METHODOLOGY

Last updated 2026-04-25

Methodology

Public-facing description of the Quantum Intelligence System — how sources are chosen, how events are scored, what’s in scope, what’s not, and how corrections work.

This page is load-bearing for credibility with the investor, researcher, and CISO audiences. It is updated whenever the underlying approach changes; the change history is the top of this file.

Current spec version: v2.3 Methodology last updated: 2026-04-23

Change history

2026-04-23 (v2.3) — Updated after second-pass forensic audit of the v2.2 repository. Material changes: insider_pattern_update is now a distinct event type separate from correction (cluster growth doesn’t pollute the corrections trail or force daily newsletter publication); the weekly roll-up filename rule is day-of-week robust and weekly catch-up now works after a failed Monday; form4_context schema unified; Pattern A at base score 5 or 6 with catalyst-adjacent escalator now gets targeted EDGAR Form-4 corroboration (the narrow case most likely to produce exercise-and-sell false positives); schema_version string format bug fixed. See spec/v2.3.md change log for the full list.
2026-04-23 (v2.2) — Initial remediation after first-pass forensic audit of the v2.1 bootstrap. Material changes: /corrections/ is now materialized per day; the insider-signal pattern detector has proper cross-day deduplication; feedback no longer nudges scoring mid-run; the EODHD scope (purchase and sale codes only) is now documented honestly.
2026-04-22 (v2.1) — Initial public methodology.

Mission

Produce high-signal, neutrally-reported daily intelligence about the global Quantum 2.0 ecosystem — quantum computing (hardware and software), quantum networking, quantum sensing, and post-quantum cryptography.

Two deliverables:

A structured event dataset — every development worth recording, captured as a JSON event with stable identifiers, source citations, and a calibrated importance score. The dataset is the asset.
A newsletter — a human-readable rendering of the day’s (or week’s) dataset when there is enough signal to warrant publication.

The newsletter is the interface; the dataset is the asset. Everything claimed in a newsletter traces to a source in a structured event.

Audience

The system is designed for three overlapping reader types:

Investors — capital allocation decisions. They care about funding flow, competitive positioning, and technical milestones with enterprise-value implications.
Researchers — the state of the art across modalities. They care about benchmarks, talent movement, and architectural progress.
CISOs — post-quantum cryptography timelines. They care about NIST standards, enterprise adoption, policy signals, and vendor readiness.

The technical floor is university-level physics and CS. Basics (qubits, superposition, gate fidelity) are not redefined. Less-common terms (e.g. magic state distillation, NV centers, lattice-based KEM) are briefly contextualized when first used.

Cadence

Daily: JSON ingestion and classification runs every day. A daily newsletter publishes only when at least one event clears an importance score of 6. On quieter days, the JSON is still written; the day is simply not published.
Weekly: A roll-up newsletter publishes every Monday, regardless of the week’s volume. Quiet weeks are acknowledged explicitly rather than padded.

The conditional-daily approach is deliberate. Training readers to ignore a feed is a worse failure than missing a daily slot.

Sources

The system draws from the open web plus one structured-data provider (EODHD). Sources are tiered by authority.

Tier 1A — Structured authoritative (public companies)

EODHD — authoritative source for earnings, fundamentals, corporate actions, insider transactions, and tracked-ticker news on public quantum names.
SEC EDGAR (US), LSE RNS (UK), ASX announcements (Australia), and equivalent regulator feeds for non-US names.

Any event scored at 7 or above involving a public company is cross-verified against the primary regulator filing, not only the EODHD structured entry.

Tier 1B — Primary (all entities)

Company investor-relations pages and press releases.
arXiv quant-ph daily listings (filtered — see below).
NIST, ETSI, IETF standards and drafts.
Government program pages: DOE, DARPA, NSF, EU Quantum Flagship, UK NQCC, Japan Q-LEAP, Australia Quantum Strategy.
Major trade publications: The Quantum Insider, Global Quantum Intelligence (Inside Quantum Technology), Physics World quantum beat, IEEE Spectrum quantum beat.

Tier 2 (queried on signal)

USPTO and EPO for notable patent grants.
LinkedIn for senior-hire signal confirmation (public profiles only).
University and national-lab news rooms.

Tier 3 (background / trend confirmation)

Industry analyst commentary (Hyperion, QED-C public reports).
Conference programs (Q2B, IEEE QCE, APS March Meeting quantum sessions).
Peer-reviewed journals (Nature, Science, PRX Quantum) for confirming earlier arXiv signals.

Non-English coverage

English is primary. Non-English sources are included only when translation is reliable and the source is the primary signal.

Mandarin Chinese — USTC announcements, Origin Quantum, SpinQ, and Xinhua English editions when available. Machine-translated sources are flagged with medium source confidence.
Russian — Rosatom quantum program when translatable.
Japanese, German, French, Dutch, Korean — most organizations publish English releases; native-language sources are used only when they contain material not in the English version.

Coverage is partial for Chinese and Russian and occasionally for other languages. This limitation is disclosed rather than hidden.

Importance scoring

Every event receives an importance_score from 0 to 10, calibrated against an anchor table of past events. Summary:

Score	Meaning
9–10	Paradigm-shifting — redraws the competitive landscape or fault-tolerance timeline.
7–8	Material — concrete advance, significant capital, major hire, or consequential policy.
5–6	Meaningful but incremental.
3–4	Minor; usually excluded from the newsletter.
0–2	Noise; should have been filtered at ingestion.

The newsletter publication threshold is 6. The full anchor table lives in the repository at /skills/quantum-intel/scoring-rubric.md.

Scoring is applied against the same anchors on every run. Operator feedback in /meta/feedback.jsonl does not silently alter scoring within any single run — the rubric is the authority for every individual scoring decision. Every fourth Monday roll-up includes a calibration note summarizing feedback patterns over the preceding four weeks; if a consistent bias is observed, the note proposes a specific edit to the rubric, and the operator applies any edit manually after review. The system does not self-modify the rubric.

Academic paper inclusion

arXiv quant-ph posts roughly 40 papers per working day. An industry newsletter cannot and should not carry all of them. A paper is logged as an event only if it meets at least one of:

(a) The first author is affiliated with a commercial quantum company or a tracked national lab.
(b) The paper reports a quantitative benchmark improvement on an established, named metric — two-qubit gate fidelity record, logical-qubit lifetime, quantum volume, boson-sampling scale, sensor sensitivity, code distance, T1/T2 coherence times, etc.
(c) The paper has been independently covered by at least two Tier-1 trade publications.

This is deliberately restrictive. Research coverage is filtered for industry relevance, not academic completeness.

Insider transactions

Insider transactions on tracked public quantum names are streamed to an append-only signal log. Individual transactions are not published; they are raw data.

A pattern-detection pass promotes transactions to first-class events only when they match one of four patterns plus an optional escalator:

Cluster sell — three or more distinct insiders sell within 14 days, materially above the 90-day baseline.
Cluster buy — two or more distinct insiders make discretionary purchases within 14 days.
Anomalous individual transaction — a single transaction at least 3× the insider’s 12-month average (only for insiders with ≥ 6 months of baseline), or representing at least 25% of the insider’s holdings.
10b5-1 plan change — plan adoption, modification, or termination by a named officer.

Plus a catalyst-adjacent escalator (+1 to the base pattern score, capped at 8) when a pattern falls within the 10 trading days preceding a known catalyst.

Known limitations. The primary structured-data source for US insider transactions (EODHD) exposes only SEC Form 4 transaction codes P (purchase) and S (sale). Other codes — including M (option exercise) — are not available through that feed. The detector mitigates this in two ways:

Any candidate pattern scoring ≥ 7 triggers a fetch of the original Form 4 from SEC EDGAR so exercise-and-sell sequences can be excluded from cluster-sell aggregation.
Any Pattern A (cluster sell) at base score 5 or 6 with the catalyst-adjacent escalator applied also triggers the Form 4 fetch. This is the narrow case most likely to produce exercise-and-sell false positives — multiple executives mechanically exercising options and selling shares in a pre-earnings window. Targeted corroboration prevents these from landing in the newsletter as “cluster sells.”

For other sub-7 patterns, the limitation is noted in the event’s why_it_matters rather than corrected at source.

Pattern growth over the 14-day window. When a cluster grows (e.g. a fourth insider joins within the open window), the system emits an insider_pattern_update event — a distinct event type from correction. Updates do NOT trigger daily newsletter publication and do NOT appear in the corrections trail. They surface in the weekly roll-up’s Insider Signals section as status updates on the original open pattern. This preserves the credibility meaning of “correction” (factual fixes only) while still tracking how patterns evolve.

Insider activity is a lagging, noisy indicator. The newsletter always frames promoted patterns with explicit uncertainty. Full rules live at /skills/quantum-intel/insider-signals.md.

Non-US tracked tickers (LSE, ASX, etc.) are covered through the equivalent exchange director-dealings announcements; parse rules are documented in the insider-signals skill file.

Voice and attribution

Declarative sentences. Past tense for events, present tense for standing facts.
No first-person voice. No “we think” or “I believe.”
Claims are attributed to a source. Quantitative claims include units and context. Interpretation is flagged explicitly.
Paraphrase, never reproduce. Direct quotes are rare and short.
No marketing adjectives imported from press releases.

Corrections

When new information contradicts or materially updates a prior event, the original event is not edited. A new event is written with type correction and a reference to the superseded event. Corrections appear prominently in the next newsletter and are catalogued in /corrections/.

This preserves the audit trail. Over time the cumulative corrections log is itself a credibility signal — it shows the system is willing to be wrong and transparent about it.

Backfill corrections (distinct from `type: "correction"`)

Occasionally an event is added to a prior day’s daily JSON not because new information contradicts a prior factual claim, but because the original processing missed the event entirely (a Step-2 or Step-3 sweep gap). These are backfill corrections — additive completeness fixes — and are recorded here for transparency. They are kept distinct from type: "correction" events: the latter requires a prior factual claim to fix, the former does not. Backfill corrections do not appear in the cumulative corrections log below and do not use type: "correction" on the added event.

2026-04-25 — added 2026-03-26-ionq-scannell-board-election to daily/2026-03-26.json. The 8-K disclosing William F. Scannell’s election to the IonQ board was filed within Batch 1’s window but missed during initial processing; surfaced via retrospective trade-press coverage. The systematic gap is now closed by the EDGAR 8-K Item-number scan procedure documented in skills/quantum-intel/source-catalog.md.

Cumulative corrections log

Corrections are recorded here as they are issued. Each entry links back to the supersedes_event_id of the original event and summarizes what changed.

This list covers only factual type: "correction" events. insider_pattern_update events (pattern-growth updates — see spec §7.6) do NOT appear here; they are tracked in state.json’s open_insider_patterns and surfaced in weekly roll-ups. Backfill-correction additions (see subsection above) do NOT appear here either.

Known limitations

No direct interviews. The system aggregates from the public record. Information available only through primary interviews is out of scope.
No proprietary research. Outside of EODHD’s structured data, all sources are public.
Partial non-English coverage. See above.
No market-making guidance. The system reports public capital events but does not issue predictions about equities or trading recommendations.
Lag between events and inclusion. Typical lag is 0–24 hours; larger lags occur when non-English sources are primary, when EODHD data arrives late, or when corroboration is required before publishing.
Insider activity is noisy. Promoted patterns are framed with explicit uncertainty; readers should weight them accordingly.

Data and redistribution

Newsletters, the methodology, structured event data, and the skill package are licensed CC-BY 4.0.
Raw EODHD API responses are not redistributed. Only derived fields appear in event records. Consumers who need the underlying structured data should subscribe to EODHD directly.
Primary sources (press releases, filings, news, papers) retain their own copyright. Citations paraphrase and attribute; they do not reproduce verbatim.

Contact and feedback

Issues and pull requests against the repository are welcome. Feedback on coverage gaps, scoring, and factual errors can also be submitted by adding an entry to /meta/feedback.jsonl. The file is read at the start of every run and informs calibration over time.

Methodology published under CC-BY 4.0. Repository: https://github.com/jlass192/quantum-info