Thesis drift monitoring is the discipline of testing, at fixed intervals, whether the reasoning that justified a position still matches what the evidence shows. It needs three components: a thesis written as falsifiable claims, indicators tied to each claim, and a review record that separates an updated thesis from a quietly abandoned one. For investors who underwrite positions on management's stated strategy, the most reliable indicators are public records, because the distance between what a company said it would do and what its filings show it did is observable without insider access. That gap is what Lacuna Index measures.
TL;DR
- Thesis drift is an unrecorded divergence between the assumptions that justified a position and the evidence now available. A documented revision is an update, not drift.
- Drift is distinct from model drift (statistical decay in a predictive model) and from ordinary variance in reported results.
- Monitor claim-level indicators: milestones, catalysts, risk exceptions and outliers, each with a threshold set before the position is taken.
- Review on a fixed cycle, and after trigger events. Each review should end in one of three recorded outcomes: hold, investigate, or reconsider.
- For company-narrative theses, compare management's stated commitments with delivery in filings, transcripts and proxy statements.
Table of Contents
- 1. What thesis drift is, and what it is not
- 2. Separating thesis drift from model drift and ordinary variance
- 3. Evidence and indicators to monitor
- 4. Building the monitoring process
- 5. Review cadence and what to document
- 6. From signal to decision: revise, investigate or reconsider
- 7. Where public-record delivery analysis fits
- 8. Our view
- FAQ
- Sources
1. What thesis drift is, and what it is not
An investment thesis is a reasoned statement of why a position should earn a return: the drivers, the timeline, the conditions under which it fails. Thesis drift occurs when the position is held for reasons that differ from the ones recorded at entry, and nobody has written that down. Three forms recur:
- Assumption drift. The evidence behind a stated driver has weakened (margin expansion, market share gains, a capital allocation commitment), but the position continues on inertia.
- Rationale substitution. The original driver fails, and a new justification (valuation support, a different catalyst) replaces it without a formal re-underwriting.
- Mandate drift. The holding or portfolio moves away from the stated philosophy or style. Public pension governance documents treat this as a review trigger: the North Dakota State Investment Board's governance process lists significant deviation from a manager's stated philosophy or style among the reasons a manager may come under review.
The operative distinction is documentation. A thesis revised with stated evidence, a date and an approver is an updated thesis. A thesis that has changed in substance while the file still reads as it did at entry has drifted.
2. Separating thesis drift from model drift and ordinary variance
The phrase "drift monitoring" also appears in machine learning operations, where it describes the decay of a predictive model as input data or target relationships change. The two concepts share a logic (a stored baseline compared against current observations) but monitor different objects.
| Dimension | Thesis drift | Model drift | Ordinary variance |
|---|---|---|---|
| Object monitored | The reasoning behind a position | A statistical model's inputs or outputs | Period-to-period results |
| Baseline | Written thesis, milestones, stated risks | Training data distribution | Expected range of outcomes |
| Signal | Evidence contradicting or bypassing a stated assumption | Shift in feature or prediction distributions | Deviation within the tolerance band |
| Response | Revise, investigate or reconsider | Retrain or recalibrate | Record and continue |
Ordinary variance is the hardest category to separate from real drift. A single quarter below guidance is usually variance. A pattern is different: the same commitment restated across several earnings calls with a moving deadline, a metric quietly replaced by a more flattering one, or a milestone dropped from the investor presentation without comment. The test is whether the miss touches a load-bearing assumption in the thesis, and whether the explanation offered is consistent with earlier disclosures.
This is the point at which a written thesis pays for itself. If the thesis names which assumptions carry the position, an analyst can classify an observation in minutes. If it does not, every disappointing result becomes an argument.
3. Evidence and indicators to monitor
Indicators fall into two families: portfolio-level controls and company-level evidence.
Portfolio-level controls
Institutional oversight guidance already prescribes much of this structure. The CFA Institute Research Foundation's 2026 Total Portfolio Approach guide recommends dashboards and scorecards that track progress toward goals, risk indicators and success measures for board and investment-committee oversight. The same logic applies at position level: each thesis needs its own scorecard.
The Office of the Comptroller of the Currency's Bulletin 2008-10, written for national bank fiduciary activities, adds two controls that transfer well. It recommends exception tracking for items such as assets outside approved or retention lists, conflicts of interest and concentrations, with follow-up and timely corrective action. It also expects investment reviews to include performance measurement and a process for handling performance outliers. The bulletin addresses bank fiduciaries, so it is a reference model rather than a rule for every investor, but its structure (define the exception, log it, assign follow-up, close it) maps directly onto thesis monitoring.
Company-level evidence
For a thesis that depends on management delivering stated commitments, the monitored evidence is the public record. Useful categories include:
- Commitment language in earnings calls. Quantified targets, deadlines and named initiatives, compared across successive quarters.
- Periodic filings. Segment disclosures, risk-factor changes, and any redefinition of reported metrics.
- Proxy statements. Whether incentive-plan metrics still correspond to the strategy the thesis relies on.
- Press releases and presentations. Initiatives that appear, vanish or are renamed.
- Unasked and unanswered questions. Topics that analysts have stopped raising, or that management consistently redirects.
The analytical unit is the aspiration-to-execution gap: the difference between what was promised and what the record shows was delivered, measured against the same commitment over time. Lacuna's published treatment of governance failure early warning signals in public filings and its governance red flags guide catalog patterns that commonly precede a widening gap.
4. Building the monitoring process
A workable process has five elements. None requires specialized software at the start; each requires discipline.

Step 1: Write the thesis as testable claims
Convert the narrative into three to seven claims, each with a metric, a time frame and a failure condition. "Management will improve returns" is not testable. "Segment operating margin reaches a stated level by a stated fiscal year, with capital expenditure held within a stated range" is. For a narrative-dependent thesis, add the management commitments themselves as claims, citing the filing or call in which each was made.
Step 2: Set thresholds before entry
Define in advance what constitutes a watch condition, an investigation trigger and a break. Thresholds set after a disappointment are rationalizations. Record whether each is based on a single observation or a persistence rule (for example, two consecutive reporting periods).
Step 3: Log observations against claims
Every new filing, call or event is logged against the claim it affects, with a source reference. Observations that fit no claim are themselves informative: they indicate either an incomplete thesis or a new driver that has entered the story.
Step 4: Track exceptions to closure
Borrowing from the OCC model, each exception carries an owner, a follow-up action and a closure date. Open exceptions are reported at every review, so unresolved items cannot age out of attention.
Step 5: Record the outcome
Each review ends in a recorded decision, with the evidence relied on. The record is what later allows a committee, an auditor or a reporter to distinguish a disciplined revision from drift.
The North Dakota State Investment Board's governance process describes an annual evaluation cycle that reviews investment policies and performance results and develops corrective-action plans. The pattern (evaluate, identify deviation, plan correction) is the same loop at institutional scale.
5. Review cadence and what to document
Cadence has two layers: scheduled and event-driven.
Scheduled. The OCC bulletin sets the floor for national bank fiduciaries: assets in each account must be reviewed at least once during every calendar year for appropriateness, individually and collectively, alongside an annual assessment of the portfolio as a whole. For active equity positions whose evidence arrives quarterly, a review aligned to each reporting cycle is a practical pattern, with a deeper re-underwriting annually.
Event-driven. Reviews should also follow material events: guidance withdrawal, management or auditor change, restatement, a major acquisition, a change in segment reporting, or a proxy contest. Each is a point at which assumptions are likely to move.
At every review, document:
- The thesis claims, with status (intact, weakened, broken, unverifiable).
- New evidence, with source and date.
- Changes in the company's own language or metric definitions.
- Open exceptions and their follow-up.
- The decision, the rationale, and the person accountable.
6. From signal to decision: revise, investigate or reconsider
Evidence does not carry its own instruction. A decision rule keeps the response proportionate.
| Observation | Reading | Response |
|---|---|---|
| Single miss on a non-load-bearing assumption, explanation consistent with prior disclosures | Ordinary variance | Record, continue |
| Miss on a load-bearing assumption, with a specific and verifiable explanation | Possible genuine change | Investigate: test the explanation against filings and peer data |
| Commitment restated with a new deadline, or metric redefined | Possible assumption drift | Investigate and widen the log to prior periods |
| Same pattern across two or more periods | Persistent drift signal | Revise the thesis formally or reconsider the position |
| Load-bearing assumption contradicted by audited disclosure | Thesis break | Reconsider the rationale, with documented decision |
| New driver supporting the position that is absent from the original thesis | Rationale substitution | Re-underwrite as a new thesis |
Two cautions apply to the rule. First, revising is legitimate only if the revision could have been wrong: a new thesis must carry its own failure conditions. Second, an investigation should be bounded by a date. An open-ended investigation functions as a decision to hold without saying so.
7. Where public-record delivery analysis fits
Portfolio dashboards tell a committee what happened to returns. They do not, by themselves, say whether the company is doing what it told investors it would do. For narrative-dependent theses, that second question is the leading one, because delivery gaps appear in the record before they appear in results.

Lacuna Index builds that comparison from public records only: earnings calls, SEC filings, press releases, proxy statements and public statements, with no insider access. Per-company analysis scores narrative against delivery, classifies companies into archetypes (earned, borrowed or undervalued) according to execution and valuation, and applies one methodology across all covered issuers, with governance logs for the scoring. A monitoring file can cite those scores as a dated observation against a thesis claim, and the logs allow an analyst to trace any score back to its evidence.
Sector benchmarks are free, and they give a cohort baseline against which a single issuer's gap can be read, for example in Financials, Health Care or Industrials. Company-level reports are available by subscription through the Cohort Pulse and Sector Sweep plans, and the per-company analysis covers individual issuers. The user guide explains how to read a report, and the methodology walkthrough sets out how scores are built.

For related reading on how the scoring supports review decisions, see the analyses of systematic accountability scoring for investors and the corporate promise gap as a governance risk.
8. Our view
Most thesis drift is not a failure of analysis. It is a failure of record. Positions outlive the reasons they were opened, and no one can say when the reasons changed.
We built Lacuna around the observation that the public record already contains the leading evidence. Management states commitments in calls, filings and proxy statements, and later disclosures either deliver on them or quietly stop mentioning them. Reading that sequence consistently, under one methodology and with every score traceable to a source, turns a vague sense that the story has changed into a dated, citable observation.
For institutional investors, governance advocates, proxy advisors and journalists, that matters in a concrete way. A monitoring file with an execution score, its evidence log and a sector benchmark beside it can be challenged, audited and compared across issuers. A file built on impressions cannot. We recommend anchoring every narrative-dependent thesis to the delivery record from the first day, and reviewing it each reporting cycle.
Glen
FAQ
Is thesis drift the same as style drift?
Style drift is one form of it. It concerns a manager or portfolio moving away from its stated philosophy, which governance documents such as the North Dakota State Investment Board's process treat as a review trigger. Thesis drift is broader: it also covers a single position whose justification has changed while its record has not.
How is thesis drift different from model drift?
Model drift is statistical decay in a predictive model as data changes, and it is corrected by retraining or recalibration. Thesis drift concerns the reasoning behind a position, and it is corrected by documented revision or exit. Both compare current observations with a stored baseline.
How often should a thesis be reviewed?
At least annually, which is the floor the OCC sets for fiduciary account reviews at national banks. Positions whose evidence arrives each quarter are better reviewed at each reporting cycle, with additional reviews after trigger events such as guidance withdrawal, auditor change or restatement.
Can a single bad quarter prove drift?
Rarely. A single miss on an assumption that does not carry the thesis is ordinary variance. Drift becomes a credible reading when the same commitment slips repeatedly, when metrics are redefined, or when an audited disclosure contradicts a load-bearing assumption.
Sources
- North Dakota State Investment Board, Section II: Governance Process
- OCC Bulletin 2008-10, Fiduciary Activities of National Banks: Annual Reviews of Fiduciary Accounts
- CFA Institute Research Foundation, The Total Portfolio Approach (TPA): A Practical Guide
