that gains from a policy has, on that fact alone, done nothing wrong. Governments act on markets constantly and someone always gains; that is what policy is. Treating “benefited” as a synonym for “captured” or “was rewarded” turns this into an insinuation engine.

Escalate — do not assert — when an independent indicator is documented: recusal failure, undisclosed interest, non-competitive award, contemporaneous private benefit to a decision-maker, or the beneficiary’s documented authorship of the instrument. At that point the finding leaves this skill and routes to the corruption trackers. The WLM records the handoff and says nothing further.

Absent an indicator: the map says the firm benefited, and stops. No adjectival load, no “notably,” and no placing a beneficiary next to an unrelated corruption finding to let the reader connect them. Insinuation-by-arrangement is this instrument’s characteristic failure — no mechanical check catches it, which is why the Claims Ombudsman reviews for it specifically.

Gate 2 — No investment advice, ever

Refused categorically: position sizing, buy/sell/hold framing, ranking firms by attractiveness, price targets, tickers presented as candidates, and the hype register — “positioned for,” “set to benefit,” “clear winner.”

A WLM describes where policy lands in a market’s structure. It is never shaped as a recommendation.

The Prior Question — always first

What did this market look like before the action, and what would it look like absent it?

This is the step that changes most conclusions, and it is the one most often skipped. No winner or loser claim is made without a stated baseline (charter §2). No baseline → no claim; log the row at confidence: low with the gap named.

Post-hoc attribution is the single most likely defect this instrument produces. A sector that grew after a policy did not necessarily grow because of it; a firm that won contracts under this administration may have won them under any administration.

Acceptable baselines, in descending preference:

1. Prior-law / prior-posture trajectory — the published trend before the instrument took effect
2. A comparable untreated market or jurisdiction — comparability argued, not asserted
3. A stated counterfactual — explicitly labeled an assumption, never an observation

Publish the baseline with the finding. A reader cannot evaluate an attribution claim whose counterfactual is hidden.

Stage 1 — Build the structure (no companies)

Full method: docs/value-chain-mapping-method.md. In brief:

1. Intake — reader, civic decision, upstream/downstream boundary, jurisdiction, horizon, explicit exclusions, and the policy domain the map is for.
2. Spine — pick one dominant axis (physical flow · technology stack · capital flow · transaction lifecycle · actor type). Hybrids are normal; make the transition point explicit.
3. Pillars — walk all 14 archetypes; 8–15 pillars. Orthogonal pillars (services & talent, capital, governance, demand, lifecycle) go at the end, not in flow order.
4. Node grammar — noun phrases; the occupancy test (3+ organizations could occupy it, fewer than ~100); sibling coherence; canonical home plus a note for genuine overlaps.
5. Depth — three levels. Level 4 is companies (Stage 2). If you are at level 4 in Stage 1, vendors crept in.

Allocate depth by constraint, not headcount. For a policy map this is the whole game: chokepoints are where a policy instrument has leverage over the entire chain. A segment with eight suppliers and veto power deserves more nodes than one with four hundred interchangeable vendors.

python scripts/build_value_chain.py docs/value-chains/<market>.txt --check

Stage 1b — The policy attribute overlay

All of these are attribute columns, never branches (one hierarchy, N attributes). A naive policy map wants a “policy” pillar containing everything policy touches; that produces two competing hierarchies and an unusable tree.

The chokepoint distribution is itself a finding. Where chokepoints cluster tells you where the chain is gated, and therefore where an instrument bites hardest — a more useful output than any single node.

Stage 2 — The event log (enacted actions only)

A proposal is not an action. A statement of intent is not an action. A press report of a forthcoming action is not an action. A post announcing an action is not the action.

Litigation status is mandatory. An enjoined action has no mechanical effect while enjoined; treating one as live is the most common error in this analysis.

Punditry, forecasts, and analyst commentary are non-information for the event log and may not be laundered by attribution.

Stage 3 — Trace the transmission mechanism

Name the specific channel. Vague causal claims are the characteristic failure of this genre.

Cash-flow channels

Civic channels — the reason this is a PU instrument and not a portfolio one

Motive is not a channel. Why a policy was enacted tells you nothing about how it reaches a market. Motive routes to malice-evaluator and the corruption trackers.

State the channel, the direction, and the magnitude with its basis. “Raises input costs” is not a finding. “Raises input costs by roughly N% of COGS, based on the tariff rate applied to the imported share disclosed in the filing” is. Magnitude asserted without a basis is a defect.

The Claim Contract

Every row carries all six fields. A row missing any of them is deleted, not softened.

Action:       <what, effective date, primary citation>
Status:       in force / enjoined / phased / vacated
Baseline:     <prior trajectory, comparable market, or labeled counterfactual>
Mechanism:    <channel> — <direction, magnitude, and the basis for the magnitude>
Exposure:     <structural nodes touched; directness tier>
Displaced:    <the counterpart position, where direction is advantaged>
Observable:   <the reported figure that would show this landing>
Refuted if:   <what would show it did not>
Confidence:   <plain words; never a computed-looking number that was not computed>

Both sides, always. A WLM of an action our readers dislike that finds only losers has failed its own method. The displaced column is the one most often omitted — incumbents whose economics the policy erodes, workers in a substituted process, regions bypassed by a reallocation. Omitting it is a tell.

Stage 4 — Population (only on an explicit editorial decision)

Publication default is structure-only (charter §4). A structure-only map can always be populated later; a published misattribution cannot be unpublished from a reader’s memory.

When a named firm is placed on a node:

• Primary sourcing — filings, procurement records, Federal Register text, court records, official trade data. Trade press is supporting, never sole.
• Pure-play vs. segment — “Company X benefits” is materially different when the affected segment is 3% of revenue. Omitting that is a misrepresentation.
• Disclosure over inference — sector-average exposure is not firm exposure.
• Private individuals excluded — route through patriot-private-citizen-inclusion-gate.
• Gate 1 at every named firm, without exception.

Sourcing method: corporate-intelligence-investigator.

Never combine, never sum

1. Never sum across instruments. A Financial Cost Tracker household figure (dollars per household per year) and a firm-side finding are different units with different payers. Same document, never the same number.
2. Never produce a composite winner/loser score. Scoring an action is LIA’s job — and LIA is itself forbidden from collapsing its two tracks into one. A single scalar “who won” is a defect.
3. Never rank firms. Ranking is the shape of investment advice (Gate 2) and the shape of an accusation (Gate 1), and it adds no civic information the mechanism does not already carry.

Output Format

POLICY WINNER–LOSER MAP — <market>

SCOPE
  Reader | Civic decision | Boundaries | Jurisdiction | Horizon | Exclusions | Policy domain

MARKET STRUCTURE (Stage 1)
  <pillar tree, levels 1–3>
  Chokepoints: <where the chain is gated — the leverage view>

EVENT LOG
  Action | Effective | Instrument | Status | Primary source

TRANSMISSION MAP
  Action → Channel → Baseline → Magnitude basis → Exposure → Observable → Refuted if

ADVANTAGED POSITIONS        (structural; named firms only per Stage 4)
DISADVANTAGED & DISPLACED   (mandatory — never omitted)

ESCALATED
  <Gate 1 indicators found> → corruption trackers. Nothing asserted here.

WHAT I COULD NOT DETERMINE
  <gaps, and why they matter>

AS OF
  Build date | Litigation-status check date

Compose rendering: Datawrapper for the pillar and chokepoint distributions, network-graph for a populated map, infographic for the reader-facing summary. A WLM is a dated artifact — see Staleness.

Composition with Other Skills

0. make route Q="<the market / instrument / sector>"     -> resolve current SKILL.md paths; never hardcode
1. policy-analyst-legislative-specialist                 -> event identity, status, compliance tracking
   federal-register-regulatory-analysis                  -> rule text and posture
2. legislative-impact-scoring (LIA)                      -> Q3 sectoral output CONSUMED as an event input
3. value-chain-mapping-method + build_value_chain.py     -> Stage 1 structure
4. source-credibility-scoring                            -> evidence tier on every pillar and event
5. corporate-intelligence-investigator                   -> Stage 4 population sourcing
   patriot-private-citizen-inclusion-gate                -> individuals
6. trump-corruption-accountability-tracker /
   trump-family-financial-tracker /
   public-corruption-ombudsman                           -> Gate 1 ESCALATION target
   malice-evaluator                                      -> all motive questions
7. citation-checker + Claims Ombudsman                   -> publish gate (runs last)

Reads: docs/policy-winners-losers-methodology.md (canonical), docs/value-chain-mapping-method.md (Stage 1 method). Hands off: scoring an action → LIA; dollars → Financial Cost Tracker; wrongdoing → corruption trackers; motive → DMA; portfolio questions → nobody at PU (refuse).

Anti-Patterns

What This Skill Is NOT

• NOT investment advice — categorically, under any framing (Gate 2).
• NOT a corruption finding. Benefiting is not wrongdoing (Gate 1).
• NOT a scorer. No composite, no ranking, no grade. It is a map.
• NOT a forecaster. Enacted actions only; no prediction of what will be enacted, no election forecasting.
• NOT a motive instrument. Mechanism is the finding; motive routes to DMA.
• NOT a person-scoring instrument (those are DMA/SMA/DRA), and not an action-scoring instrument (that is LIA).

Common Mistakes to Avoid

Staleness

A WLM is a dated artifact. An action may since have been amended, expired, enjoined, or vacated; a map that was correct in March can be actively misleading in September.

• Every published map carries its build date and its litigation-status check date.
• Re-verify litigation status quarterly, and before any republication or reuse.
• Re-run the constraint, failure-mode, and policy-instrument pressure tests quarterly. A bottleneck appearing in the trade press with no node is the leading indicator that the taxonomy needs revision.

Examples

Example A — a tariff on an intermediate input (structure-only, publishable)

Stage 1 map of the affected chain; the tariff logged with its HTS schedule citation, effective date, and “in force, one challenge pending, not enjoined.” Input-cost channel onto the fabrication and assembly nodes, magnitude basis = the rate applied to the imported share disclosed in segment reporting. Baseline = the pre-tariff import trend. Advantaged: domestic producers of the substituted input (primary tier). Displaced: downstream converters with no domestic supply and long qualification cycles — stated, not omitted. Observable: margin compression in the next two quarterly filings absent a price increase. No firms named; publishes at Stage 1.

Example B — a procurement shift where a Gate 1 indicator appears

Same method through Stage 3. At Stage 4, procurement records show an award made without competition to a firm whose principal met the deciding official during the award window. The WLM does not characterize this. It records the structural finding (this node became advantaged via the demand channel), notes that a Gate 1 indicator is present, and escalates to trump-corruption-accountability-tracker, which owns the evidentiary standard for that claim. The published WLM says the node was advantaged and that a related inquiry is separately documented — nothing more.

Example C — refused request

“Which stocks benefit from the administration’s energy policy?” — refused under Gate 2. The nearest thing PU produces is a structural map of where energy policy lands in the value chain, with no firms named, no ranking, and no framing a reader could act on as a trade. Offer that; do not produce a ticker list under a civic label.