Disclosure Continuity
Bloomberg and FactSet give you the normalized financial series. Disclosure Continuity monitors whether the reporting path behind that number just changed — and tells you, and your AI, whether to ignore it, understand it, or read the filing.
A maintained, versioned intelligence layer over SEC disclosure — not a data dump and not a black-box flag. Every event is source-linked to the filing and reproducible as of any past date. The engine stays server-side: consumers receive answers and evidence, never the underlying lineage knowledge base.
The problem
A company's financial series can keep flowing smoothly while the SEC filing underneath it changes: a concept moves to a new XBRL tag, migrates across tag families, is replaced by a custom tag, drops out of structured data into prose, or genuinely ends. To a raw machine-readable feed all of these look identical — the data disappeared — yet they are economically very different. This is a data-quality problem before it is an investment problem, and it is exactly what trips up an AI research agent:
Raw filing: 2024 purchase obligations = $45bn · 2025 = NULL
LLM: “The company stopped reporting purchase obligations.” — often wrong.
Three outcomes, not twenty classifications
The engine keeps detailed, evidence-tiered classifications internally. The analyst — or the agent — sees three.
| Outcome | Meaning | Worked example |
|---|---|---|
| IGNORE | Reporting representation changed; continuity is fully explained. | SFBS — a $12.6bn NotesReceivableGross line left its XBRL tag; the same exposure continues under a successor financing-receivable concept. 71% of assets · no economic disappearance · no action. |
| UNDERSTAND | The disclosure moved, and we can say where. | NVIDIA — the tagged purchase-commitment series ended at $45.1bn; the commitment did not shrink, it grew to $122.2bn, disclosed in the notes with no XBRL tag. Track capex off the tagged number and you miss $77bn. |
| READ | Continuity cannot be established — a human must read the filing. | CURLF — a $1.75bn contractual-obligation disclosure disappeared; no structured successor confirmed, text verification incomplete. 61% of assets · high priority · read the filing. |
A READ is a workflow instruction — “this needs a human” — not an economic verdict. It does not, on its own, imply misconduct or a bearish signal.
What you avoided reading
Across one scan of all US filers, the value is the compression: surfacing the handful of changes that matter and suppressing the ones that don't.
| Stage | Count |
|---|---|
| Apparent reporting discontinuities (all US filers, one scan) | 16,774 |
| Durable, adjudicated continuity events | 201 |
| · resolved automatically as re-tags → IGNORE | 71 |
| · resolved as moved / located → UNDERSTAND | 78 |
| · could not be resolved → READ (analyst queue) | 52 |
| Changes the analyst did NOT have to review | 149 |
Engineering-verified this cycle: 203 candidates → 201 admitted to the durable ledger (2 rejected as unsupported re-tag claims — the system refusing to serve a conclusion it can't evidence). 100% of events source-linked at accession level and point-in-time reproducible. Buyer-facing adjudication metrics (Precision@Review, blind-discard rate, true new-lineage rate) are measured by human review and are deliberately not asserted until that review is complete.
Why the trust model is the product
| Property | What it means |
|---|---|
| Fail-closed | When evidence is insufficient the engine returns INDETERMINATE with a reason, never a manufactured PASS. For teams fighting hallucinated certainty, refusing to guess is the feature. |
| Diligence-ready | Every event carries prior tag/value → successor tag/value, both accessions, the lineage and text tests that ran, and the verification dates. |
| Point-in-time | ?as_of= returns exactly what the engine knew on any past date — the evidentiary property that makes results reproducible. |
| Maintained | A versioned, adjudicated dataset, not a script. The value compounds as accumulated lineage lets more changes resolve automatically. |
For AI research platforms
Disclosure Continuity sits between the filing and the LLM. Before your agent concludes “the company stopped reporting X,” it checks whether X was re-tagged, moved to prose, or genuinely ended — so a broken data series is never mistaken for a change in the business. Delivered as a REST API and a remote MCP tool: the model receives the question’s answer and its evidence, never your knowledge base. Access is per-tenant and entitlement-scoped; consumption is metered as a licence boundary, not sold as raw query volume.
How it is delivered
Evaluation is a paid pilot: $5–10k over 8–12 weeks, private API, defined coverage (100–250 issuers, recent window, full evidence packets, sample historical as_of), credited against a subsequent annual licence. No bulk history export during the trial.
Positioning. Keep Bloomberg for the numbers. Use Disclosure Continuity to know when the reporting structure underneath those numbers has changed — and whether it matters.
To arrange an evaluation on your own names, or to discuss an OEM route for a financial-AI platform, write to tnk@newwaycapital.com.
Further reading
Practices
Contact
- tnk@newwaycapital.com
- +41 76 381 47 48
- Geneva, Switzerland
Disclosure Continuity reports how companies report financial information through time. It is a data and intelligence service, not investment advice, and no result is a recommendation to buy or sell any security. Outcomes such as READ indicate that a filing warrants human review, not that any conclusion about a company has been reached.