Abduction Log¶
Artifact — instantiates Abductive Explanation Selection
A running record of explanation changes, retired rivals, new evidence, and revision triggers.
The Abduction Log is an append-only, timestamped ledger of how an explanation evolves. Each entry records the current best-so-far account, which rivals were retired and why, what evidence arrived, how the confidence label shifted, and which pre-set revision triggers fired. Its defining move — the thing no snapshot document does — is preserving the trajectory: a reader can see not just what is believed now but the whole path that led there, so settled ground is not re-fought and, crucially, the slow accumulation of ad hoc patches becomes visible instead of hidden inside a currently-confident conclusion.
Example¶
A satellite operations team notices intermittent telemetry dropouts. Day 1 entry: best-so-far = a ground-station antenna fault; rivals logged = onboard transponder degradation, solar-array power sag during eclipse, a software watchdog reset; status = working, low confidence. Day 3: dropouts recur while a second, independent ground station is in contact — the pre-set trigger "recurs cross-station" fires; the antenna-fault explanation is retired and the entry records why. Day 5: the dropouts correlate tightly with eclipse entry; best-so-far becomes power sag. The same entry flags that one off-pattern dropout required an extra assumption to fit — an early ad-hoc patch, noted as a watch item. Weeks later, reading only the latest line would show a confident "power sag" verdict; reading the log shows a disciplined narrowing with one unexplained residue still flagged. No one reopens the antenna question, because the log shows exactly why it died.
How it works¶
- Append, never overwrite. Every entry is timestamped and additive; the history is the point.
- Each entry carries five things: current best, rivals retired (with reason), new evidence, updated confidence label, and any triggered revision condition.
- Pre-declare revision triggers so that when one fires, the log demands a reopening rather than a footnote.
- Count the patches. The log flags when the leading explanation survives only by accreting ad hoc auxiliaries, surfacing degeneration a snapshot would conceal.
Tuning parameters¶
- Entry cadence — per-event versus scheduled. Per-event captures every twist but is heavy; scheduled is lighter but can miss fast-moving reversals.
- Append-only strictness — strictly immutable versus editable. Immutability protects the audit trail; edits keep it readable but invite quiet sanitizing.
- Patch-count sensitivity — how many ad hoc rescues accumulate before the log flags "reconsider the whole explanation."
- Confidence vocabulary — the label set the status field draws from; a shared vocabulary makes trajectories comparable across cases.
When it helps, and when it misleads¶
Its strength is threefold: it stops teams re-litigating retired rivals, it makes revision auditable, and — uniquely — it exposes a degenerating explanation that is being propped up by ever more special-case fixes.
Its failure mode is the mirror image: an explanation kept alive by endless ad hoc rescue can look robust to anyone who reads only the latest entry, because each patch individually seemed reasonable — a degenerating problemshift dressed as resilience.[n1] The classic misuse is back-dating or editing entries to launder the trail after an outcome is known. The guarding discipline is strict append-only entry, an explicit patch count, and treating a fired trigger as a mandate to reopen the comparison, not a note to file.
How it implements the components¶
best_so_far_explanation— logged at each entry as the current provisional account, versioned over time.rival_comparison_record— records which rivals were retired and the reason, preserving them until genuinely dead.defeasibility_status_label— the confidence tag that shifts entry to entry, making changing certainty legible.revision_trigger— pre-declared conditions whose firing the log captures and acts on.
It does not implement candidate generation or the explanandum_boundary (Anomaly-to-Hypothesis Workshop), the explanatory_fit_criteria comparison (Inference-to-Best-Explanation Matrix), or the discriminating_evidence_plan (Disconfirming Probe Plan) — it records evidence as it lands but does not design the probes.
Related¶
- Instantiates: Abductive Explanation Selection — the archetype's temporal record of explanatory change.
- Consumes: Inference-to-Best-Explanation Matrix — each entry snapshots the current comparison the matrix produced.
- Sibling mechanisms: Inference-to-Best-Explanation Matrix · Differential Diagnosis Workup · Explanatory Case Memo · Anomaly-to-Hypothesis Workshop · Disconfirming Probe Plan · Forensic Scenario Reconstruction · Model-Debugging Hypothesis Loop
Editorial Notes¶
Form Classification¶
Form family: Record, Log & Register
Rationale: The mechanism is a running record of explanation changes, retired rivals, new evidence, and revision triggers, so its operative form is an accumulating history or provenance record.
Independent corroboration: The frozen evidence defines Abduction Log as 'A running record of explanation changes, retired rivals, new evidence, and revision triggers', so its operative form is Record, Log & Register.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Philosophy
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Universal
Rationale: Abduction as inference to the best explanation is a philosophical and logical tradition, especially Peircean, and supplies the record's explanation-selection logic.
Related originating lineages:
- Accounting & Auditing — Append-only, timestamped audit-ledger practice contributes the non-overwriting evidentiary history.
- Security Studies & Intelligence Analysis — Competing-hypotheses practice materially contributes the tracking and retirement of rival explanations as evidence changes.
Review resolution: Peircean abduction supplies the governing inference, while append-only audit practice and competing-hypothesis investigation materially shape the log; historiography is a use context rather than a necessary lineage.
Attribution caveat: The named log is an Encyclopedia synthesis built from a philosophical inference rule and later investigative recordkeeping practices.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; medium confidence.
Notes¶
[n1] In Imre Lakatos's account, a research programme is degenerating when it survives disconfirmation only by adding ad hoc auxiliary hypotheses that predict nothing new — the mirror of a progressive one that anticipates novel facts. An abduction log's patch count is the practical instrument for catching this drift in a single investigation. ↩