Reference

Glossary

Every Field 3.0 term defined in one place. If you've heard a word in the docs and don't have a definition for it, it's here.

1 min readUpdated May 12, 2026Edit on GitHub

Field 3.0: Atlaso's knowledge model — append-only, dispersion-aware deposits with polarity, evidence grade, and a six-facet scope. The numeric version reflects three internal rewrites; the current shape is production-locked.
Deposit: One immutable record of something the system observed. Carries content, polarity, evidence_grade, scope, tags, provenance, and any contradicts edges. Deposits never update — to revise a fact you call contradict().
Polarity: A deposit's direction: positive (X works) / negative (X doesn't) / cautionary (X works but…) / open (a question or note, no direction). The first three are directional and feed conflict detection.
Evidence grade: How well-supported a deposit is. anecdotal → observed → replicated → verified. The gate weighs this against scope breadth.
Scope: Six optional facets (model, dataset, env, version, n, seed) plus a free-form note. Pins the deposit to a context. The scope is the grouping key for conflict.
Scope bag: The set of deposits sharing the same six scope facets. The bag is what gets a precision score, a conflict check, and the is_confident verdict.
Gate: The write-time policy that decides whether a deposit may enter the store. Bases acceptance on polarity × scope_breadth × evidence_grade. See Architecture · The gate.
Scope breadth: Narrow when ≥2 scope facets are non-null; otherwise broad. Broad positives need higher evidence; narrow negatives only need provenance.
Provenance: At least one artifact_refs entry OR a scoped env / version. Required for narrow-negative and cautionary deposits.
Conflict: A scope bag contains two or more distinct directional polarities drawn from {positive, negative, cautionary}. Atlaso flags it before your agent sees the data.
is_confident: Bag-level verdict, true iff bag_precision >= 0.99 AND not is_single_sample AND no has_conflict. Three conditions, strict.
agreement_score: Public alias for bag_precision— fraction of the bag's directional deposits matching the dominant polarity, in [0, 1].
FMI (Field Maturity Index): A 0–100 score: geometric mean of Coverage × Precision × Resolution × Density over a rolling window. Geometric, not arithmetic, so a single weak pillar drops the score sharply. See Diagnostics & FMI.
Coverage: FMI pillar — rolling fraction of recall calls that returned at least one confident hit. Drops when the field can't answer the questions being asked.
Resolution: FMI pillar — fraction of conflict bags that have been adjudicated (a deposit in the bag is the target of a contradicts edge).
Density: FMI pillar — 1 − singleton_bags / total_bags. Drops when most bags have a single observation.
Idempotency key: Per-item dedup key required by add_many(). Helper idempotency_key(*parts) returns a content-addressed ak_… hash. 24-hour replay window.
Soft retract: Adds an atlaso:retracted=<reason> tombstone tag. The deposit becomes invisible to recall() but the row survives — the audit chain is preserved.
Hard delete: Irreversible. Removes the row from deposits and clears the FTS index. Reserved for GDPR / PII erasure. Triggered by retract(..., hard_delete=True).
UserHandle: Frozen per-user view returned by Memory.for_user(...). Pre-fills user_idon every call so identity can't be fumbled per-method.

Was this page helpful?

Reference

Glossary

Every Field 3.0 term defined in one place. If you've heard a word in the docs and don't have a definition for it, it's here.

1 min readUpdated May 12, 2026Edit on GitHub

Field 3.0: Atlaso's knowledge model — append-only, dispersion-aware deposits with polarity, evidence grade, and a six-facet scope. The numeric version reflects three internal rewrites; the current shape is production-locked.
Deposit: One immutable record of something the system observed. Carries content, polarity, evidence_grade, scope, tags, provenance, and any contradicts edges. Deposits never update — to revise a fact you call contradict().
Polarity: A deposit's direction: positive (X works) / negative (X doesn't) / cautionary (X works but…) / open (a question or note, no direction). The first three are directional and feed conflict detection.
Evidence grade: How well-supported a deposit is. anecdotal → observed → replicated → verified. The gate weighs this against scope breadth.
Scope: Six optional facets (model, dataset, env, version, n, seed) plus a free-form note. Pins the deposit to a context. The scope is the grouping key for conflict.
Scope bag: The set of deposits sharing the same six scope facets. The bag is what gets a precision score, a conflict check, and the is_confident verdict.
Gate: The write-time policy that decides whether a deposit may enter the store. Bases acceptance on polarity × scope_breadth × evidence_grade. See Architecture · The gate.
Scope breadth: Narrow when ≥2 scope facets are non-null; otherwise broad. Broad positives need higher evidence; narrow negatives only need provenance.
Provenance: At least one artifact_refs entry OR a scoped env / version. Required for narrow-negative and cautionary deposits.
Conflict: A scope bag contains two or more distinct directional polarities drawn from {positive, negative, cautionary}. Atlaso flags it before your agent sees the data.
is_confident: Bag-level verdict, true iff bag_precision >= 0.99 AND not is_single_sample AND no has_conflict. Three conditions, strict.
agreement_score: Public alias for bag_precision— fraction of the bag's directional deposits matching the dominant polarity, in [0, 1].
FMI (Field Maturity Index): A 0–100 score: geometric mean of Coverage × Precision × Resolution × Density over a rolling window. Geometric, not arithmetic, so a single weak pillar drops the score sharply. See Diagnostics & FMI.
Coverage: FMI pillar — rolling fraction of recall calls that returned at least one confident hit. Drops when the field can't answer the questions being asked.
Resolution: FMI pillar — fraction of conflict bags that have been adjudicated (a deposit in the bag is the target of a contradicts edge).
Density: FMI pillar — 1 − singleton_bags / total_bags. Drops when most bags have a single observation.
Idempotency key: Per-item dedup key required by add_many(). Helper idempotency_key(*parts) returns a content-addressed ak_… hash. 24-hour replay window.
Soft retract: Adds an atlaso:retracted=<reason> tombstone tag. The deposit becomes invisible to recall() but the row survives — the audit chain is preserved.
Hard delete: Irreversible. Removes the row from deposits and clears the FTS index. Reserved for GDPR / PII erasure. Triggered by retract(..., hard_delete=True).
UserHandle: Frozen per-user view returned by Memory.for_user(...). Pre-fills user_idon every call so identity can't be fumbled per-method.

Was this page helpful?