forensic-accounting-redflags — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited forensic-accounting-redflags (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Role: defensive capital-protector gate. This screen runs BEFORE conviction, BEFORE IV. A fraudulent or severely-mismanaged business does not earn a conviction grade — it earns AVOID or TOO_HARD.
The binding frame (why this gate exists): accounting red flags are incentive-caused bias made visible ([[forensic-redflags-are-incentive-bias-made-visible]]). Every dimension below is a place where management's incentives to flatter the numbers leave a forensic fingerprint — Munger's [[incentive-caused-bias-distorts-judgment]] and Buffett's [[avoid-accounting-gimmicks]] are the binding anchors, and the nine quantitative tests are simply how that bias becomes measurable. The gate's authority therefore ladders up to the binding CIO, not down from academic canon. This is what makes Hard Rule 4 (CIO management-integrity supersedes a forensic CLEAN) load-bearing: the Munger/Buffett vault is the binding CIO; this skill supplies the forensic technique, not the verdict. The CIO's management-integrity and owner-orientation principles override any forensic "pass" that conflicts with them.
The SEBI-Satyam, SEBI-IL&FS-CRA, and Damodaran-forensic method-atomics live in the canon layer. Pull them before running the checklist:
set -a && source /Users/pw/invest/.env && set +a && /Users/pw/invest/.venv/bin/python \
/Users/pw/invest/data/scripts/32_consult_brain.py \
--corpus canon \
--company "<company name> cash flow PAT related party pledge auditor provisioning NBFC group entity insider trading forensic" \
--model <bank|general|...> \
--step circle \
--json-out extracted/grilling/<TICKER>_forensic.jsonCite only slugs that appear in the returned principles[].slug. If principles comes back empty or thin, flag it and stay conservative — abstain rather than fabricate.
(Use --corpus blended instead of --corpus canon when you also want the desk-synthesis connective atomics — [[forensic-redflags-are-incentive-bias-made-visible]], [[acquisition-accounting-is-a-shared-forensic-and-capital-allocation-redflag]], [[a-cheap-turnaround-framing-cannot-reverse-a-governance-flag]], [[a-forensic-flag-must-become-an-immutable-kill-criterion]] — surfaced alongside the binding layer. blended keeps the Munger/Buffett CIO primary and down-weights pitch-side voices.)
STEP 0 pulls the forensic technique from canon. STEP 0b pulls the binding authority — run the Munger/Buffett consult and confirm the four binding anchors are returned before you trust any verdict:
set -a && source /Users/pw/invest/.env && set +a && /Users/pw/invest/.venv/bin/python \
/Users/pw/invest/data/scripts/32_consult_brain.py \
--corpus binding \
--company "<company> incentive-caused bias accounting gimmicks operating cash flow faulty accounting management integrity" \
--model <bank|general|...> \
--step conviction \
--json-out extracted/grilling/<TICKER>_forensic_binding.jsonThe four binding anchors this gate ladders up to are [[focus-on-operating-cash-flow]], [[avoid-accounting-gimmicks]], [[incentive-caused-bias-distorts-judgment]], and [[do-not-rely-on-faulty-accounting-or-ratings]]. The CIO's read on management character supersedes the forensic checklist: a technically-CLEAN screen still yields CAUTION if the binding consult flags integrity. The forensic dimensions are how the bias shows up; these anchors are why the gate has authority.
Channel-check handoff (don't stay filing-desk-bound). The filing is what management chose to disclose; triangulate it ([[triangulation-of-banker-and-auditor-feedback]]). Where a flag turns on counterparty behaviour — lenders quietly cutting limits, an auditor's informal reputation, supplier or distributor whispers — hand the channel-check to the primary-research-sentiment skill and fold its read back into the dimension that raised the flag. A filing that looks CLEAN while bankers are de-risking is a CAUTION, not a pass.
Run every dimension. Mark each CLEAR / FLAG / HARD-FAIL and note the evidence or absence.
Canon: [[earnings-quality-gap-analysis]], [[accounting-malfeasance-red-flags-checklist]]
reliable signal of earnings inflation (aggressive accruals, revenue recognized before cash arrives, deferred cost capitalisation).
management change.
hiding spots for cash diversion that inflate PAT while OCF bleeds.
Canon: [[cash-balance-fabrication-test]], [[test-for-inflated-cash-balances-and-bank-confirmations]], [[auditor-confirmation-reliability-check]]
(a) interest income earned (implied yield on reported balance — an implausibly low yield signals fictitious cash that earns nothing); (b) auditor bank-confirmation notes in the audit report (Satyam lesson: confirmations can be fabricated if the auditor doesn't independently verify with the bank); (c) schedule of cash balances vs bank certificates in the annual report.
multiple banks; bank balances concentration in a single related-party bank.
Canon: [[related-party-loan-and-pledge-chain-analysis]], [[related-party-entity-mapping]], [[forensic-footnote-mining-for-hidden-assets-and-liabilities]]
loans given to / received from promoter-owned entities, purchases/sales between group companies, guarantees extended.
loans (circular funding = siphoning red flag).
and any "advances" that have been outstanding ≥1 year without repayment.
receivables; RPT volumes > 15% of revenue without clear arm's-length rationale.
Canon: [[promoter-share-pledge-tracing]], [[pledge-margin-trigger-risk-assessment]]
level); which lenders hold the pledge; any margin-call trigger history.
cascade can force a block sale that decimates the stock and the promoter's control.
below 40% (limited buffer before loss of control).
(defaults, court orders against promoter entities).
Canon: [[auditor-confirmation-reliability-check]], [[cross-verification-of-management-disclosures]]
risk; Satyam's PW signed off for years); auditor resignations mid-year (Indian law requires explanation — a resignation before completing the audit is a hard signal); qualifications or emphasis-of-matter paragraphs in the audit report; restatements requiring re-audit.
auditor downgraded from Big-4 to a small/unknown firm without explanation.
replacement was appointed for ≥30 days; SEBI/NFRA enforcement action against the auditor on this engagement.
Canon: [[provisioning-and-accounting-policy-scrutiny]], [[accounting-malfeasance-red-flags-checklist]], [[earnings-quality-gap-analysis]]
YoY for silent changes in depreciation life, provisioning rates, revenue recognition, or capitalisation of R&D / pre-operative expenses.
use 120+ DPD under regulatory forbearance); compare reported Gross NPA% with SEBI stress-test disclosures and peer NPA ratios at similar ticket sizes.
investments" used to prop operating profit are Damodaran's flagged red-flag patterns.
re-expense any capitalised R&D or pre-operative cost back through the P&L and ask — does reported profit survive? If margins collapse once the capitalisation is unwound, the "quality" was accrual-manufactured. Run this as a binary red-flag trigger (yes/no), not a valuation re-build — IV stays downstream and is voided entirely on a RED verdict.
margin × capital-turnover. A "high-ROC" business whose return is driven by an implausibly thin capital base (aggressive capitalisation, off-balance-sheet assets, related-party leverage) is showing manufactured quality, not a real franchise — flag the mismatch and hand the durability question to the moat-analysis skill (the ROIC-vs-WACC signature test there).
median with no credit-quality explanation; growing deferred revenue or contract liabilities that never convert to cash; reported margin does not survive the R&D re-expensing.
provisioning adequacy in emphasis-of-matter.
Canon: [[asset-liability-mismatch-detection]], [[stress-testing-liquidity-under-adverse-scenarios]], [[alignment-of-fund-inflow-timing-with-obligation-dates]]
debt-funded subsidiaries.
< 1yr) funding long-term loan assets (mortgage, infra) is the IL&FS pattern — a liquidity crisis can crystallise overnight even when solvency looks fine.
disclosed committed liquidity backstop; rising CP issuance in a rising-rate environment.
action on liquidity.
Canon: [[asset-liability-mismatch-detection]], [[provisioning-and-accounting-policy-scrutiny]]
Look for evergreening (fresh loans issued to a stressed borrower to keep an account "standard"), serial restructuring, large "standard restructured" or SMA-1/SMA-2 pools, and a divergence between RBI's Asset Quality Review (AQR) GNPA and the reported GNPA.
peer median; recurring "technical write-offs" that flatter GNPA without real recovery.
borrower kept standard through related-party refinancing.
verdict on whether the book is honestly stated belongs to the bank-valuation skill plus SEBI/RBI canon — NOT to pitch-side perspective voices commenting on the lender. This dimension raises the flag; bank-valuation quantifies it (cost-of-funds, credit-cost normalisation, Ke vs ROA). Carry the flag forward; do not resolve it here.
Canon: [[group-entity-interdependency-analysis]], [[complexity-scorecard-financial-statements]], [[forensic-footnote-mining-for-hidden-assets-and-liabilities]]
extended to group entities, off-balance-sheet vehicles, and intra-group eliminations that are material.
cannot be understood standalone (its cash flows depend on opaque inter-company flows) is TOO_HARD or AVOID regardless of headline metrics.
([[acquisition-accounting-is-a-shared-forensic-and-capital-allocation-redflag]]): serial acquisitions, large goodwill that is never impaired, earn-outs that mask organic decline, and consolidation that buries acquired-entity weakness are forensic flags here AND capital-allocation flags. Raise the forensic flag and hand the quality of the deal-making to the capital-allocation-judge skill — the two desks share this signal.
eliminates ≥20% of gross revenue; auditor's report on subsidiaries includes qualifications not visible in the standalone.
subsidiary that is a significant borrower is unaudited or has a different (smaller) auditor.
Canon: [[insider-trading-sale-pattern-analysis]], [[insider-trading-pattern-detection]], [[promoter-share-pledge-tracing]]
and key-managerial-personnel (KMP) transactions.
results, bullish analyst commentary) that predated a negative event — the Satyam pattern of insider liquidation while concealing fraud.
connected entity, even in a different company, is a character signal.
statements; block deal at a significant discount to market; SEBI investigation (even pending, not convicted).
in any entity.
After running all nine dimensions, tally the signals and issue one verdict:
| Verdict | Criteria | Effect on pipeline |
|---|---|---|
| CLEAN | 0 HARD-FAILs, ≤2 FLAGs, no flags in dimensions 2, 3, or 4 | No cap; proceed to conviction normally |
| CAUTION | 0 HARD-FAILs but ≥3 FLAGs, OR any FLAG in dim 2/3/4 | Cap conviction at 2; requires explicit disclosure in dossier; WATCH/AVOID bias |
| RED | Any HARD-FAIL in any dimension | Hard-fail the thesis → AVOID or TOO_HARD; do not compute IV; note the specific trigger |
State the verdict, the dimension(s) that drove it, and the specific evidence (filing, year, source).
Post-fraud disposition (RED pathway): once fraud is established (not merely suspected), the intrinsic value is voided, not discounted ([[intrinsic-value-adjustment-rejection-in-fraudulent-context]]). Do NOT net any residual book value, "salvageable" segment, or sum-of-parts floor back into the thesis — fraudulent financials cannot be partially trusted, so there is no reliable base to discount from. The name is AVOID/TOO_HARD with conviction 1, full stop.
Return a structured block to embed in the dossier's ## Governance / Forensic Screen section:
GOVERNANCE-RISK VERDICT: <CLEAN | CAUTION | RED>
Dimensions assessed: 9/9
Flags: [list dimension numbers that FLAGged]
Hard-fails: [list dimension numbers that HARD-FAILed, or NONE]
Key evidence: [1-3 sentences on the most material finding]
Conviction cap: [NONE | 2 (CAUTION) | HARD-FAIL (RED)]
Pipeline effect: [no cap / cap at conv-2 / hard-fail → AVOID|TOO_HARD]
Canon cited: [slug list from the returned principles]This screen runs at Stage 1 (dossier), dimension 2 of the dossier template, BEFORE the conviction consult at Stage 3. The output is a frozen input — it does not get re-run at re-ranking unless a new governance event occurs (restatement, SEBI action, pledge surge).
no IV computed. The one-line reason goes into the dossier kill_criteria field.
which by the MoS thresholds (conv2 ≥40%) means the price must be ≥40% below IV-base to buy. Effectively, CAUTION names rarely clear the bar — that is intentional.
alone (the rubric in stock-onboarding-pipeline KEY REGISTRY).
if the vault flags character issues and the forensic screen is technically CLEAN, CAUTION applies.
prerequisite to proceed.
mandatory; do not skip them citing "not applicable."
new material governance event has occurred since the original dossier date.
The 30 perspective voices are pitch-side/promotional (down-weighted 0.85) and frame balance-sheet stress as upside optionality; they may add context but they may not argue around a flag ([[a-cheap-turnaround-framing-cannot-reverse-a-governance-flag]]). For a forever-hold mandate, a balance-sheet-stressed special situation is structurally disqualified ([[special-situation-vs-compounder-distinction]]; [[do-not-rely-on-faulty-accounting-or-ratings]]). Only the binding Munger/Buffett CIO arbitrates the verdict.
kill_criteria atdiscovery time and is immutable** ([[a-forensic-flag-must-become-an-immutable-kill-criterion]]; [[decision-journal-method]]). It cannot be re-litigated or softened by a later bullish thesis — face the reality on the record rather than rationalising it away ([[avoid-denial-and-face-reality]]).
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.