Sequential Forecast Update¶
Temporal update protocol — instantiates Bayesian Belief Updating
Revises a forecast as new observations arrive while preserving a record of prior forecast states and reasons for movement.
Sequential Forecast Update keeps a single forecast question alive over time: it folds each new observation in on a regular cadence, carries each posterior forward as the next round's prior, and preserves a running trail of what the forecast was at each step and why it moved. Its defining idea, false of both the one-shot diagnostic and the formal single-step engine, is that it is a temporal protocol — a discipline for repeated updating — whose value lies as much in the stable question and the movement history as in any single posterior. It is the machinery that makes a forecast a coherent time series rather than a sequence of unrelated guesses.
Example¶
A meteorology desk maintains one forecast: the probability that the storm's center makes landfall in Region R within 72 hours. Every six hours a new ensemble model run and fresh observations arrive. The current probability — say 35% — becomes the prior; the new data updates it to 48%; that value becomes the prior for the next cycle. The question stays fixed even as the storm itself moves, so the numbers are comparable across the whole sequence.
Each update is logged with its driver: "steering ridge weakened, ensemble members shifted west, +13 points." Forecasters and the public then see not just the current number but its trajectory and the reasons behind each move. The discipline of many small updates guards against the standing danger — swinging the whole forecast on a single noisy model run that a later cycle walks back.
How it works¶
- Fix the forecast question and its resolution criteria so successive posteriors are genuinely comparable.
- On each cadence tick, treat the last posterior as the prior and update on the new observation.
- Prefer many small updates to few large ones, so a single noisy signal cannot swing the forecast.
- Log each step's value and the reason it moved, building the movement history the protocol is prized for.
Tuning parameters¶
- Cadence — how often you update; frequent tracks change but amplifies noise if signals are correlated.
- Increment discipline — a cap on how far one observation may move the forecast in a single step.
- Question fixity — how strictly the target is held constant vs. allowed to be re-specified.
- Carry-forward decay — how much old evidence is discounted as the world moves.
- History granularity — how much of each step's rationale is recorded.
When it helps, and when it misleads¶
Its strength is that it produces a coherent, comparable time series and — echoing the superforecasting[n1] finding that the best forecasters update frequently in small increments — resists both stubbornness and overreaction. Its failure mode is that correlated sequential signals treated as independent make the forecast overconfident (the same wind field observed twice is not two updates), and a drifting or silently re-specified question makes the sequence incomparable. The classic misuse is re-anchoring hard on the latest dramatic observation each cycle and discarding the prior. The guarding discipline is to hold the question fixed, cap per-step movement, and check that each new observation is genuinely new information rather than a restatement of what already moved the forecast.
How it implements the components¶
update_question— it fixes the belief being revised and its resolution criteria so posteriors stay comparable across time.update_cadence— it sets the regular tick on which the last posterior becomes the new prior and the forecast is revised.posterior_estimate— it produces the running probability at each step and carries it forward as the next prior.
It decides no action — whether to order an evacuation is a decision_threshold set by Adaptive Decision Threshold. It does not audit whether each starting prior's source is legitimate (prior_source_record) or re-run under alternative priors (prior_sensitivity_check) — that is Prior Sensitivity Analysis; and it does not verify the sequence against realized outcomes (calibration_check, Posterior Risk Estimation).
Related¶
- Instantiates: Bayesian Belief Updating — repeated posterior-to-prior updating over time with an audit trail.
- Consumes: Likelihood-Ratio Reasoning weighs how much each fresh observation should move the running forecast.
- Sibling mechanisms: Bayesian Diagnosis · Likelihood-Ratio Reasoning · Posterior Risk Estimation · Adaptive Decision Threshold · Prior Sensitivity Analysis · Bayesian Model Update
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Sequential Forecast Update operates by updates the previous posterior with each new observation to compute a comparable current forecast. That concrete deployed or enacted form is Analysis, Modeling & Optimization under the frozen taxonomy.
Nearest alternative: Monitoring, Sensing & Alerting — Although Monitoring, Sensing & Alerting can support this mechanism, the frozen evidence makes its operative form the act that updates the previous posterior with each new observation to compute a comparable current forecast; the alternative is therefore secondary rather than defining.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Universal
Rationale: Revising probabilistic expectations as observations arrive is sequential Bayesian inference and time-series forecasting.
Related originating lineages:
- Data Science & Analytics — Online prediction systems operationalize rolling updates and retain model-state history.
- Economics & Finance — Forecast vintages and revisions are central to economic and financial forecasting practice.
- Futurism & Strategic Foresight — Recording why expectations moved supports accountable anticipatory judgment rather than retrospective storytelling.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: revises a forecast as new observations arrive while preserving a record of prior forecast states and reasons for movement.
Review resolution: The blind reviewers agree that statistics_experimental_design is the primary origin and differ only on alternate origin disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain convergent because the combined record shows independent disciplinary development. The broader reach of universal records portability separately from historical provenance, and encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
Sequential Forecast Update is a protocol, not a formula. It says nothing about how any single step's posterior is computed — that arithmetic can be a full Bayesian calculation, a likelihood-ratio update, or a structured qualitative rule. In particular it is not Bayesian Model Update, a mechanism under a different archetype that performs the single-step formal posterior computation; this protocol governs the sequence — question stability, cadence, and movement history — around whatever engine does the per-step math.
[n1] Superforecasting — Philip Tetlock's finding that the most accurate forecasters revise their probabilities often and in small steps as evidence arrives, rather than making rare large jumps; frequent incremental updating is a hallmark of calibration. ↩