Review Soul Sp Bp

Review SOUL SP/BP

Public distribution

This package is a smart-hardware-focused implementation. The public distribution contains templates; a private deployment may supply an approved organization SOUL and populated goal.md. Never treat bundled templates as live facts, and never infer private targets or thresholds that are not present in the supplied source.

Install from GitHub with:

npx skills add binbingoes/review-soul-sp-bp

For a global install, add -g; to select the skill explicitly, add --skill review-soul-sp-bp.

Data boundary: SOUL versus goal.md

• The user-supplied soul.md and goal.md are the review’s primary configuration; bundled files are templates only.
• references/soul-decision-kernel.md contains only stable SOUL language and qualitative SP strategic pull. It must not be used as the source of BP periods, budget amounts, quotas, or current financial targets.
• goal.md contains the time-sensitive quantitative layer: SP traction goals, BP results, budget lines, evidence thresholds, owners, release conditions, and stop conditions.
• If SOUL and goal.md disagree, expose the conflict. Do not silently copy BP/budget data into SOUL or treat a goal as a timeless principle.

Audience and review modes

Support three equivalent modes with the same evidence contract:

1. Author self-check: the circle or plan owner supplies a self-check pack. Do not accept its verdict as evidence; independently test facts, conflicts, and gates.
2. Peer/reviewer check: a peer, executive, finance partner, or Agent reviews the plan and returns exact revision actions.
3. Portfolio check: review several circles or business lines separately first, then reconcile shared products, channels, customers, resources, and financial targets. Never merge conflicting numbers or assign the same result to multiple circles without an ownership rule.

The portable output must state source versions, coverage, review mode, confidence, verdict, unresolved gaps, and next decision authority. If an original source is unavailable, mark it GAP; do not imply full coverage.

Turn an SP, BP, or budget draft for a smart-hardware or AI-hardware business into an evidence-based decision review. Apply hard gates before scoring so revenue ambition, presentation quality, or seniority cannot offset red lines, missing unique value, or an untraceable goal/budget chain.

Resource routing

Required user inputs

At the start of every review, look for the user-supplied configuration files in the current review directory:

1. ./soul.md — the organization or business SOUL and SP strategic pull;
2. ./goal.md — the organization or business SP/BP/budget goals and quantitative gates.

If the files are not named exactly this way, ask the user for their paths or use paths explicitly provided in the prompt. Do not silently substitute the bundled templates for a formal review. If either file cannot be read, mark the related pack GAP and stay in preliminary mode.

Read only what the review needs, in this order:

1. Read the user-supplied soul.md for stable SOUL language and qualitative SP pull.
2. Read the user-supplied goal.md for approved SP/BP periods, quantitative goals, metric definitions, budget lines, evidence thresholds, owners, release conditions, and stop conditions.
3. Read references/soul-decision-kernel.md only as a smart-hardware reference or fallback template; it contains no company-specific targets.
4. Read goal.md only as a schema/example reference; never treat it as live organizational data.

• Read references/strategy-frameworks.md when analyzing SWOT/TOWS, Ansoff, growth curves, or Where to Play/How to Win.
• Read references/finance-pull.md for finance evidence intake questions and definitions. It is not a goal source.
• Read references/baseline-intake.md when the plan lacks operating metrics, comparable baselines, or causal targets. Select questions by business type, stage, Ansoff quadrant, and growth curve.
• Read references/review-rubric.md before scoring or assigning a verdict.
• Read references/input-schema.md before running the deterministic scorer.
• Use references/output-template.md for the final report.

Workflow

Step 1: Lock scope and source priority

Identify:

1. review scope: company, business line, product line, platform, or function;
2. artifact type: SP, BP, budget, or combined pack;
3. version, owner, decision body, and review date;
4. planning periods: the SP cycle, BP cycle, and current budget period stated in the user-supplied goal.md;
5. user-supplied SOUL revision and goal.md version.

For a multi-circle or multi-channel review, also identify the boundary pack: circle/series ownership, user and scenario boundary, price band, channel role, shared capability, cannibalization rule, and the decision authority for overlap. A missing boundary pack is a strategic-choice gap, not a formatting issue.

Use sources in this priority order:

1. the latest explicit instruction from the user or decision owner;
2. the user-supplied SOUL and goal.md;
3. the bundled templates;
4. the plan under review.

If versions conflict, expose the conflict. Do not silently choose an older target.

Label every important statement:

• FACT — source-backed observation or number;
• JUDGMENT — management choice or target;
• ASSUMPTION — plausible but unverified causal belief;
• GAP — missing information required for a decision.

If a live document is inaccessible, ask for an export or pasted text. Continue with available material, lower confidence, and keep missing content as GAP.

Do not silently carry a current rule backward into an older plan. Record the user-supplied SOUL revision and goal.md effective period/revision before judging a historical plan.

Step 2: Collect evidence packs, then classify

Do not ask the product/business owner or finance owner to choose a stage, Ansoff quadrant, or growth curve. Ask them for evidence in their own domain; the Agent makes the classification and shows its reasoning.

Collect three separate packs:

1. Product/business pack: own historical performance; users and problems; product, channel, region, and price-band results; market change; competitor and alternative evidence; capabilities and constraints; proposed opportunities and explicit sacrifices.
2. Finance pack: comparable revenue, gross profit/margin, operating profit, cash, inventory/collection, unit efficiency, investment, AI package/human package, definitions, sources, and forecast/budget baselines.
3. Goal pack: populated goal.md, source version, SP/BP/budget periods, quantitative goals, metric definitions, thresholds, owners, and review dates.

For a portfolio or cross-circle review, collect a fourth boundary pack containing:

• who owns each product, customer, channel, price band, and result;
• overlaps between circles or series and the proposed resolution;
• channel requests versus strategic choices;
• shared platform, GTM, marketing, service, and delivery capacity;
• cannibalization, price integrity, and resource-allocation rules.

Read references/baseline-intake.md. Ask no more than six first-round questions to each owner. Do not send the full question bank.

After receiving all applicable packs, the Agent must:

1. reconcile definitions, periods, and sources;
2. classify business type, stage, primary constraint, Ansoff quadrant, and growth curve;
3. cite the facts supporting each classification and state confidence;
4. mark an ambiguous classification provisional and ask only for the missing factual distinction;
5. show conflicts between business and finance inputs side by side; never average conflicting facts or silently choose one owner.

For every requested metric require:

• definition and calculation;
• comparable baseline period and value;
• source and data owner;
• SP/BP target period and value;
• gap, causal driver, responsible owner, and next update date.

Before calculating completion, reconcile the metric subject. At minimum distinguish circle revenue, regional/full-market revenue, sell-in/SI, sell-out/SO, recognized management revenue, operating profit, net profit variants, contribution margin, and cash. A shipment, order, or channel commitment is not a business result unless the plan defines how sell-out, collection, margin, quality, and trust are verified.

Use the baseline periods and target milestones defined in goal.md; if they are absent, record GAP rather than inventing a cycle.

If finance cannot provide a number, record GAP, owner, and due date. A plan with a critical missing baseline may receive a preliminary review, but cannot receive 通过.

🔴 CHECKPOINT — baseline: Before scoring financial quality, show the selected metric dimensions, definitions, baselines, and unresolved gaps to the plan owner or named finance authority. Do not infer a baseline from an aspirational target.

Smart-hardware focus

For every device or AI-hardware opportunity, check the complete system rather than the device claim alone:

• user scenario, device interaction, embedded model, cloud/service dependency, installation, and after-sales loop;
• BOM, tooling, yield, test, certification, firmware/OTA, warranty, returns, inventory, logistics, and service economics;
• activation, adoption, model quality, failure/defect, privacy/security, recommendation, trust, and service-SLA evidence;
• channel price integrity, lifecycle, cannibalization, and one-owner-per-result rules.

Step 3: Normalize the plan

Extract no more than three strategic opportunities or platform initiatives. For each one record:

• target user/customer and problem;
• unique value evidence;
• SWOT facts and TOWS option;
• market-driven or technology-driven;
• Ansoff quadrant;
• first, second, or third growth curve;
• Where to Play, How to Win, and core control point;
• E0-E4 evidence level;
• SP and BP half-year milestones;
• revenue or downstream value, healthy gross margin, cash, trust, unit-efficiency, hardware reliability, and capability assumptions;
• SP/BP/budget goal references from goal.md, including release condition, stop condition, owner, and review date;
• explicit choice and explicit sacrifice.

Treat channel or customer requests as evidence inputs, not automatic priorities. For every request, record the requesting party, reachable user problem, incremental value, full-cost economics, strategic fit, and what the plan will not do. If a product/price/channel boundary overlaps another circle, create a conflict row and pause the affected strategic conclusion until the named authority resolves it.

Treat a platform or function initiative as an opportunity. Evaluate cycle time, quality, reuse, adoption, downstream economic value, and risk avoided instead of demanding direct revenue.

Step 4: Run hard gates in fixed order

Run these gates before scoring:

1. Red lines: safety, integrity, privacy, quality, and compliance.
2. Unique value: horizontal difference, vertical progress, user perception, and payment/trust evidence.
3. Strategic choice: no more than three opportunities; clear choice, sacrifice, and control point.
4. Evidence: facts are sourced; E0-E4 is explicit; authority, consensus, and sunk cost are not treated as evidence.
5. Commercial closure: growth, healthy gross margin, cash, trust, and delivery can form a viable loop.
6. Budget traceability: every budget maps to an SP opportunity, BP milestone, evidence purchase, release condition, and stop condition.
7. Accountability: result owner, decision body, exception owner, review date, and result verification are explicit.

The strategic-choice gate also checks portfolio integrity: one owner per result, no ungoverned circle or series overlap, no channel-exclusive promise without lifecycle and price rules, and no double counting of shared revenue or capability benefits.

Approval evidence boundary

Treat approval as a separate evidence type. AI revisions, AI summaries, dry-runs, generated documents, planned delivery, meeting minutes, executive sponsorship, and verbal support are working artifacts; they do not prove that an SP/BP, budget, exception, or operating result is approved.

Only record approved when the pack identifies the named decision authority, exact approved artifact/version, approval status, effective period, and a verifiable decision receipt or system record. If any element is missing, label approval GAP/unknown, keep the accountability/evidence gate from passing, and use preliminary mode without a numerical score. Never infer operating evidence from an approved plan: approval authorizes a decision; actual results still require dated source records.

Apply the override rules in references/review-rubric.md. A hard-gate failure cannot be repaired by a higher total score.

Step 5: Apply the strategy frameworks

Use the fixed chain:

SOUL SP pull → goal.md → SWOT/TOWS → Ansoff → growth curve → E0-E4 → staged gates → review and reallocation

Reject framework theater:

• SWOT without sources is a list of opinions, not analysis.
• Ansoff classification without a higher evidence bar for farther moves is only labeling.
• A second or third curve without migration milestones is a slogan.
• Where to Play without a sacrifice is not a choice.
• How to Win without a control point, owner, or metric is not a strategy.
• A channel request list without a choice, sacrifice, and economics is not a GTM strategy.
• A product matrix with overlapping price bands and no user/value boundary is not portfolio strategy.

Step 6: Check SP, BP, and budget traceability

For SP, require every milestone in goal.md, target portfolio, growth-curve migration, required capabilities, choices, exits, and multi-year assumptions.

For BP, require every BP milestone in goal.md, key actions, evidence upgrades, financial, hardware reliability, and trust metrics, resource ceilings, and correction against SP.

For the current budget, require the line-level evidence purchase, amount, release condition, stop condition, owner, authority, and review date from goal.md. Do not approve irreversible investment before the goal file’s E3-equivalent gate. Allocate third-curve work as a limited, time-boxed option until evidence supports expansion.

For low-price, ODM, or channel-specific plans, require a full-cost view: product cost, dedicated people, quality/testing, warranty/service, returns, inventory, collection, marketing, channel incentives, brand externality, and shared-capability cost. Do not approve a plan because a narrow marginal-contribution view is positive.

Check company-level portfolio defaults only on the configurable strategic growth and investment budget:

• market-driven / technology-driven: 80/20;
• first / second / third curve: 70/20/10;
• exclude safety/compliance, basic operations, and rigid commitments from the denominator.

Do not force every unit to mechanically match company-level ratios. Require any exception to state reason, owner, validity period, and review date.

Step 7: Score and determine the verdict

Choose review mode before scoring.

Use preliminary mode if the business/product, finance, or goal pack is missing/partial, a critical operating baseline is missing, any hard gate is unknown, or opportunity ownership/release/stop fields are incomplete. In preliminary mode:

• do not output a numerical score or nine-dimension score table;
• do not run the scorer with invented placeholder scores;
• return 暂不评分（资料不完整）;
• keep the report to five sections and at most two main tables;
• ask only for the missing evidence that can change the next decision.

Use preliminary mode as well when a multi-circle review has no independent circle pack, when a source is inaccessible, or when definitions conflict. State which circle, metric, or decision is uncovered; do not imply full portfolio coverage.

Use formal mode only when all three packs, comparable baselines, seven hard gates, opportunity cards, owners, and review dates are complete. Score the nine dimensions in references/review-rubric.md, create the structured card in references/input-schema.md, and run the deterministic scorer.

Run:

python scripts/score_review.py review.json --format markdown

If Python is unavailable, calculate the weighted score manually and apply the same override rules. Never change a hard-gate result to fit the score.

Use only these verdicts:

• 通过 — evidence and operating closure support the approved scale;
• 验证 — direction is plausible, but only validation budget is justified;
• 暂停 — critical evidence, capability, traceability, or ownership is missing;
• 终止 — a red line is hit, unique value is absent, or commercial closure is disproven.

Step 8: Return a revision-ready report

Use the matching preliminary or formal format in references/output-template.md. Lead with:

1. verdict, weighted score, and confidence;
2. hard-gate results;
3. the three most important defects;
4. exact actions to keep, add, delete, validate, or stop;
5. the questions that require the named strategy, product, finance, delivery, red-line, or final decision authority to decide.

Keep management language direct. Do not rewrite the entire plan unless asked. Point to the exact claim, table, milestone, or budget line that must change.

For an author self-check, return two layers: (1) the plan participant's claimed status, and (2) the Agent's independent status. Never let consensus, sponsorship, or prior spend substitute for evidence.

Do not invent a submission date, threshold, score, evidence level, Ansoff quadrant, or curve. Use GAP or provisional until the responsible source supplies the deciding fact.

Scope red-line stops precisely. Stop real-data collection, user exposure, launch, or budget release when those actions create the risk. Do not automatically stop safe offline design, fictional-data testing, or remediation work unless the red-line owner requires it.

🔴 CHECKPOINT — formal decision: Before changing a live SP/BP, releasing budget, stopping a project, or creating a people consequence, present the review record to the named decision authority. An Agent may recommend 终止; it may not silently make the formal company decision.

Failure recovery

Anti-pattern blacklist

Do not:

• reward polished slides, high revenue targets, or executive sponsorship as evidence;
• treat an AI revision, dry-run, created document, meeting summary, verbal support, or planned delivery as formal approval or operating evidence;
• accept six priorities disguised as three themes;
• let SWOT adjectives stand without facts, sources, dates, and implications;
• classify an opportunity with Ansoff or a growth curve but ignore the resulting evidence and budget threshold;
• approve growth that lacks healthy gross margin, cash discipline, trust, or delivery capacity;
• treat AI spend as progress without an adopted workflow, measurable leverage, and result accountability;
• penalize a platform for lacking direct revenue when downstream value is measurable;
• approve full budget because money was already spent;
• invent missing market, competitor, user, margin, cash, or adoption data;
• output a numerical score when source packs, baselines, hard gates, or required fields are incomplete;
• invent dates or thresholds because a current-cycle snapshot contains a nearby deadline;
• hide contradictions by averaging scores;
• output only a score without exact revision actions and decision questions;
• treat a channel, customer, or executive request as a strategy without user, economics, and sacrifice evidence;
• double count the same revenue, SKU, channel, customer, or shared platform result across circles;
• approve a low-price product from contribution margin that excludes dedicated people, quality, service, inventory, cash, or brand cost;
• leave overlapping product-series price bands unresolved while releasing channel or tooling budget;
• publish internal targets, named cases, or confidential financial data outside the authorized audience.

Completion gate

Call the review complete only when:

• source version, scope, owner, and periods are explicit;
• the approved SOUL source and populated goal.md version are explicit;
• business type, stage, quadrant/curve, metric definitions, baselines, sources, owners, and target periods are explicit;
• no more than three opportunities or initiatives are reviewed;
• all seven hard gates have evidence and a status;
• cross-circle boundaries, ownership, overlaps, and double-counting rules are explicit when applicable;
• every opportunity has Ansoff, curve, E-level, WTP/HTW, control point, milestones, and a budget gate;
• SP, BP, and budget traceability is visible;
• financial, unit-efficiency, trust, and AI Native assumptions are separated from facts;
• the deterministic scorer passes or manual fallback is disclosed;
• verdict, confidence, top gaps, revision actions, authority, and next review date are present.
• numerical scoring appears only in formal mode after the scorer confirms input completeness.