@1aboveio/skills 1.0.10 → 1.0.11

package/skills/risk-compliance/tld-detection/SKILL.md

@@ -1,245 +1,130 @@
 ---
 name: tld-detection
 description: >
-  Explicit-only. Potential TLD (transaction-laundering) clusters from an
-  authorization extract: payment-pattern inference (direct or FX-converted
-  top-up hypotheses, e-commerce patterns, fixed-amount characteristics), then
-  descriptor-local exact original-currency amount pairs plus a separate
-  BIN-country base-currency FX sensitivity when requested or indicated, with
-  same-descriptor baseline contrasts, high/medium/low risk
-  tiers, exposure sizing, and a plain-language investigation conclusion
-  with a ≤3-sentence summary.
-  Not payment-analysis, fraud-analysis, or website research (similar-domain).
+  Explicit-only. Investigate potential transaction-laundering patterns in
+  transaction data using amount concentration, escalating amounts and pricing
+  patterns, with USD and BIN-country currency views, group profiles and priorities.
+  Not payment-analysis, fraud-detection, or website research.
 disable-model-invocation: true
-release-version: 0.44.1
+release-version: 0.45.0
 ---
 **Updates:** Once/session run `node <skill-base-dir>/release-check.mjs`.
-# TLD Detection
-
-Find **potential TLD clusters** in an authorization extract. TLD here is
-transaction laundering, not a domain name: hidden activity stuffed under a
-normal MCC, often mixed across descriptors.
-
-First partition the cleaned authorizations by **descriptor**, then infer
-**payment patterns** inside each descriptor: a billing name is not a verified
-merchant identity. Patterns describe observed amount and behavior features;
-they do not partition canonical detection. The clustering key remains the
-original **currency-amount-pair** within the whole descriptor. Geometric
-relationships between exact anchors are a stuffing signal (掺量), not proof
-of hidden business. Presets converted from an inferred top-up currency can
-vary across dates and fragment exact ladders; pattern inference provides
-separate context for those observations. BIN country and inferred patterns
-are not canonical membership keys.
-
-The original-currency run is canonical. When the user requests home/base-
-currency detection, or when date-varying transaction amounts may conceal fixed
-prices in card-origin currencies, add the separate BIN-country base-currency
-FX sensitivity in [references/fx-conversion.md](references/fx-conversion.md).
-Never merge its `FXC*` membership, exposure, or risk tiers into canonical
-`C*` results.
-
-**Produce:** a concise investigation report, complete linked evidence, and a
-separate run-specific methodology document
-under `reports/tld-detection/`.
-**Do not produce:** confirmed laundering, a legal finding, or supervised
-fraud-rule packages (`fraud-analysis`).
-
-Paths under `references/` are relative to this skill.
-
-## Workflow
-
-For a new analysis, follow these steps in order. For presentation-only work,
-first locate the accepted saved results, complete evidence, methodology, and
-original chart inventory. Verify their revision and scope, then go directly to
-step 5; do not rerun steps 1–4 or reinterpret accepted findings. If needed facts
-or charts are unavailable, disclose the gap and request the source artifacts
-instead of silently rebuilding them. Detail:
-[references/workflow.md](references/workflow.md).
-
-| Step | Do | Gate |
-|---|---|---|
-| 0 Pre-flight | File, grain, columns, window | Stop if auth grain, descriptor, amount, or currency column is missing; caption missing BIN country |
-| 1 Clean | Dedup | Record drop counts; do not filter BIN country |
-| 1b Patterns | Infer payment patterns with `scripts/detect_streams.py`; use direct top-up, FX-converted top-up, e-commerce, or other names; fixed amounts are a supplementary characteristic | Show coverage, inferred top-up currency, actual charge currency, and tested rate consistency; retain other and small-sample groups; names are hypotheses |
-| 2 Cluster | Run canonical detection on all cleaned attempts in each descriptor, irrespective of inferred pattern; add FX sensitivity separately; freeze exact keys and same-descriptor baseline | Partition = descriptor only; ≥3 levels; adjacent levels = 2x; ≥100 attempts; absolute residual cap per candidate currency |
-| 3 Risk | Assign high/medium/low investigation priority | In canonical results, Mixed BIN countries are neutral and exact 100% single-country concentration is a standout; FX concentration is descriptive only and cannot raise tier |
-| 4 Decision | Who to review, why, what is unknown, and what to check next | Summary ≤3 short sentences; supporting explanation allowed; no new evidence |
-| 5 Report | Answer highest-risk, cluster-characteristic, and exposure questions; write; `editorial-review`; `polish-cn` on Chinese; proofread | Preserve facts and tiers; sort high → medium → low; no confirmed-TLD wording |
-
-### 0 — Pre-flight
-
-Authorization extract, one row per auth attempt. Record paths, time field,
-amount, currency, BIN/issuer country, descriptor, MCC, response. Grain and
-column aliases: [workflow.md](references/workflow.md).
-
-### 1 — Clean
-
-[references/clean.md](references/clean.md). Deduplicate first. Do not
-FX-convert the canonical clustering key. Do not drop non-US BIN rows.
-
-### 1b — Payment patterns
-
-[references/streams.md](references/streams.md). Run
-`scripts/detect_streams.py` on the cleaned universe with a daily FX cache.
-Use **Direct top-up pattern / 直接充值模式**, **FX-converted top-up pattern /
-换汇充值模式**, **E-commerce pattern / 电商模式**, and **Other pattern /
-其他模式** in reader-facing reports. Compare the inferred top-up currency
-with the actual charge currency, not BIN-country legal tender, to distinguish
-direct from FX-converted top-ups. Fixed amounts are an additional feature,
-not a separate business class or proof of a fixed conversion rate. Raw `stream`
-labels remain available for audit. These inferred patterns cannot change
-canonical members, thresholds, or the same-descriptor baseline.
-
-### 2 — Cluster
-
-[references/clustering.md](references/clustering.md). Build normalized
-descriptor partitions, then rank currency-amount-pairs inside each descriptor. Use
-`scripts/detect_ladders.py` to detect geometric anchor relationships with
-adjacent, currency-specific absolute residual caps. Every cluster must contain
-at least three theoretical levels with a fixed 2x relationship between adjacent
-levels (for example 1x, 2x, 4x, 8x); search both up and down from every candidate
-seed. A theoretical level may
-keep multiple exact anchors, but each additional anchor must clear the stated
-portfolio support floor. Freeze membership by exact descriptor, currency,
-and amount; keep only clusters with ≥100 authorization attempts. Compare each
-cluster with the baseline formed by removing that cluster from the same
-descriptor. Do not invent clusters from MCC or comparison features.
-
-If the FX sensitivity applies, run it only after the cleaned universe and
-canonical partitions are frozen. Follow
-[references/fx-conversion.md](references/fx-conversion.md) and its mandated
-Frankfurter v2 request/cache contract. Within each frozen canonical descriptor
-partition, `|P_FX|` is only its exact-date-converted attempts; use only `P_FX`
-for FX high-count and rung-pool thresholds and the FX same-descriptor baseline.
-Excluded rows stay outside FX detection and appear only in exclusion coverage
-and reconciliation to canonical `U`. Use `FXC1`, `FXC2`, ... ids and publish
-separate result and methodology files. Because BIN country constructs the
-target currency in this view, country concentration, including 100%, is
-descriptive only and cannot raise the FX tier.
-
-### 3 — Risk
-
-[references/decision.md](references/decision.md). Assign every identified
-currency-amount pair exactly one investigation-priority tier: High, Medium,
-or Low. In canonical results, BIN-country mixture is neutral and exact 100%
-single-country concentration is a standout; separately report descriptor-
-country slice coverage without a threshold or risk points. In FX results,
-country concentration remains descriptive because it constructs the target
-currency and is not independent risk evidence.
-
-### 4 — Decision
-
-Write a business-readable investigation conclusion following
-[references/decision.md](references/decision.md): who to review first, why,
-what remains unknown, and what evidence to check next. Limit the summary to
-**at most 3 short sentences**, not the full explanation. Keep detailed amounts
-and exposure in supporting tables; no qualifying cluster does not mean low risk.
-
-### 5 — Report
-
-[references/report-template.md](references/report-template.md). Use three main
-sections: Period conclusion, Main findings, Next checks (Month conclusion for
-monthly reports). Within each finding, explain observed evidence before the
-payment-pattern hypothesis and its reconciled relationship to amount clusters.
-Pattern count is not a count of upstream businesses. If a supported menu has
-non-doubling presets, explicitly explain in the finding why its prices need
-not satisfy the formal cluster gate; do not leave that distinction implicit
-in separate amount lists. Terms:
-[references/terminology.md](references/terminology.md).
-
-Use the main-report, evidence, and methodology filenames listed in
-[report-template.md](references/report-template.md); provide matching Chinese
-Markdown and HTML where required.
-
-Keep the main report selective: supported findings and useful original charts,
-not repeated conclusions or exhaustive tables. Preserve every identified
-cluster, its descriptor share, comparisons, and by-tier/by-descriptor exposure
-in complete linked evidence. Sort High → Medium → Low, then by evidence
-strength and attempts inside each tier. Put the
-actual run's cleaning, partitioning, parameter selection, sensitivity evidence,
-baseline definition, and audit details in the methodology documents; link the
-three document types together. HTML is preferred when there are charts.
-Use new revision-qualified paths for delivered revisions; never overwrite prior
-outputs. Presentation-only revisions reuse frozen results without rerunning
-inference, detection, or risk assignment. Retain all valid original charts,
-putting important amount/rate charts inline and secondary views in expandable
-sections; verify chart payload fidelity and desktop/mobile rendering.
-Review the written documents with `editorial-review`
-(`{editorial-review baseDir}/SKILL.md`), naming
-[references/report-template.md](references/report-template.md) as its template. For Chinese deliverables,
-including a Chinese final chat handoff, then run `polish-cn`
-(`{polish-cn baseDir}/SKILL.md`) with the coverage, protected terms and guardrails in
-[references/chinese-polish.md](references/chinese-polish.md). Then verify factual
-fidelity, synchronize Markdown and HTML, and grep banned strings before delivery.
-
-## Progress
-
-Track a task plan. One item is `in_progress` at a time.
-
-1. Pass pre-flight
-2. Freeze cleaned universe and descriptor partitions
-2b. Infer payment patterns and record them per descriptor without splitting detection
-3. Assign High / Medium / Low risk tiers
-4. Publish complete cluster evidence + same-descriptor baselines
-5. Write main report: conclusion (≤3 sentences), findings, next checks
-6. Link report, evidence, and run methodology; retain charts and prior revisions; run `editorial-review`, then `polish-cn` on Chinese; verify and proofread
-
-## Boundary
-
-| Need | Skill |
-|---|---|
-| Potential TLD clusters, currency-amount pairs, stuffing / 掺量, MCC miscoding, descriptor mixing, upstream streams behind one descriptor | `tld-detection` |
-| Which websites, mirrors and operators sit behind a merchant domain | `similar-domain` |
-| Volume, journey Sankey, auth rate, decline mix, settlement overview | `payment-analysis` |
-| Proxy/confirmed fraud labels, holdout precision/recall, rule packages | `fraud-analysis` |
-
-A suspected-fraud *rate contrast* in this skill is not a fraud-analysis
-rule package.
-
-## Reference index
-
-| Need | Read |
-|---|---|
-| Grain, columns, window | [workflow.md](references/workflow.md) |
-| Dedup | [clean.md](references/clean.md) |
-| Payment-pattern names, direct vs FX-converted top-ups, fixed-amount characteristics, behavior metrics | [streams.md](references/streams.md) |
-| Currency-amount pairs, anchors, same-descriptor baseline | [clustering.md](references/clustering.md) |
-| BIN-country base-currency FX sensitivity | [fx-conversion.md](references/fx-conversion.md) |
-| Business-readable conclusion and ≤3-sentence summary | [decision.md](references/decision.md) |
-| Report chapters | [report-template.md](references/report-template.md) |
-| Chinese polish coverage, protected terms, and fidelity checks | [chinese-polish.md](references/chinese-polish.md) |
-| Language / banned strings | [terminology.md](references/terminology.md) |
-
-## Sanity check
-
-- [ ] Auth grain; dedup counts are stated; no BIN-country row filter
-- [ ] Partition key is normalized descriptor; missing descriptor is an explicit bucket
-- [ ] Payment-pattern inference ran; direct/FX-converted top-up, e-commerce, and other labels are hypotheses; inferred top-up and actual charge currencies are separate
-- [ ] Fixed amounts are a supplementary characteristic; fixed FX requires tested source-to-top-up rate movement, not same-currency prices or untested `live` labels
-- [ ] Canonical detection, exact membership, thresholds, and baseline use the whole descriptor; inferred patterns and BIN-country groups never filter members
-- [ ] Clustering key is original currency + amount, not USD
-- [ ] Canonical `C*` results use original amounts; any FX run is separate, uses `FXC*` ids, and does not alter canonical membership or exposure
-- [ ] FX run records as-of-date country→legal-tender mapping, the versioned immutable Frankfurter v2 exact-date request/cache facts, formula, minor-unit rounding, exclusions, and converted-universe coverage
-- [ ] Within each frozen descriptor, `|P_FX|` contains only exact-date-converted attempts; FX thresholds and same-descriptor baselines use only `P_FX`, while exclusions reconcile only to canonical `U`
-- [ ] FX country concentration is descriptive only because BIN country constructs the target currency; it does not independently raise risk
-- [ ] Each rung-pool currency has an explicit adjacent absolute-residual cap and sensitivity evidence
-- [ ] Multi-anchor levels state their portfolio support floor; final membership uses exact selected amount keys
-- [ ] Cluster attempts ≥100
-- [ ] Every cluster has a High / Medium / Low risk tier and results sort in that order
-- [ ] Currency-amount pairs show seed and representative anchors; `*` marks a multi-amount level
-- [ ] Same amount pair may recur across descriptor partitions; never merge them automatically
-- [ ] BIN country is post-membership reporting data, not a membership key; Mixed is neutral
-- [ ] Canonical exact 100% single-country concentration is a standout; FX concentration is descriptive only; descriptor-country coverage has no risk threshold
-- [ ] Each cluster baseline is the same descriptor after removing that cluster
-- [ ] Complete linked evidence, not the main report, uses one consolidated profile/comparison table per cluster; descriptor, seed, anchors, attempts, descriptor totals/share, MCC, and baseline metrics are not repeated in separate definition/characteristic/metric tables
-- [ ] Anchor table reports attempts, authorization success rate, and suspected-fraud share for every level, including n < 30
-- [ ] Characteristics contain descriptor total, cluster attempts/share, MCC, and cluster-vs-baseline auth rate, suspected-fraud share, top 3 banks, and top 3 BIN countries
-- [ ] Summary is ≤3 short sentences; explanation answers who to review, why, what remains unknown, and what to check next without decoding technical labels
-- [ ] No qualifying cluster is not described as low merchant risk; unqualified leads have no cluster tier or cluster exposure; current-period findings are separate from historical results
-- [ ] Report answers highest-risk, identified-cluster characteristics, and by-tier/by-descriptor exposure
-- [ ] Main report has exactly three sections: conclusion, findings, next checks; findings explain observations before hypotheses and reconciled cluster relationships
-- [ ] Main report, complete evidence, and methodology are separate and mutually linked; prior revisions and frozen facts remain intact
-- [ ] Valid original chart payloads are retained; important amount/rate charts are inline, secondary charts render on expansion, and desktop/mobile views are nonblank without page overflow
-- [ ] Zero families → ranking only; no descriptor-invented clusters
-- [ ] Language matches terminology.md; no confirmed-laundering claim
-- [ ] Documents passed `editorial-review`; Chinese report, evidence, methodology, chart captions, and handoff passed `polish-cn`; the model that ran is recorded, facts are unchanged, and Markdown / HTML agree
+# TLD detection
+
+Find patterns that warrant investigation, not proof of transaction laundering
+(TLD). Declare selection thresholds and rating criteria before interpreting results.
+
+## Data contract
+
+One row represents one **transaction**. Require transaction time, transaction ID,
+currency, amount, MCC, billing descriptor, MID, card scheme, card BIN, card issuer,
+and card BIN country. Record column aliases, types and missingness. Use the requested
+reporting period; otherwise state the cleaned data's observed time range.
+
+Missing required columns or unresolved transaction identities block analysis.
+Request the data or an explicit aggregation rule; do not substitute authorization
+attempts or orders. A constant MID needs an established value and scope. The prepared
+script input requires nonempty identity and business fields; resolve missing values
+before running it. BIN country may be null: retain that transaction in USD if valid,
+and explicitly exclude its BIN-country view. Status fields may independently be
+unknown. Never invent missing dimensions or merge unknown scopes.
+Follow [transaction preparation](references/clean.md) for deduplication and normalization.
+
+Rates additionally require transaction-level authorization status and a suspected-
+fraud indicator or documented response-code mapping. Missing status is unknown,
+not failure or zero fraud. Report coverage and withhold complete-population rates
+when classifications are incomplete. A separately labeled known-only rate may
+accompany coverage, but cannot replace the requested total-denominator rate.
+
+## 1. Pattern sniffing
+
+Prepare two views from the same cleaned transactions: **USD-converted amounts**
+and **BIN-country-currency amounts**. Follow [FX conversion](references/fx-conversion.md)
+for dated mappings, exact-date quotes, rounding and exclusions. Partition each view
+by MID, normalized billing descriptor and target currency within the reporting period.
+Keep the views distinct even when both target USD.
+
+Look independently for:
+
+- **Amount concentration:** repeated prices such as 8.00 or 9.9. Declare minimum
+  transaction count and share before selection; a concentration needs no ladder.
+- **Escalating amounts:** doubling prices such as 8, 16, 32, 64. Default to at least
+  three levels and 100 transactions. Keep the multiplier at 2x; record any changes
+  to level count, minimum support or discovery tolerances.
+- **Pricing patterns:** supported preset menus such as 10, 20, 50, FX-converted
+  menus, or retail-like prices sharing a fitted multiplier. Keep inferred pricing
+  currency separate from charge currency and BIN-country currency.
+
+Read [pattern.md](references/pattern.md) for detection methods, eligibility and
+membership. Freeze exact transaction IDs for every group and amount keys where
+applicable. Retain below-threshold and insufficient-evidence candidates without
+priority tiers. Pricing groups follow the same profiling and rating steps; they do
+not need to pass the escalation gates.
+
+Supply prepared transaction views; these scripts do not fetch rates or prepare
+live FX conversions. Follow the [script and metric contract](references/metric-contract.md)
+for the dataset schema, exclusions, output schemas and formulas. Replace `{baseDir}`
+with the absolute skill directory and use thresholds agreed for the analysis. Input and
+output paths below are relative to the working directory; these values are examples:
+
+```bash
+python3 "{baseDir}/scripts/detect_patterns.py" dataset.json groups.json \
+  --pattern concentration --concentration-min-count 30 --concentration-min-share .2 \
+  --pattern escalating --cap USD=.15 \
+  --pattern pricing --pricing-currency EUR --pricing-rates rates.json
+python3 "{baseDir}/scripts/profile_groups.py" dataset.json groups.json profile.json
+python3 "{baseDir}/scripts/render_charts.py" profile.json charts.html --overwrite
+```
+
+Give `--cap` for every target currency in the prepared views. `--pricing-rates`
+is optional and supplies an exact-date adapter, not a live feed; see
+[pattern.md](references/pattern.md). Pricing quotes use UTC transaction dates;
+profiles use the declared reporting timezone. `profile_groups.py` accepts optional `--tier-edges`
+followed by sorted decimal boundaries. Outputs refuse replacement unless
+`--overwrite` is explicit; omit it above for a new chart path.
+
+## 2. Group profiling
+
+Compare each eligible group with the **out-group**: converted transactions in the
+same MID, descriptor, view, target currency and period, minus that group's IDs.
+Other groups remain in the out-group; conversion exclusions do not. Save membership
+and population counts. Groups can overlap across patterns or views: use the union
+of transaction IDs for combined totals and state how overlapping tiers are handled.
+
+For both populations, present daily amount-tier counts, success and suspected-fraud
+rates, transaction counts and shares, and 24-hour counts. Declare the timezone,
+default UTC, and show status coverage and unavailable rates. See the
+[metric contract](references/metric-contract.md) for denominators, zero-filling and
+reconciliation. Add issuer, BIN-country, MCC and card-scheme composition where useful.
+
+## 3. Rating and decision
+
+Assign **investigation priority** from independent, material profile differences:
+
+- **High:** several independent, materially adverse observations converge.
+- **Medium:** at least two independent material differences warrant follow-up,
+  but High is not supported.
+- **Low:** adequate observed evidence provides little corroboration beyond the
+  amount pattern. Insufficient evidence leaves a group unrated, not automatically Low.
+
+State the basis, sample size, materiality criteria and confidence. Avoid counting
+correlated observations twice or inventing a calibrated numeric score. BIN country
+constructs the target currency in that view, so its concentration cannot independently
+raise priority. MCC and descriptor alone do not establish a business mismatch.
+No qualifying group does not establish low merchant risk.
+
+Explain whom to review, why, what remains unknown, and which orders, products,
+sellers or fulfillment records to check next. Do not claim unperformed checks or
+recommend enforcement from these patterns alone. Keep current and historical findings
+separate.
+
+## Deliver
+
+Write a concise report in the requested language, preserving evidence, methodology
+and useful charts under revision-qualified paths in `reports/tld-detection/`.
+Read `{house-style baseDir}/references/analytical-reporting.md` and its required
+references, and use `{write-report baseDir}/SKILL.md` with the frozen analysis for the
+report stage. If a required skill or reference is unavailable, report the blocker.
+Follow [reporting](references/report-template.md) for evidence contents, terminology,
+review and presentation-only revisions; do not rerun analysis for report writing. The scripts produce detection, profiles and
+charts; interpretation, priority and the final report still require analyst judgment.

package/skills/risk-compliance/tld-detection/agents/openai.yaml

@@ -1,6 +1,6 @@
 interface:
   display_name: "TLD Detection"
-  short_description: "Risk-tiered TLD clusters from exact currency-amount pairs"
+  short_description: "Investigate transaction amount patterns across currency views"
 
 policy:
   # Codex counterpart to SKILL.md disable-model-invocation: true

package/skills/risk-compliance/tld-detection/references/clean.md

@@ -1,51 +1,18 @@
-# Data cleaning
-
-Run on the authorization extract before any clustering. Record every drop
-count in the run-specific methodology and retain the cleaned census in linked evidence.
-
-Do not filter to US-BIN here. Issuer country remains available for the
-cluster-versus-baseline Top 3 BIN-country comparison.
-
-## 1. Exact-row dedup
-
-Drop bitwise-duplicate rows (same values in every column used). Count
-them. They are extract artefacts, not extra attempts.
-
-## 2. Identity dedup
-
-If `Transaction ID` / `Auth ID` is unique when present, keep one row per
-id (first by time).
-
-Otherwise, if `Order ID` is present, keep one row per
-(`Order ID`, time, amount, currency, masked PAN / last4 if present).
-Retries of the same order at different times stay distinct.
-
-If none of those identities exist, **skip this step** — keep the
-exact-row-deduped table and say identity-dup was skipped. Do not stop
-the run. Do not invent a customer id from BIN + amount.
-
-## 3. Amount and currency
-
-- The canonical clustering key uses the **original** amount and currency. A BIN-country base-currency sensitivity may derive a separate key only after this cleaned universe is frozen; see [fx-conversion.md](fx-conversion.md).
-- Normalize `10`, `10.0`, `10.00` to the same decimal key.
-- Drop rows with missing amount or currency; count them.
-- Do not FX-convert before clustering. A 9.99 EUR rung is not a 9.99 USD
-  rung. Amount sums later stay inside one currency; never add EUR to USD.
-
-## 4. Missing dimensions
-
-Rows with missing descriptor or BIN / issuer country stay in the
-clustering universe. Missing descriptors enter the `UNKNOWN_DESCRIPTOR`
-partition; missing countries enter the `UNKNOWN_BIN_COUNTRY` reporting
-bucket after membership is frozen.
-
-## Output of this step
-
-A cleaned table plus a one-line census:
-
-```
-raw N → exact-dup drop n1 → identity-dup drop n2 (or skipped)
-→ missing amount/currency n3 → cleaned universe U
-```
-
-All later attempt counts use `U`, not `raw N`.
+# Transaction preparation
+
+Record source rows, duplicate removals, unresolved IDs, exclusions and the final
+transaction count. Remove only demonstrably identical extract duplicates. Inspect
+all rows sharing a transaction ID: conflicting times, statuses, amounts or other
+events may represent retries or state changes. Never keep the first row by default
+or collapse conflicting records without an explicit, documented transaction-level
+aggregation rule. If no rule resolves them, stop and ask; do not silently switch
+to authorization-attempt or order grain. Preserve a row-to-transaction audit map.
+
+Normalize currency codes, exact decimal amounts and descriptor whitespace/case;
+do not fuzzy-merge descriptors or strip suffixes without documented semantics.
+Keep MID separate. Resolve missing identity and business values before preparing
+script input; the validator rejects incomplete required values instead of silently
+dropping rows. BIN country and metric statuses are the documented exceptions.
+Keep rows with missing BIN country in the USD view when otherwise valid;
+they cannot enter the BIN-country view without a valid mapping. See
+[FX conversion](fx-conversion.md). Never sum amounts in different currencies.