Evidence has a time boundary.
A preserved record does not become false merely because it is old. But its ability to support a present claim can weaken, expire or be displaced. Temporal validity asks whether the evidence still fits the claim at the exact moment a decision is made.
Temporal fit is sufficient for the present claim.
Confidence declines as change becomes plausible.
The record remains historical evidence.
The earlier record stays preserved, not erased.
Evidence does not decay. Its support relation does.
Age alone is not a defect. Temporal decay is a reduction in how strongly a record supports a claim at a later decision time, given the volatility of the observed state and the existence of newer evidence.
Old can remain fully valid.
A historical event record may remain definitive forever. A dated contract can still prove what was agreed on that date. The question is whether the claim is historical or current.
New can still be weak.
A recent observation can be poorly scoped, methodologically weak or detached from provenance. Freshness controls time fit; it does not repair quality defects.
Stale evidence stays evidence.
A record that cannot support “is true now” may still strongly support “was true at T₀.” Preserve it, narrow its claim and change its status.
Replacement creates a sequence.
A newer authoritative record may supersede an older state without proving the earlier record false. Both belong to the temporal history.
Move decision time. Watch support change state.
The model below is an explanatory instrument, not a universal truth score. It combines declared age, domain volatility, independent freshness and explicit supersession to expose the temporal reasoning behind a decision.
One record can contain four different times.
Temporal errors often begin when publication time is treated as observation time. Every material claim should preserve the clocks required to reconstruct when the state existed, when it was captured and when it was evaluated.
The state occurred.
The transaction closed, a page ranked, a policy entered force or an incident happened.
The state was measured.
The instrument or researcher observed the subject. This is usually the anchor for freshness.
The record became available.
A report published today may describe data collected months earlier. Recency of publication can mislead.
The evidence is used.
Validity is assessed here. The distance from relevant observation time determines temporal risk.
Never overwrite history. Version the support state.
A superseded record remains part of the evidence chain. The ledger preserves each observation, its effective interval, refresh events and the point at which a later record assumes current authority.
Make assumptions visible. Never disguise them as certainty.
A decay function can help schedule re-collection and compare temporal risk, but only after its parameters are declared. It cannot manufacture truth or replace domain judgment.
F(t) estimates temporal support fitness at elapsed time t. B is initial claim fit, λ is the declared decay rate, ρ is bounded fresh corroboration and N is the current-record factor. The function exposes assumptions; thresholds and half-lives still require calibration against observed change.
Initial claim fit after scope, quality and provenance checks. Time cannot rescue weak starting evidence.
Domain volatility translated into an expected loss of temporal relevance.
A bounded contribution from recent independent confirmation. It can reduce temporal uncertainty without rewriting the age or identity of the original record.
A supersession factor. A valid authoritative replacement can reduce current use to zero.
The same age means different things.
Freshness windows should follow how quickly the underlying state changes—not how convenient the collection schedule is.
SERP position
A page observed at position four seven days ago may already be unsuitable for a current ranking claim. Search results vary by query, location, device and time.
- VALID CLAIM
- “Position 4 in the preserved US desktop result at T₀.”
- INVALID LEAP
- “The page ranks fourth now.”
- ACTION
- Re-collect the same configuration before a present-state decision.
Live price
A captured price can remain perfect evidence of a past offer while losing present purchasing relevance within hours.
- VALID CLAIM
- “The displayed price was €149 at T₀.”
- INVALID LEAP
- “The product currently costs €149.”
- ACTION
- Preserve the capture and retrieve a new current offer.
Executive position
An official profile may strongly establish a role at publication time, but organizational appointments and departures can supersede it.
- VALID CLAIM
- “The company identified X as CEO on T₀.”
- INVALID LEAP
- “X is CEO” without a current check.
- ACTION
- Inspect current first-party records and effective dates.
Dated event
A verified filing proving that an acquisition closed on a specified date does not decay merely because years pass.
- VALID CLAIM
- “The acquisition closed on the recorded date.”
- INVALID LEAP
- Inferring the present ownership structure without later records.
- ACTION
- Retain for history; extend the chain for current state.
Refresh because risk changed. Not because a calendar blinked.
Scheduled refresh is useful, but event-driven re-collection is more precise. Trigger it when the claim, environment, source or decision consequence materially changes.
State-change signal
A new release, filing, appointment, incident or observed result shift indicates that the subject may have changed.
Claim expansion
A historical or local observation is being reused for a broader, current or higher-consequence claim.
Replacement record
A newer authoritative record appears and must be compared for continuity, correction or supersession.
Decision escalation
The same evidence is entering a more consequential decision that requires a narrower freshness tolerance.
Seven controls. One defensible time boundary.
The protocol converts an undated citation into an inspectable temporal support decision.
Type the claim
Separate historical, current, trend and forecast claims.
Recover times
Capture event, observation, collection and publication times.
Estimate volatility
Declare how and how quickly the subject can change.
Set freshness
Define an evidence window appropriate to the claim and risk.
Find successors
Look for correction, update, withdrawal or newer authority.
Assign state
Mark current, aging, stale, superseded or unresolved.
Act visibly
Use, qualify, re-collect, replace or retain historically.
Freshness is not a page property. It belongs to the claim-state relation.
The same source date can be acceptable for one proposition and unacceptable for another. This matrix separates the type of state being claimed, the mechanism by which it can change, the temporal anchor that must be preserved and the event that should trigger a new observation.
TEMPORAL TEST →
Rankings can change as the index, query interpretation, competitors, location and result features change.
Preserve query, locale, language, device and capture timestamp.
A recent capture may describe only one result environment.
Re-collect when the decision depends on the present result.
“The page appeared at position four in the captured result.”
Promotions, stock, taxes, currency and seller rules can alter the displayed state quickly.
The offer must be checked at the point it is needed.
A screenshot remains historical proof but may lose purchasing relevance within hours.
Refresh before committing cost, availability or delivery claims.
“The seller displayed €149 at the recorded time.”
Leadership changes discretely and may be reported after an effective date.
Publication date alone may not identify when the role changed.
Old biographies and directory pages often remain live after succession.
Compare current first-party records and authoritative registries.
“The organization identifies X as CEO as of the stated date.”
Terms, technical behavior and requirements change through named or silent updates.
Capture the version, release, effective date and preserved text.
Stable documents can persist for years; active products may change weekly.
Reassess when a successor record modifies the relevant clause.
“Version 3.2 declares the following behavior.”
The event does not change, although later evidence may refine what is known about it.
Separate when the event occurred from when it was reported.
Risk concerns interpretation, completeness or later correction—not ordinary freshness decay.
Review when higher-quality evidence materially changes the reconstruction.
“The filing records that the transaction closed on that date.”
Direction can reverse, seasonality can repeat and measurement rules can break comparability.
Preserve every observation period plus any methodological break.
A trend is valid only for the interval and measure used to estimate it.
Re-estimate when additional data or a structural change enters.
“The measured series increased during the defined six-month window.”
Most time errors are not arithmetic. They are boundary errors.
A timestamp can be perfectly accurate while the conclusion remains temporally invalid. These failure modes identify the exact point where an otherwise credible record is stretched beyond the state, interval or version it actually observed.
Publication time replaces observation time.
A report released in September may summarize data collected from January through March. Calling the evidence “September data” shortens its apparent age and hides the lag between reality, collection, analysis and publication. Recover the relevant observation window before making a current-state claim.
A live URL is treated as a current record.
Pages can remain accessible while their facts, authors, screenshots or specifications are obsolete. The absence of an archival date does not prove freshness. Look for effective dates, revision history, page changes and current first-party confirmation.
A newer record exists but is not connected.
Two records may describe different valid states in sequence. Without a supersession edge, a researcher may select the older statement, misclassify a normal update as a contradiction or erase evidence that the earlier state existed.
A recent summary makes old observations look new.
Republishing, quoting or retrieving an older datum does not refresh its observation time. The visible page may be new while the underlying measurement remains years old. Trace the temporal lineage to the original evidence-generating event.
A convenient schedule is called a validity window.
Weekly or annual review cycles are operational choices, not empirical evidence of how quickly a state changes. Calibrate windows using observed update frequency, change magnitude, decision cost and tolerance for stale conclusions.
Superseded evidence is deleted instead of reclassified.
Removing an older record destroys the sequence needed to explain change. Keep it as historical evidence, mark the interval during which it governed and connect it to the successor that now supports the present-state claim.
Apply temporal validity without inventing false precision.
The answers below clarify how freshness, decay, history, trends and supersession should be handled in ordinary research and decision work.
How old is too old for evidence?
There is no universal age limit. Evidence becomes too old for a claim when the underlying state could have changed enough to affect the decision and no sufficiently recent evidence confirms continuity. Define the state, its volatility, the decision consequence and the required currentness before setting a window.
Does evidence itself decay?
The preserved record does not physically lose what it documented. What weakens is the support relation between that record and a later claim. A price capture from last year can remain perfect evidence of last year’s displayed price while being unusable for today’s purchasing decision.
What is evidence half-life?
Evidence half-life is an analytical model for the time over which support for a specified claim is expected to fall materially under declared assumptions. It is not a natural constant. It should be calibrated from observed change and used to schedule review, not presented as a hidden truth score.
Can corroboration make old evidence current?
Fresh independent corroboration can support continuity of the state, but it does not alter the timestamp or identity of the original record. Treat the new observation as a separate evidence line and explain how it reduces uncertainty. If it measures a different population or condition, it may not refresh the original claim.
What is the difference between stale and superseded?
Stale evidence has exceeded the accepted freshness boundary without a confirmed replacement. Superseded evidence has a newer, sufficiently authoritative record that now governs the current-state claim. Both remain usable for historical reconstruction when their intervals are preserved.
Can an old source still be the best source?
Yes. Primary records of completed historical events, signed agreements, archived specifications and original research can remain the best evidence for what happened or was declared at that time. A new summary is not automatically better merely because it is recent.
Which date should be cited?
Cite the date that corresponds to the claimed state and preserve other dates needed for interpretation. That may include event time, observation window, collection time, publication time, effective date, retrieval time and decision time. Publication time alone is often insufficient.
How should silent page updates be handled?
Preserve dated captures, compare versions and avoid assuming that a current page reproduces its earlier content. If a material change cannot be dated precisely, mark the transition interval as uncertain rather than assigning an invented effective moment.
When does a trend claim expire?
A trend claim remains valid for the series and interval it describes. It should not automatically be projected forward. New observations, measurement changes, seasonality or regime shifts can alter direction. Re-estimate the trend whenever the decision concerns a later window.
How often should current claims be rechecked?
Use risk-based scheduling plus event triggers. Faster-changing states and higher-consequence decisions require shorter intervals. Recheck immediately when a release, filing, correction, incident, market shock or successor record indicates that the state may have changed.
Should a superseded source be removed from the page?
Usually no. Preserve it where it helps readers understand historical state or change, clearly label its interval and link to the current record. Remove or isolate it only when continued display would materially mislead and the historical value is negligible.
How should uncertainty about time be written?
State the known interval and the missing clock directly: “observed no later than,” “published from data collected between,” or “effective date not confirmed.” Do not replace temporal uncertainty with a precise date inferred from crawling, retrieval or page availability.
Time bounds support. Claims make the relation explicit.
EVD/08 determines whether evidence remains temporally fit. EVD/09 maps each bounded record to the exact claim, proposition or claim component it supports.
Evidence objects, claims, context and support relations.
EVD / 02 CLASSIFY Evidence Types & ClassesObserved, documentary, computational and derived evidence.
EVD / 03 LINEAGE Source ProvenanceOrigin, custody, transformation and derivation paths.
EVD / 04 QUALITY Source QualityAuthority, directness, independence and claim fit.
EVD / 05 CAPTURE Evidence Collection & PreservationAcquisition context, integrity and reproducible records.
EVD / 06 CORROBORATE Corroboration & TriangulationIndependent convergence across sources and methods.
EVD / 07 CONFLICT Conflicting Evidence ResolutionScope, time, meaning and method divergence.
EVD / 08 TIME Temporal Validity & Evidence DecayFreshness windows, volatility, supersession and re-collection.
EVD / 09 MAPPING Claim–Evidence MappingExplicit support paths from records to propositions.
EVD / 10 CONFIDENCE Confidence CalibrationBounded confidence aligned with evidence strength.
EVD / 11 GAPS Evidence Gaps & UnknownsMissing observations, blind spots and unresolved states.
EVD / 12 SYNTHESIS Evidence Synthesis & Decision ReadinessIntegrated evidence states for bounded decisions.