See all
54 items
Before you flip an AI agent live, score its guardrails. Five operational controls — approval gate, logging, rollback, bounded scope, escalation — one verdict: READY / FIX / DO NOT DEPLOY. A destructive action with no human approval is a hard stop, no matter how high the rest scores. Deterministic, offline, runnable.
Decide whether a proposed AI agent is worth building before you spend on it. Scores six use-case-fit dimensions (genuine autonomy need, bounded scope, integration readiness, measurable outcome, ROI survives mitigation, failure fallback) weighted to 100 and returns BUILD / PILOT FIRST / DON'T BUILD per agent, with a slate rollup (SLATE READY / SOME TO PILOT / STOP-FIRST). Two worsen-only structural gates force DON'T BUILD — the mitigation erases the ROI (the controls cost as much as the agent returns), or there's no genuine autonomy need (automation rebranded as an agent) — so an agent can score 94/100 and still not be worth building. Names the one thing to fix first. Runnable Python engine + workbook + facilitator & build/pilot/stop playbooks + a 6-project sample. A planning decision aid from your own numbers; it never builds or runs an agent and scores no person. Not financial or legal advice.
Stop paying for AI seats nobody uses. Score every license on one tool's roster across five weighted signals — KEEP / DOWNGRADE / RECLAIM per seat — with a dead-seat gate that forces RECLAIM on inactive, cold licenses no matter how they score, and a fleet waste rate that surfaces what the average hides. Deterministic, offline, runnable.
AI makes a fabricated application read as cleanly as a true one. Triage how verifiable an application is — the document, not the person — across six signals into LOW-RISK / FLAG TO VERIFY / HIGH-RISK, with a keystone gate that flags any uncorroborable core claim no matter how polished the rest. Every verdict routes to verification, never to a reject. Not a hiring, screening, or ranking tool; not legal advice.
A thorough postmortem isn't a closed incident. Grade an AI-incident postmortem across six sections of a blameless writeup into CLEARED TO CLOSE / FINISH ACTIONS / NOT CLOSEABLE, with an open-critical-action gate that refuses to close while a single critical fix is still open — no matter how complete the writeup. Grades the document and remediation state, not people. Deterministic, offline, runnable.
Before you install a third-party MCP server or agent skill, grade whether it should be trusted at all. Mark six trust signals (publisher provenance, change history, declared-vs-actual scope, credential handling, approval behavior, sandbox containment) — the verdict is the weakest signal, TRUSTED / REVIEW / DO NOT INSTALL — with a supply-chain gate that forces DO NOT INSTALL whenever credential handling AND approval behavior are both below full, the clean-then-poisoned pattern, even at 75/100. Names the signal to vet first; rolls a registry up to GOVERNED / DRIFTING / EXPOSED. Runnable Python engine + workbook + vetting & trust-hardening playbooks + a 5-artifact sample. Grades the artifact, never people. Not a code scanner or malware detector.
Would a deepfake call or spoofed-vendor email actually be stopped before the money left? Grade a payment-approval process on six out-of-band verification controls (callback to a known-good number, second approver, sender/domain check, escalation & stop path, code-word protocol, bank-change re-verify) — the verdict is the weakest control, DEFENSIBLE / TIGHTEN / NOT DEFENSIBLE — with an independent-validation gate that forces NOT DEFENSIBLE whenever the callback AND second-approver controls are both below full, since a single person can still push an urgent request through, even at 79/100. Names the control to fix first; rolls a portfolio up to GOVERNED / DRIFTING / EXPOSED. Runnable Python engine + workbook + facilitator & control-hardening playbooks + a 5-process sample. Grades the process, never people. Not fraud detection or legal advice.
Most businesses overpay for card processing by 20-40%, and in 2026 several processors met interchange cuts by adding padded line items to claw the savings back. This audit grades a merchant-processing statement on the two axes that decide whether you're overpaying: your realized effective rate (total fees / total volume) against the well-optimized band you confirm for your channel, and the fee padding a clean-looking rate hides. Each statement scores FAIRLY PRICED / RENEGOTIATE / BEING BLED on rate, then a worsen-only padding gate floors the verdict to BEING BLED whenever junk fees hit 15% of the bill or unexplained downgrades hit 20% of volume — so a statement can sit inside its rate band and still read BEING BLED, the money a rate calculator never surfaces. The shipped teaching case: a 2.10% card-present statement inside the 2.20% band floors to BEING BLED on a 16% junk load; strip the padding and it releases to FAIRLY PRICED. It names the one fix to make first, totals the reclaimable dollars per statement, and rolls a portfolio up to ALL FAIR / OVERPAYING SOMEWHERE / MONEY LEAVING THE BUILDING. Runnable zero-dependency Python engine + workbook (Start Here / Dashboard / Fee-Padding Audit) + Statement-Audit SOP + Fee-Renegotiation Runbook + a 4-statement sample. Deterministic and offline; audits the statement you paste, sets no rate, moves no money, contacts no processor. Grades a fee statement, not a person. Not financial, accounting, or legal advice.
Could a poisoned web page, email, ticket, or document quietly take over your AI agent? Grade an agent's exposure to indirect prompt injection on six design controls (weighted to a 0-100 score, CONTAINED / HARDEN / HIGH EXPOSURE) with a dispositive kill-chain gate that forces HIGH EXPOSURE when untrusted input can enter, an unscoped tool can act, and no human is in the loop to catch it, even at 72/100. Names the control to harden first; rolls a fleet up to GOVERNED / DRIFTING / EXPOSED. Runnable Python engine + workbook + exposure-mapping & control-hardening playbooks + a 5-agent sample. Grades the agent's design, never people. Not a scanner or red-team.
A backup you've never restored isn't a recovery plan. A pre-incident drill: would you actually recover from ransomware or a critical AI-dependency outage? Score six recovery controls per system (tested timed restore, immutable/isolated backups, AI-dependency fallback, RTO defined & validated, incident runbook, comms plan) for a weighted 0-100 number and RESILIENT / GAPS / WOULD NOT RECOVER, with a two-trigger untested-recovery gate that forces WOULD NOT RECOVER when you've never run a timed restore OR have no fallback for a load-bearing AI dependency — even at 78/100. Names the control to drill first; rolls a portfolio up. Runnable Python engine + workbook + recovery-readiness & drill-and-fix playbooks + a 6-system sample. The pre-incident companion to the AI Incident Postmortem & Readiness Gate. Grades the posture you describe, never people. Not a backup tool, DR platform, or the rehearsal itself.
Grade your non-human-identity governance — service accounts, API keys, tokens, OAuth apps, agent credentials — on the six controls that decide whether a leaked or orphaned secret becomes a breach. The verdict is the weakest control (GOVERNED / SPRAWLING / UNMANAGED, mean shown for context only), with a dispositive leaked-key gate that forces UNMANAGED when secrets are both un-vaulted and un-revocable — reachable and un-killable at once — so a class averaging 83% still reads UNMANAGED. Names the control to harden first; rolls the org up to GOVERNED / DRIFTING / UNMANAGED. Runnable Python engine + workbook + inventory & credential-hardening playbooks + a 5-class sample. The credentials layer beneath the Connector Access Auditor's reach layer. Grades the setup, never people. Not a scanner, vault, or discovery tool.
The post-pilot scale-or-kill decision: grade a running AI pilot on whether it earned production. Six weighted signals from your own marks (measured outcome vs. target and cost-vs-value are gates) for a 0-100 score and one verdict — SCALE / EXTEND PILOT / KILL — with a worsen-only two-trigger gate that forces KILL when a pilot was never measured or loses money, even at 76/100. Names the signal to fix first; rolls a portfolio up to ALL SCALE / SOME TO EXTEND / KILL-FIRST. Runnable Python engine + workbook + facilitator & proof-out/kill playbooks + a 6-pilot sample. The post-pilot companion to the Agent Use-Case Fit Gate (which decides whether to build before you spend). Grades the pilot you describe, never people. Not financial, investment, or legal advice.
Decide whether a live support bot is fit to stay live before it quietly costs you customers. Mark six fitness signals 0/1/2 (grounded answering and verified resolution are gates) — the verdict is the WEAKEST signal, KEEP LIVE / RESTRICT SCOPE / PULL BACK, with the mean shown for context only. A worsen-only vanity-deflection gate forces PULL BACK when grounding AND verified resolution are both only partial — the bot improvises answers and counts abandoned chats as resolved — so a bot averaging 83 still reads PULL BACK. Names the signal to fix first; rolls a fleet up to ALL LIVE / SOME TO RESTRICT / PULL-BACK FIRST. Runnable Python engine + workbook + fitness-review & keep/restrict/pull-back playbooks + a 6-bot sample. The fitness lane beside the Support Deflection Honesty Grader (is the number real) and the Customer-Facing AI Output Risk Triage (legal exposure). Grades the bot's fitness, never people. Not a compliance certification or legal advice.
The founder's go/no-go on the scariest hire — the first one. Most owners hire on exhaustion, not readiness, then can't make payroll or hire before the work is documented and stay the bottleneck. Mark six weighted signals 0/1/2 (fully-loaded cost coverage and work-documented are gates) for a 0-100 score — HIRE NOW / BRIDGE WITH A CONTRACTOR / NOT YET — with a worsen-only two-trigger gate that forces NOT YET when revenue can't cover the fully-loaded cost or the work lives only in your head, even at 76/100. Names the one thing to fix first; rolls a slate up to ALL CLEAR TO HIRE / STAGGER THE HIRES / HOLD HIRING. The BRIDGE band is the honest middle — contractor first, convert when coverage is solid. Runnable Python engine + workbook + readiness & get-to-go playbooks + a 6-role sample. A decision aid that grades the hiring decision, never a person. Worker classification varies by agency/state — confirm affordability with your CPA and classification with counsel. Not financial, tax, or legal advice.
The risk that doesn't show in a monthly report until a renewal goes sideways and a third of your revenue walks. Paste your revenue-by-customer list (and your monthly cost base) and the gate computes single-customer share, top-5 share, and a 0-100 safety score — DIVERSIFIED / CONCENTRATED / DANGER — weighting the single-customer share heavier (100 under 10%, 0 by 40%). A worsen-only survival gate forces DANGER when losing your #1 would drop you below break-even, no matter how the percentages look — the Consultancy sample scores 51 and still reads DANGER. No cost base? It degrades to flagging any single customer over 35%. Names the account to address first and the diversification target to get your top under 25%. Rolls a portfolio up to ALL DIVERSIFIED / WATCH THE MIX / DANGER PRESENT. Computed from your own numbers, not a mark grid; runnable Python engine + workbook + audit & diversification playbooks + a 6-business sample. Grades a revenue mix, never a person. Not financial, investment, or accounting advice.
Customer concentration is the risk everyone screens for — this is its mirror, and the buyer fear is sharper: lose a critical supplier and you don't trim revenue, you can't make the product. Enter your spend by supplier and the gate computes each supplier's share, a 0-100 supply-security score on the lower diligence line (100 under 10%, 0 by 30%), and a per-supplier verdict — SECURED / EXPOSED / SINGLE-SOURCE RISK. A worsen-only survival gate forces SINGLE-SOURCE RISK only when a supplier is both material (≥18% share) AND uncontracted — the contract is the release valve, so a 30% supplier on a long-term contract is a managed risk while an 18% handshake is a single point of failure; a ready second source relaxes the line, and blank contract status degrades to a share-only threshold flag. Names the fix to make first and rolls the base up to SUPPLY SECURE / WATCH THE BASE / SUPPLY-SHOCK RISK. The matched pair to the Customer-Concentration Risk Gate — run both before diligence. Runnable Python engine + workbook + concentration-audit & supply de-risking playbooks + a 6-supplier sample. Grades a supply structure from your own numbers, never a person. Not financial advice.
Grade whether your plant's OEE is a real number or a story your reports tell. Computes each line's OEE as reported and as true, exposes the hidden factory, and forces INFLATED on any line whose availability quietly omits lost time. Deterministic, offline, your own shift facts.
Grade every critical process on whether its know-how is captured in a form someone else can run — or lives only in one head. Forces SINGLE-POINT on any critical, solo-held process with no verified SOP, and names the one to capture first before the person who holds it retires. Deterministic, offline.
Your CMMS shows 96% PM compliance right until the gearbox fails. This tripwire reads your export and catches the two lies inside compliance — deferral and pencil-whipping — forcing RUN-TO-FAILURE on any critical asset that's been double-deferred or signed off in less time than the work could take. Deterministic, offline.
Scrap, rework, containment, warranty, and expedite live in five GL accounts nobody totals. This ledger sums all five per cell, projects the figure monthly and annually against your own revenue, and names the worst cell and the one stream driving it — with a concentration gate that flags a runaway stream a modest ratio hides. Deterministic, offline, your own numbers.
A storeroom's fill rate averages a $12 fuse with a 26-week-lead coupling that downs the line if it stocks out. This watchtower reads your storeroom export and gives two independent reads per spare — coverage and exposure severity — forcing SINGLE-FAILURE RISK on any crit-A spare with no supply, no substitute, and a lead past tolerable downtime. Deterministic, offline.
Most states send no renewal reminder, and a single missed annual report escalates quietly: late fee, then loss of good standing, then administrative dissolution — reinstatement runs 3-10x a timely filing, and some states release your business name. This watchtower reads every entity and license renewal from your own confirmed dates against a pinned evaluation date. Urgency is the headline — CURRENT / DUE SOON / PAST DUE — and severity rides alongside as a separate read (from a flat late fee up to dissolution or name loss). A dispositive good-standing dependency cascade forces a license to BLOCKED when its parent entity is past due, past grace, and dissolution-bound — because you can't renew a license under a dissolved entity: the shipped sample's Florida contractor license reads BLOCKED 149 days from its own renewal because the Florida LLC underneath it is 64 days past due. It ships no deadline database — every date and consequence is buyer-confirmed, so it stays right when a state changes the rules. Rolls the book up to ALL CURRENT / RENEWALS DUE / STANDING AT RISK and names the one obligation to fix first. Zero-dependency Python engine + workbook with a fixed eval-date cell + reading & monthly-run playbooks + a 5-obligation sample. Grades obligations, never a person; files nothing. Not legal advice.
Founders don't get trapped by one big thing — they get trapped by a plate full of small tasks they keep doing because it's faster themselves, and the business bottlenecks on them. Mark each task on six weighted signals 0/1/2 (core founder work and willing-to-transfer-the-decision are gates) for a 0-100 leverage score — OFF YOUR PLATE / DOCUMENT FIRST / KEEP — measuring what you'd reclaim by removing it. Two worsen-only trap gates each force KEEP and name which fired: the CORE-WORK trap (only you can carry it) and the PHANTOM-DELEGATION trap (you'd keep the decision, so it routes back) — the sample's Investor updates (53) and Refund approvals (72) both read KEEP. Rolls your whole plate up to LEVERAGED / STRETCHED / FOUNDER-BOUND, folding in the catch every framework names: reclaimed hours only count if you redeploy them. Runnable Python engine + workbook + leverage-audit & clean-handoff playbooks + a 6-task sample. The step before the hire; grades how your time is spent, never a person. Not financial, legal, or management advice.
A signed LOI isn't the finish line — about a third never close, half of small-business sales die in diligence, and most of the rest get re-traded on price. This is sell-side diligence run on yourself: mark six weighted dimensions a buyer's team stress-tests 0/1/2 (validated financials/QoE and owner independence are deal-killer gates) for a 0-100 survival score — DEAL-READY / GAPS TO CLOSE / WOULD NOT SURVIVE DILIGENCE. Two worsen-only structural killers each force WOULD NOT SURVIVE: PHANTOM-EARNINGS (books won't validate) and OWNER-IS-THE-BUSINESS (nothing survives your exit) — the polished agency sample scores 75 and still fails. Customer concentration is a heavily-weighted scored dimension, not a gate — it re-trades the price, it doesn't end the deal. Names the one thing to fix first; rolls a portfolio up to ALL DEAL-READY / GAPS TO CLOSE / WOULD NOT SURVIVE. Runnable Python engine + workbook + sell-side diligence & deal-killer fix playbooks + a 6-business sample. Run it 12-36 months before you sell, when the skeletons are cheapest to fix. Grades a deal, never a person. Not a business valuation, brokerage, or legal, tax, or investment advice.
Could the business keep running if this person vanished for a week? The bus-factor question — cheapest to ignore, most expensive to discover late: replacing a key employee runs 100-300% of salary and acquirers visibly discount single points of failure. Score each key person on six continuity dimensions 0/1/2 (relationships, undocumented knowledge, sole authority, trained backup, continuity protection, revenue share) for a 0-100 resilience score — COVERED / EXPOSED / SINGLE POINT OF FAILURE. A two-condition kill-chain gate refuses to cry wolf: it forces SINGLE POINT OF FAILURE only when someone is both load-bearing (holds irreplaceable relationships or knowledge) AND unbacked (no trained successor) — the precise bus-factor definition, key and irreplaceable. A brilliant person with a tested backup is covered; the gate releases the moment either half clears, so it always names the cheapest fix. Rolls everyone up to RESILIENT / KEY-PERSON GAPS / FRAGILE. Runnable Python engine + workbook + risk-audit & de-risking playbooks + a 6-person sample. Grades a continuity posture, never a person — not HR or legal advice, not grounds for any employment decision.
Is your founding equity structured to survive a co-founder leaving? Skipping vesting is the single most-regretted founder decision — the co-founder who walks at month 11 still holding a big unvested stake becomes a cap-table problem you pay for at every future round. Score each co-founder on six structure dimensions 0/1/2 (vesting schedule, one-year cliff, stake size, acceleration, 83(b) election, documentation) for a 0-100 equity-health score — SOUND / EXPOSED / DEAD-EQUITY RISK. A two-condition dead-equity gate fires only when someone holds a material stake AND has no vesting — the precise cap-table-nuking pattern; a large stake that's properly vesting is protected, and the gate releases the moment either half clears. Names the one structural fix first (never a 'right split') and rolls everyone up to CLEAN CAP TABLE / GAPS TO FIX / FRACTURED. Runnable Python engine + workbook + equity-health & de-risking playbooks + a 6-founder sample. Grades a structure, not a deal or a person — and routes every fix to counsel. Not legal, tax, or investment advice.
Cheap and well-loved isn't the same as worth renewing. Score every AI/SaaS vendor on reliability and whether the spend is justified — RENEW / RENEGOTIATE / DO NOT RENEW — with a reliability-floor gate that refuses to justify renewing a business-critical vendor you can't rely on, however cheap or well-adopted. Grades the relationship from your own data, not people. Deterministic, offline, runnable.
A deterministic first-pass triage for any contract or SOW someone has asked you to sign. Grades six clause-risk areas (limitation of liability, IP assignment, indemnification, auto-renewal & termination, payment terms, scope clarity) weighted to 100 and returns SIGN / NEGOTIATE / DO NOT SIGN per contract, with a portfolio rollup. A structural deal-breaker gate hard-stops the two clauses that carry existential exposure — uncapped liability or a full pre-existing-IP assignment — no matter how clean the rest reads. Names the one clause to redline first. Runnable Python engine + workbook + reviewer & redline-negotiation playbooks + a 7-contract sample. Grades the document's risk surface from your own reading, never your legal rights. Not legal advice — have counsel review before you sign.
Loom in → SOP + Claude skill + automation map out. The ABS Framework's doc method, productized.
Turn invoices, receipts, and forms into structured data with AI that flags low-confidence fields for review instead of guessing. A runnable extractor (zero deps, runs keyless in demo, exits non-zero on review), editable JSON schemas, a QA tracker, and a playbook. Flag, don't guess — no accuracy guarantee.
The deterministic layer that judges extracted or keyed document fields against fixed rules — format, range, required, allowed-value, and cross-field math (subtotal + tax = total) — and returns PASS / REVIEW / REJECT per record. A runnable Python engine + a workbook reproduce the same rules, with a hard-fail gate that rejects any record with a critical failure (a missing total, a math error) however high it scores — a record can't be 90% valid. Type-agnostic: swap the rule pack for POs, applications, claims, intake forms. The validation step the AI Document Extraction Kit hands off to. Validates your own documents, not people. Not legal advice.
The deterministic pre-share safety gate for documents. A pattern pack of fixed detectors scans a document's text for PII across three sensitivity tiers — SSNs, cards, bank routing, health markers, dates of birth, addresses, emails, phones, IPs — and returns CLEAR / REDACT FIRST / DO NOT SHARE per document, with a library rollup. An exposure gate blocks any document with an unhandled high-sensitivity finding however high its coverage — one open SSN is dispositive. It flags what to redact and what's still exposed; it does NOT redact, is not a certified tool, and does not score or rank people. Tunable patterns and tiers. Validates your own documents; not legal advice. You remain responsible for what you share.
The deterministic AP money-gate: match every invoice line against the purchase order (what you authorized) and the goods receipt (what you received), run seven fixed checks, and return APPROVE / HOLD / BLOCK PAYMENT per invoice, with an AP-run rollup. A payment-block gate stops any line that bills more than was received, references no PO, has no receipt, or is overcharged — regardless of the match score; one over-bill is dispositive. Four editable tolerances separate a real exposure from rounding. It gates the pay decision; it moves no money, doesn't score people, and is not accounting advice. The general-AP three-way match — not a manufacturing receiving check, not a month-end close. Runnable engine + workbook. Not legal advice.
The deterministic front door for document intake: a priority-ordered routing rule pack matches each incoming document on fixed signals (filename, keyword, sender, amount, date), assigns a type + destination first-match-wins, and returns ROUTE / QUARANTINE / UNMATCHED, with a batch rollup. An ambiguity gate quarantines any document two rules send to two different destinations — regardless of priority — so a signed contract is never silently first-match-won into AP. A confidence floor quarantines a too-weak single match. QUARANTINE (conflict) is distinct from UNMATCHED (no rule). No model, no probabilities — rules you control. It decides the destination; it moves nothing and doesn't score people. Runnable engine + workbook. Not legal advice.
The deterministic “do we have everything to start?” gate. Define a requirements checklist for an intake type (new-client packet, loan file, vendor onboarding, patient packet) with required/optional + blocker flags and an acceptance rule (present / non-expired / matches-format). Each case gets a per-item status (OK / MISSING / EXPIRED / INVALID) and a verdict — COMPLETE / CHASE / HOLD INTAKE — with a blocker gate that holds a case missing a non-negotiable (a signed agreement, a consent, a valid ID) regardless of the completeness percent; 83% with an expired ID is still HOLD INTAKE. CHASE = required paperwork outstanding but blockers OK (safe to start prep). Offline workbook + runnable engine; reads a packet you enter, not your systems. Doesn't score or rank people. Not legal advice.
The pre-send check for a mail-merge. Point it at a template and a data list; it returns READY / FIX / BLOCK per row and MERGE READY / FIX ROWS / HOLD per template, with a structural-dominant gate — an unmapped placeholder or a malformed token forces HOLD no matter how many rows are clean, because it breaks every row identically. Each row is graded on its own fields (a missing required value is BLOCK, a malformed email is FIX); nothing is ever filled or guessed. Catches the literal {{first_name}} before it sends. Runnable Python engine + a workbook that reproduces it + two playbooks + a worked sample. It's the check, not the sender — your ESP/Word/GHL still sends. Offline, deterministic. Not legal, compliance, or deliverability advice.
The front door to the document-ops line: run it first to learn which fix to buy. Score your workflow across six stages — intake, classify, extract, reconcile, merge, govern — from your own 0–5 marks; it takes the worst stage as the headline (MIN-as-headline — a pipeline is only as automated as its weakest stage), returns AUTOMATED / MANUAL DRAG / STALLED, and routes the constraint stage to the exact RedHub drop that fixes it. A hard stall gate forces STALLED if any stage falls to 40 or below regardless of the rest; the mean is context only. Ties break to the earliest stage. Runnable Python engine + a workbook that reproduces it + two playbooks + a worked sample. Deterministic and offline — your inputs, no AI scoring, no lift multiplier. A gap analysis; you run the pipeline. Not legal or accounting advice.
Faster monthly reporting built around a workbook that checks its own math — enter the figures and every tie-out is labeled Ties out, Review, or Does not reconcile, with one close status. Self-checking reconciliation workbook, close SOP, finance prompts. You keep the numbers; not accounting advice.
Grade an existing SOP step by step on whether an autonomous agent could actually run it unattended. Mark six 0/1/2 signals per step (unambiguous action, named tool/system, explicit decision branches, defined success criterion, defined stop/escalation path, no undefined references) weighted to 100 and get per-step READY / GAP / BLOCKER plus an SOP verdict: RUNNABLE / TIGHTEN / NOT EXECUTABLE, with a library rollup. A keystone gate forces NOT EXECUTABLE whenever one step has no checkable success criterion or no stop/escalation path — the two things a human improvises around and an agent can't — so an SOP can score 96/100 and still be held. Names the one step to fix first. Runnable Python engine + workbook + step-review & fix-the-step playbooks + a 6-step worked SOP. The agent-era companion to the AI-First SOP Engine. A readiness aid that grades the document, never people, and never runs an agent; not legal, safety, or compliance advice.
Grade whether an exported CSV is structurally fit to be processed before any row-level tool touches it. Per-column CLEAN / DIRTY / MISSING plus one file verdict — READY / FIX / BLOCK — with a hard structural gate: a ragged row, mangled header, or missing required column blocks the file no matter how clean the cells that parsed look (the sample is all-CLEAN, quality 100, and still BLOCKs). Runnable Python engine + workbook. The upstream gate for the whole Document Processing line; grades the file, not people.
Reconcile 3+ sources on one key — a bank export, a processor export, and your ledger; or three feeds that must agree. Classifies every key AGREED / DISPUTED / INCOMPLETE (separating a value conflict from a coverage gap, which a two-list tool can't) and rolls up to RECONCILED / FIX / NOT RECONCILED, with a one-disagreement gate: a single disputed key blocks the reconciled verdict however many keys agree. Runnable Python engine + workbook. Deterministic, offline; reconciles data values, not people.
A column-aware PII readiness check for CSV / CRM exports. Classifies each column by the PII it holds and whether it's handled to the minimum its class requires — HANDLED / MASK FIRST / EXPOSED per column, SAFE TO USE / REMEDIATE / HOLD for the dataset — with a regulated-exposed gate: one open regulated column (an SSN with no handling) holds the whole dataset, 83% handled notwithstanding. Column-grain, not a cell-by-cell text scan. Runnable Python engine + workbook. Deterministic, offline; flags columns, not people. Not legal advice.
Audit a spreadsheet's formula layer — not its numbers — for the faults that silently corrupt a model: error cells, broken links, hardcoded magic numbers, copy-paste inconsistencies, and volatile dependencies. Reads a real .xlsx and returns SOUND / REVIEW / BROKEN per formula cell and a model gate where one error cell breaks the whole model — the shipped sample scores 92 and is still BROKEN because one cell carries a #REF!. Runnable Python auditor + workbook. Deterministic, offline; audits formulas, not people. A clean total is not a sound model.
Diligence-grade SaaS metrics kit for founders raising in 2026. Burn Multiple, Rule of 40, LTV done right, AI gross margin defense. Sourced 2026 benchmarks.
The MCP-driven workflow that turns an email backlog into a finished morning in ~30 minutes. Four Claude Skills read your inbox via the Gmail MCP, triage every thread (Reply now / Delegate / Schedule / Archive), draft the replies that need you in your voice, and hand you one ordered plan — with a delegate gate that routes off your pile anything that matters but isn't yours. It reads and drafts only; it never sends or deletes. Bring your own Gmail MCP; not legal advice.
Your founder's chief-of-staff, on a cadence — the flagship of the Executive Suite. Four Claude Skills read your calendar, inbox, CRM, and metrics every Monday and deliver a one-page operating report to your own inbox: the three priorities, the at-risk flags, what slipped, and the one decision to make. It grades every area against the target and floor you set (On track / Watch / Off track) and rolls the week up to Steady / Needs attention / Off plan — with a floor gate that calls a crossed red line (runway below your minimum, churn above your maximum) OFF PLAN, however the week otherwise feels. It reads and synthesizes; you act. Operating hygiene, not financial advice.
The knowledge cockpit of the Executive Suite, beside the Operating Cadence Engine's metrics cockpit. Four Claude Skills read your own docs, notes, CRM, and calendar (plus optional live signals) and self-deliver one Monday briefing to your inbox. A deterministic engine grades every tracked thread DECIDE NOW / KEEP WATCHING / PARKED — with a decision-trigger gate that surfaces the quiet-looking threads that are genuinely due: blocked on you, or a deadline this week with the ball in your court. Reads and self-delivers to your own inbox only; acts on nothing. Not financial advice.
The advisor built to tell you no — the Executive Suite's anti-sycophancy product. Bring one hard decision (a pivot, price change, raise, or new market) and a five-member skeptical board each builds the strongest case against it and scores how well the case survives, for an honest verdict: Go / Go with conditions / Not yet / No-go. The anti-sycophancy core is a veto gate — any one lens scoring a fatal objection forces NO-GO no matter how high the conviction score. A stress-test, not a decision-maker; the call is yours. Not financial, legal, or investment advice.
Your AI CFO — the Executive Suite's cash inspector. Grades your cash position on runway and trajectory, not the bank balance: reads your recent monthly numbers, computes net burn, the current rate versus the trailing average, revenue coverage, and your real runway (cash divided by current burn), then returns a verdict — HEALTHY / TIGHT / AT RISK / CRITICAL — with the concrete fix. The anti-happy-ears core is two gates: a hard runway floor (under three months is CRITICAL regardless of trajectory) and a trajectory gate (accelerating burn bumps the verdict one level worse, because a static runway number is optimistic while burn is climbing). Read-only: it computes and delivers the read; you make the calls. Pairs with The Forecast Floor and feeds the OCE Monday report as the cash line. Not financial, accounting, investment, or tax advice.
The capstone of the Executive Suite — the orchestration layer that ties your executive systems (Operating Cadence Engine, Devil's-Advocate Board, Pipeline Commander, Cash-Flow Sentinel, or any executives you run) into one honest company status. Each system's native verdict maps to one common 0–3 severity; the company status is the maximum — the worst executive, not the average — with a compounding gate that escalates to CRITICAL when two or more domains are AT RISK or worse at once, because cross-functional problems compound. It names the binding constraint (ties broken toward the domain that kills fastest), the cross-functional throughline, and Monday's top item. Orchestration only: it reads each system's verdict and delivers the briefing; it never runs the systems, moves money, or takes an action. Works standalone or with the full Suite feeding it. Not financial, legal, or investment advice.
The membership that runs your Executive Suite on a rhythm. You own the five systems; this is the standing cadence that convenes them — a weekly board briefing, a monthly executive review, and a quarterly strategic reset — on a Command Deck that counts from the last time you actually ran each one. The keystone is the weekly board briefing: miss it once and you're DRIFTING, miss two full cycles and the rhythm has broken to OFF RHYTHM, no matter how current the monthly and quarterly are. Includes all five Executive Suite systems, the operating kit (engine, workbook, four Claude Skills, template, and two playbooks), and updates while subscribed. It schedules and reports; you run the company. Not financial, legal, or investment advice.
