Methodology
The dashboard turns public news coverage into two instruments. A per-axis escalation index tracks how serious reported actions are over time, and a threat ledger checks whether explicit threats were followed by action. A local language model scores every article against a fixed rubric, and pure SQL aggregates the scores. No step is manual, and any change to the rubric bumps its version and triggers an explicit re-score of all articles.
A 0–10 ladder shared by diplomacy and force
Diplomacy occupies the low bands, posturing the middle, direct engagement the top. Each article receives one score per axis it covers.
| Score | Band | What counts |
|---|---|---|
| 0 | None | No Iran-related military or diplomatic action. The article is excluded from the index. |
| 1 | Routine diplomacy | Scheduled meetings, courtesy statements, no new commitments. |
| 2 | Diplomatic friction | Condemnations, protest notes, ambassador recalls, threatened sanctions. |
| 3 | Significant diplomatic act | Sanctions imposed, embassy level changed, formal talks opened or broken down, IAEA board action. |
| 4 | Military posturing | Large exercises targeting Iran, carrier-group deployments, official threats of force, air-defense activation. |
| 5 | Indirect / proxy engagement | Proxy strikes attributed to Iran or its allies, attributed cyberattacks, attacks on Iranian-linked shipping. |
| 6 | Limited direct action | A single strike on Iranian territory or assets, shootdowns of aircraft or drones, attacks on Iranian vessels. |
| 7 | Retaliation cycle | Reciprocal strikes begin. Missiles or drones are fired directly at an adversary. |
| 8 | Sustained campaign | Days-long state-on-state exchange, multiple fronts opened simultaneously, mass force deployments. |
| 9 | Major escalation | Strike on a nuclear facility, formal declaration of war, mass-casualty event in one incident. |
| 10 | Existential / great-power | Nuclear use or credible imminent use, direct combat entry by a great power, regional total war. |
Scoring rules
- The score is the highest action described as fact in the article. An article covering a strike and a diplomatic condemnation scores at the strike's band.
- Rumors and speculation score one band below, marked unverified. "Sources say Iran is preparing X" is not the same as "Iran did X".
- Symmetric trust means corroboration, not nationality. No outlet is trusted by default because of its country. Military claims sourced from any party's official channels are unverified until independently corroborated, whether the outlet sits in Tel Aviv, Tehran, or London.
Each claim carries a status. Confirmed means at least two independent sources, official confirmation by a party involved in its own action, or independent evidence such as imagery. Reported means a single credible source not yet corroborated. Unverified is the default for military claims relayed from official channels, anonymous sourcing, and explicit speculation.
Aggregation
index_axis(t) = median of article scores on that axis, published within [t − 7d, t]momentum(t) = index(t) − index(t−7d)corroboration(t) = mean of source weight × claim coefficient over the same window (0–1)source weights: independent = 1.0 · official relay = 0.5claim coefficients: confirmed = 1.0 · reported = 0.8 · unverified = 0.6The median, not the mean, is the headline number so that one outlet's spike cannot move it. It reflects what the field of coverage is reporting. Corroboration sits next to the index on purpose. A high median with low corroboration means the conflict is hot but claims are thin or one-sided, which is information rather than noise.
The index has no explicit de-escalation input. It falls by window mechanics. High-scoring articles age out of the trailing seven-day window and are replaced by lower-scoring coverage, or coverage thins. A day with zero in-scope articles produces a gap in the series rather than a zero, so an empty week is never mistaken for calm.
Threats and what counts as follow-through
A threat is an explicit statement by one actor directed at another, extracted from article text. A threat converts when a qualifying action on the same axis (a strike, deployment, proxy action, or cyber incident) is published after issuance and inside the measurement window.
- Maturity. A threat enters calibration only once its seven-day window has fully elapsed plus a two-day grace period for late-ingested articles, nine days after issuance in total.
- Windows. Calibration reports conversion at 24 hours, 72 hours, and seven days. The all-time column adds no time limit. It is the most lenient measure: on a busy axis any qualifying action after issuance counts, so it answers "did anything follow eventually" rather than "did this threat drive near-term action".
- Known limitation (work in progress). Matching happens at the axis level, not the actor pair. An action by another party on the same axis can count as a conversion for a given threat. A confirmed marker requires the action article to name the threatening or threatened party, which mitigates but does not remove this weakness.
Honest limits
- Scores, summaries, and the index are LLM-assisted analysis of public reporting produced by a local model. They are not official information from any government or organization.
- The escalation index is an analytical construct about reported coverage, not a forecast of future events.
- A high score means outlets report serious actions as fact or report. It does not independently verify them. The corroboration value shows how much independent support the claims carry.
Data sources and their permission status are listed on the Sources page. Publishers or subjects can request removal or correction of specific articles via Contact. Requests are logged and acted on within seven days.