When the West moves, how will Beijing answer?

Sixty years of doctrine (Mao to Xi) treated as a retrieval corpus. Given a global event — an export control, a Taiwan incident, a talent restriction — the predictor emits a structured ex-ante claim about the rhetorical frame, escalation tier, primary policy lever, and timing window of the PRC response, scored on every dimension once the response lands.

Forward Indicators· Brief No. III–01

One timeline. Four eras. The Xi era carries the modern weight.

Each anchor is dated, sourced, and tagged with its dominant rhetorical themes. The three highlighted in accent are the top-3 retrieved passages for the live May 2026 event — the audit trail for the prediction below.

What did the state actually do?

Nine episodes from the 2018–2026 window backfill the eval set. Each row records the realized response on all four dimensions. The §2.6 BIS-2022 case calibrates the predictor’s response-window expectations: the substantive cascade ran ten to twelve months past the trigger, much longer than the 2018 immediate-tariff template.

The first ex-ante scored claim.

Bloomberg’s 26 May 2026 report on the PRC expansion of AI-talent travel curbs to private firms is the launch event for Project III. §2.9 of the dossier treats the event itself as a same-day self-reliance / talent-restriction response. The forward call below is on the substantive cascade— the state-media codification and subsidy follow-on the predictor expects over the next eight months, on the §2.6 BIS-2022 lag template.

Persisted record: data/doctrine/predictions.jsonl · emitted by python -m forward_indicators_doctrine against prior doctrine v0.1.0.

Baselines are weak. That is the whole point.

The most-common-by-category and nearest-neighbor baselines clear two-thirds framing accuracy and ~0.44 tier MAE on a 1–5 ordinal — respectable for a 9-case n — but lever F1 is at floor across the closed 7-way set because the modal response inside each event category masks the lever the realized response actually used. Doctrinal retrieval has to beat this on lever F1 first, then window calibration. The v0 stub is the staging ground; the LLM swap in v0.2 is the experiment.

Seven choices the v0 gate does not foreclose.

• i.
  Translation of recordWhen PRC English diverges from CSIS / MERICS / Brookings, which translation does the public-facing corpus show by default? PRC text with a translator’s-note toggle exposing the divergence is the working recommendation.
• ii.
  Common-prosperity coverageThe v0 eval set is scoped to external-event-triggered responses and underweights common_prosperity. Add 2–3 internally-triggered events (2021 platform crackdown, real-estate intervention) as a separate sub-set so the external-event scoreboard stays clean.
• iii.
  NQPF vs. classical self-relianceFrom 2024 forward, state media use new-quality-productive-forces and self-reliance language interchangeably. Keep them as two frames: the lever implications differ (NQPF clusters to subsidy, classical self-reliance to retaliation_export_control + talent_restriction). The empirical separation is itself a prediction signal.
• iv.
  Hu Jintao corpus weightHu-era doctrine is rarely invoked in current state media but is diagnostically valuable as the peaceful-rise counter-anchor. Keep it in the corpus with low retrieval weight; when invoked, the predicted lever shifts toward rhetorical_only and the tier drops.
• v.
  Real-time corpus ingest cadenceDaily ingest of People’s Daily / Xinhua / Qiushi with weekly editorial-vector recomputation, matching the auto-research loop cadence.
• vi.
  Passage source weightingv0 weights Xi-era state media at 1.0 and historical anchors at 0.6 — modest, not aggressive, per the flat-prior counter-recommendation. The auto-research loop is the better mechanism for empirically learned weights.
• vii.
  Live-case publication policyPublish live ex-ante predictions before the realized response, even at the cost of being publicly wrong. The May 2026 talent-curb prediction is the proof of concept; publishing only post-hoc would compromise the falsifiability claim.