The anatomy of one investment decision, from approval to LP verification.
Most platforms describe what they do. This page shows what one decision looks like as it moves through the Capital Refinery engine — from a buy-side IC approval through 18 months of monitoring, through a forced reopen of the original thesis, through outcome lock, through LP verification. T+0 is anchored on the real Falcon Services Q1 2026 proof case (healthy, APPROVE WITH CONDITIONS); the stages that follow are a modeled lifecycle showing how the engine would carry a position through drift, reopen, and outcome — not events that have happened to Falcon. Every primitive visible. Every artifact reproducible from the same engine.
One decision, eighteen months, on the record.
From the sealed IC anchor through an event-triggered reopen to an LP-verifiable outcome — every stage is a live product primitive, not a slide. The anchor is a cryptographic snapshot; the outcome at the end locks against it.
- T+0Investment Decision LedgerAnchoredIC anchor sealed at approval · decision typed acquirebasis sealed · 0d8c…9e8e
- T+~30KPI Lineage · Why-This-NumberMonitorEvery figure traced to method, confidence, audit-id
- T+~90Continuous IC MemoMonitorDrift against the entry assumptions surfaced
- T+~180Risk Signals · firm policyAdvisoryAn observed value crosses a firm threshold
- T+~270Scenario Propagation EnginePre-actionShock pack runs; covenant headroom thins
- T+360Decision Fragility · InvalidationReopenA registered entry tripwire trips — the thesis is reopenedTripwire: the 15-yr anchor-offtake assumption · crossed when the hyperscaler invokes its capacity-flex clause and defers Phase-2 power draw 9 months · auto-reopened — an event, not a calendar review
- T+540Decision Lock · Decision TimelineLockedThe outcome locks against the original anchorintervene: negotiated a take-or-pay floor + interconnect-milestone credit · locked against 0d8c…9e8e · IRR re-underwritten, hold extended — the delta between entry basis and outcome is itself the artifact
- T+540+Outside the firm boundary/p/<token> public verificationVerifiedThe LP resolves the token and verifies independentlyoutside the firm boundary · no login, no SaaS dependency
The reopen fires because a named entry assumption was registered as a tripwire at T+0 and an external event crossed it — not because a quarter elapsed. The outcome is scored against the original cryptographic anchor, and the LP verifies the final record on the public rail, outside the firm’s walls.
Why this page exists
Decision quality is not decision outcome. A good decision under uncertainty can produce a bad outcome; a bad decision can produce a good outcome. Sophisticated investors — Annie Duke in Thinking in Bets, Michael Mauboussin in The Success Equation, Daniel Kahneman in Noise — separate the two explicitly. The infrastructure for telling them apart over time is what this page shows.
The walkthrough below traces one buy-side decision through the engine. T+0 is the IC approval; T+540 is the LP review eighteen months later. Every stage shows which primitive is active, what artifact is produced, and how the decision basis stays inspectable as conditions change. The reader will see how the original IC anchor remains intact through 18 months of evolution — and how a new decision-commit at T+360 doesn’t overwrite the historical record but extends it.
The eight stages
One decision, eight stages, eighteen months. The order is fixed; the artifacts are reproducible.
Falcon Services Q1 2026 closes with an APPROVE WITH CONDITIONS recommendation. The pillars, conditions, signals, KPIs, covenants, and downside scenarios are all structured fields the engine will re-evaluate against every operator update. The decision is typed (acquire / monitor / intervene / hold / exit). The IC anchor — a cryptographic snapshot of the decision basis — is committed the moment the committee approves.
Operator data starts flowing — accounting feeds, KPIs, covenant tests. Every KPI carries candidate_id, source_ref, and method. A reader clicks a number and sees the source document and the reasoning tree. Per-cell provenance is the structural alternative to 'the CIM said.'
Operator submits Q2 financials. The engine re-evaluates every IC-approved assumption against the new data. Drift is detected on margin trajectory (300 bps below underwriting) and customer concentration (top-1 customer increases share by 4 pts). The 'Since last IC' panel surfaces the diffs in human-readable language — derived from the anchored snapshot, not reconstructed by the analyst the night before the quarterly.
The scoreboard renders observed-vs-firm-policy across portfolio LTV, customer concentration, expense creep, debt-service runway, and lease at-risk rent. Falcon's customer concentration crosses the firm policy threshold (top-1 above 12%). Severity: advisory. The engine flags but does not escalate — the gradient is observable not yet decisionable.
A macro shock pack — interest rate jump, demand contraction in a specific sector channel — runs against the live position. The engine returns break order, headroom loss, and time-to-consequence on each covenant. Falcon's stress posture shifts from 'watch' to 'pre-action' based on the propagation result. The fixture is exactly what runs against customer deals.
The engine signals that the original IC thesis is no longer defensible at the current decision posture. The Continuous IC Memo aggregates the drift accumulated over 12 months. The original anchor remains intact (the historical record); a new committee convenes to type a new decision (intervene / hold / exit). The new decision-commit creates its own anchor.
Falcon completes the intervention path the committee approved at T+360. The outcome — measurable, recorded, traceable — locks against the decision anchor. The Decision Timeline shows the full chain: original IC → drift evidence → reopen → new decision → execution → outcome. Each row is a decision-commit anchor with sanitized metadata.
The LP requests a portfolio review. The GP issues a share token for Falcon's position. The LP resolves the URL at /p/<token>, sees an LP-sanitized snapshot of the decision timeline (operator metadata stripped), and verifies independently that the artifact matches the engine state. Decision history is inspectable, not reconstructed from emails.
What this page is structurally arguing
- Decisions don't drift silently. Every assumption is structured; every assumption is re-evaluated against every operator update; drift is surfaced explicitly in the Since last IC panel.
- Decisions don't disappear into notes. The IC anchor at T+0 remains intact through 18 months of evolution. The Decision Timeline shows the full chain — original IC, drift evidence, reopen, new decision, execution, outcome — as a structured record, not a narrative reconstruction.
- Decision quality is decoupled from decision outcome. Whether the outcome at T+540 was good or bad, the artifact at T+0 documents what the team knew, what they tested against, and what they approved — the basis for the call as it stood at the moment of commitment.
- Provenance is per-cell, not per-document. 'The CIM said' is insufficient. Every figure traces to a page, a method, and an adjudication trail. The Why-this-number overlay exposes the reasoning tree on click.
- Output inherits context. The IC memo at T+0, the covenant forecast at T+180, the new IC memo at T+360, the LP verification page at T+540 — all rendered from the same engine, against the same anchored record, with no manual rebuild.
What this kills (in operational terms)
Three operational failures the architecture structurally eliminates:
- The 'where did we land on this?' email chain. Decision state lives in the structured record, not in inbox archaeology. Reopening a position pulls the anchor and the timeline, not a quarter's worth of forwarded messages.
- The 'this was approved on a different model' problem. Output is rendered from the live record at issuance. The board deck the team produces in Q3 inherits the same context as the IC memo it approved against in Q1. Approved-on-a-spreadsheet-someone-lost is no longer a structural condition.
- The 'what did we know at the time?' LP question. The IC anchor at T+0 is a cryptographic snapshot. When the LP at T+540 asks what the team knew when they approved, the answer is the anchored record — not the analyst's memory.
The same engine, from the other readers’ seats
The anatomy above traces the decision from the buy-side investment team’s seat. The same engine renders the artifact for each adjacent reader: the seller’s IRA reads the same axes the buy-side IC reads; the lender’s covenant forecast reads against the same engine’s stress propagation; the LP’s public verification page resolves the same fingerprint that anchored the IC’s commit. See same engine for the unifying architecture, or Cedarbrook for the seller-side walkthrough of the same engine on a corp.distribution business.
Sources cited
- Annie Duke — Thinking in Bets (2018); decision-making under uncertainty, separating decision quality from decision outcome
- Michael Mauboussin — The Success Equation (2012) and Counterpoint Global research; luck vs skill in investment decisions
- Daniel Kahneman — Noise (with Sibony & Sunstein, 2021); decision discipline under variability; foundation for the engine's refusal of imputation
- Philip Tetlock — Superforecasting (with Gardner, 2015); expert judgment quality and forecasting discipline
- Lovallo & Sibony — McKinsey Quarterly 'The Case for Behavioral Strategy' (2010); HBR 'Are You Solving the Right Problem?' (2017); decision quality in institutional settings
- Atul Gawande — The Checklist Manifesto (2009); decision discipline as structural primitive, not personal habit
- ILPA Principles — Institutional Limited Partners Association governance and reporting standards → https://ilpa.org/principles/
- SEC Private Fund Adviser Rules (2023 final rule, subsequently vacated by the 5th Circuit in 2024). Not in force; the conceptual framework remains useful as a reference point alongside ILPA Principles, which are the active governance counterpart.
See the engine running on a deal you bring us.
The eight stages above are reproducible. Bring a real position — a deal you're underwriting, a deal you closed last quarter, a portfolio company under watch — and we'll show you the artifact at every stage. No demo data.