Synthetic foresight and analytic tradecraft
Kill Web Anticipatory Intelligence Lab
Build a transparent system-level forecast, compare competing hypotheses, inspect source provenance and correlation, test calibration, and see why anticipatory intelligence must not become individualized pre-crime. Every scenario and value is fictional, aggregate, and non-operational.
Research basis KW-RPT-015 KW-RPT-016 KW-RPT-018 KW-RPT-021 KW-RPT-022
Ethical demarcation
Forecast systems and physical transitions—not a person’s hidden intent.
The lab permits only fictional, aggregate questions about infrastructure, logistics, communications, environmental conditions, and institutional processes. It rejects named-person risk scoring, guilt by association, protected-trait proxies, watchlisting, coercive recommendations, and any claim that probability creates authority.
- Object
- Aggregate system condition
- Inputs
- Synthetic evidence only
- Output
- Inspectable probability and uncertainty
- Permitted next step
- Review, collect, challenge, or abstain
Current analytic posture
Sustained system transition
Will the fictional North Meridian logistics network enter a sustained high-throughput phase within 21 days?
Permitted analytic next step
Review, collect, challenge, or abstain
The synthetic forecast is sufficiently evidenced for bounded analytic review. It still creates no authority and supports only further system-level monitoring, collection, and contingency analysis.
The result is synthetic and creates no authority.
Competing hypotheses
Do not let one fluent narrative become the whole decision space.
Each observation can support several explanations. The transparent calculation preserves alternatives rather than generating a single persuasive story.
Temporary exercise pattern
Declared prior: 43%
Sustained system transition
Declared prior: 32%
Civil maintenance and seasonal demand
Declared prior: 25%
Provenance and bias audit
Inspect what entered the forecast—and what did not.
Source count is not evidence diversity. The ledger exposes correlation discounts, stale observations, unverified derivatives, and whether contrary evidence was admitted.
| Observation | Role | Domain | Provenance | Age | Lineage discount | Effective quality |
|---|---|---|---|---|---|---|
| Expanded aggregate storage footprint | Supporting | Physical change | Synthetic multispectral series A | 18 hours | 1 | 81.0% |
| Rail throughput above seasonal band | Supporting | Transport | Synthetic transport ledger B | 10 hours | 1 | 74.8% |
| Published short exercise window | Contrary | Administrative context | Synthetic public notice C | 6 hours | 1 | 74.9% |
| Temporary storage capacity added | Supporting | Physical change | Synthetic multispectral series A | 30 hours | 0.52 | 31.1% |
| Severe-weather preparation signal | Contrary | Environmental context | Synthetic climate baseline D | 12 hours | 1 | 70.5% |
Calibration bench
A probability is accountable only after outcomes accumulate.
The fixed archive demonstrates a strictly proper scoring rule. Lower Brier scores are better; confident wrong forecasts are penalized more than cautious wrong forecasts.
Lower is better. A binary forecast of 50% receives 0.25 whether the event occurs or not; confident wrong forecasts are penalized more heavily.
| Archive case | Forecast | Outcome | Brier score |
|---|---|---|---|
| CAL-01 · Synthetic transport transition | 72% | Occurred | 0.0784 |
| CAL-02 · Synthetic relay interruption | 31% | Did not occur | 0.0961 |
| CAL-03 · Synthetic infrastructure delay | 64% | Occurred | 0.1296 |
| CAL-04 · Synthetic data-age breach | 42% | Did not occur | 0.1764 |
| CAL-05 · Synthetic throughput recovery | 78% | Occurred | 0.0484 |
| CAL-06 · Synthetic weather disruption | 57% | Did not occur | 0.3249 |
| CAL-07 · Synthetic storage transition | 81% | Occurred | 0.0361 |
| CAL-08 · Synthetic policy backlog | 36% | Did not occur | 0.1296 |
| CAL-09 · Synthetic supply shortfall | 68% | Occurred | 0.1024 |
| CAL-10 · Synthetic monitoring artifact | 24% | Did not occur | 0.0576 |
Pre-crime boundary
Function matters more than the label “decision support.”
A nominal human reviewer does not cure a system that allocates suspicion or coercion without objective, inspectable evidence and genuine authority to reject the machine output.
| Object or output | Lab status | Reason |
|---|---|---|
| Aggregate infrastructure phase change | Allowed in this lab | The object is a synthetic physical system condition, not a person. |
| Aggregate service interruption probability | Allowed in this lab | The output concerns system reliability and remains non-operational. |
| Named person’s future offending | Blocked | Rare-event person prediction creates severe base-rate, due-process, discrimination, and feedback-loop risks. |
| Religion, ethnicity, disability, ideology, or association as danger proxies | Blocked | Protected or identity-linked characteristics are not legitimate substitutes for objective system evidence. |
| Watchlist, detention, surveillance, targeting, or force recommendation | Blocked | A forecast cannot create legal authority, probable cause, command authority, or permission to apply coercion or force. |
Federated architecture concept
Share bounded evidence products without pretending all raw data belong in one lake.
Local custody
Raw records remain under the fictional source node’s control. The public lab never ingests external or classified data.
Inspectable summary
Only declared aggregate observations, provenance, age, and uncertainty enter the teaching calculation.
Privacy and leakage
Federation can reduce centralization risk, but model updates and summaries can still leak information or carry poisoned assumptions.
Quarantine and abstention
Failed provenance, anomalous lineage, or missing contradiction should remove an input or hold the forecast—not trigger a more confident story.
Transparent method
The entire teaching calculation is bounded and inspectable.
| Step | Rule |
|---|---|
| 1. Define the object | Use only a fictional aggregate system condition with a declared horizon and resolved outcome. |
| 2. Declare priors | Begin with an explicit base rate for each competing hypothesis rather than a hidden default. |
| 3. Admit evidence | Select synthetic observations and retain provenance, age, sensitivity, reliability, and collection lineage. |
| 4. Discount dependence | Repeated observations from one lineage receive less weight than genuinely independent evidence. |
| 5. Surface contradiction | Contrary evidence and missing alternatives remain visible; fluent narrative does not replace inspectable support. |
| 6. Combine cautiously | The transparent result and independent analyst estimate are combined, then shrunk toward the prior when evidence quality is weak. |
| 7. Score after resolution | The Brier score evaluates probability accuracy after the synthetic outcome is revealed. |
| 8. Preserve authority boundaries | The only permitted outputs are review, collect, challenge, or abstain. No operational or coercive decision follows. |
Answer-ready summary
Direct answers
What does the Anticipatory Intelligence Lab forecast?
Only fictional, aggregate system conditions such as infrastructure phase change, logistics throughput, communications-service stress, or evidence-sharing delay. It does not predict a named person’s intent or criminality.
Read the supporting pageHow does the lab reduce automation bias?
It asks for an independent human estimate, preserves competing hypotheses, identifies contrary evidence, exposes provenance and correlated sources, and shows reasons to hold or abstain.
Read the supporting pageDoes a high forecast probability authorize action?
No. Probability never creates legal authority, command authority, probable cause, target validity, or permission to surveil, detain, target, or apply force.
Read the supporting page