Concepts and glossary

Canonical definitions for TheSpider campaigns, slices, leases, findings, and identity

Audit structure

TermDefinition
ProjectTenant-scoped container for repository identity, mode, methodology versions, campaigns, and finding identity.
MethodologyImmutable configuration version containing buckets, categories, passes, scopes, and prompt sections.
CampaignAudit of a target branch and commit using one methodology version. Campaign types are full, incremental, and retest.
Audit runOne campaign cell for a pass and bucket.
SliceSmallest execution unit: one audit run assigned to one model. A slice moves through pending, running, completed, or failed state.
LeaseTime-bounded claim granting one worker the right to execute a slice. Expiry requeues the slice; first accepted result wins.
ReportOne model-emitted observation parsed from a slice's raw stdout. Reports retain model, pass, bucket, file, category, CWE, and evidence provenance.

Finding identity

TermDefinition
FindingProject-level deduplicated issue that teams triage. Security-mode projects call it a finding or vulnerability; bug-mode projects call it a bug.
Finding fingerprintDeterministic identity derived from normalized, scrubbed report content. A project maps each known fingerprint to one canonical finding.
Fingerprint aliasAdditional fingerprint mapped to an existing canonical finding after adjudication or merge. Aliases preserve identity when wording or evidence changes.
SightingEvidence that a finding appeared in a campaign, branch, and commit. Comparison uses sightings rather than the campaign where the finding was first created.
KeeperCanonical finding that remains active after a merge.
Soft mergeMerge that points a losing finding to its keeper instead of deleting it. Reports, aliases, sightings, and provenance resolve through the keeper while the loser remains an auditable tombstone.

Finding identity is project-global, not branch-local. A finding can have sightings on several branches and campaigns while retaining one triage state.

Campaign comparison

TermDefinition
BaselineCampaign selected as the prior side of a comparison. An explicit baseline wins; otherwise the preceding campaign is used where supported.
New findingLive keeper sighted in the selected campaign but not the baseline.
Resolved findingLive keeper sighted in the baseline but absent from the selected campaign. Absence should not be treated as proof while campaign slices remain incomplete.
Carried-forward findingOpen keeper sighted in both selected and baseline campaigns.
GateCheap authenticated campaign read returning pass, fail, or pending from slice completion and new findings. CI polls the gate; it does not run models.

Remediation and retest

TermDefinition
Fix recordRemediation record with description, pull request, ticket, and one or more commits.
False-positive recordDecision record with reason, decider, notes, and optional commit through which the decision applies.
Retest campaignFinding-scoped campaign that runs the finding's bucket and passes matching its files.
Fix verificationRetest conclusion attached to a fix. confirmed means the covered retest did not reproduce the finding; regressed means the same or deduplicated finding reappeared.

Execution boundary

TermDefinition
BrainServer-side logic for prompts, parsing, scrubbing, fingerprints, deduplication, triage, usage, and persistence.
AgentThin local process that detects Git metadata, claims leases, runs model commands in a checkout, and uploads raw stdout plus a short stderr tail.
RunnerCommand bound to a model code on an agent host. It consumes a prompt file and emits marker-delimited findings on stdout.
ArtifactContent-addressed prompt, output, normalized output, or stderr object stored for provenance.

Continue with campaign workflows, the runner contract, or the API reference.