Elamur Groupantsmathlib-oai
The groupant beside tools/mathlib/src/oai.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.
Facts
Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.
Components
| component | label | where |
|---|---|---|
| approximant | partial | model.md |
| ast | partial | syntax.json |
| composition | partial | Composition |
| cost | partial | invariant-and-cost |
| descriptant | present | descriptant |
| invariant | partial | invariant-and-cost |
| tests | bounded checks | tests |
Obernants
The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.
| id | type | label | where |
|---|---|---|---|
behavior_complement | behavior_complement | partial source derived model | behavior_complement |
reverse_reconstruction | reverse_reconstruction | partial source derived model | reverse_reconstruction |
checkpoint_order | checkpoint_order | partial source derived model | checkpoint_order |
token_selection | token_selection | partial source derived model | token_selection |
retry_partition | retry_partition | partial source derived model | retry_partition |
record_count_observation | record_count_observation | partial source derived model | record_count_observation |
Parts
Model text
descriptant
mathlib-oai
run creates directory, replays state, returns early for terminal token; otherwise initializes state, checks STOP/page limit, forms encoded URL, handles bounded consecutive 503 responses, parses OAI error/token, writes compressed page before durable state append.
Epoch 2026-10-31
Prospective model retained before changes. Existing production bytes are preserved; additions are bounded inline tests. Nine source groupants expose ingestion and routing boundaries; typed children bind exact source bytes and reuse their parent syntax projection. Native tests run twice per profile; behavior mutants run as compiled source changes and exact bytes are restored before positive checks.
Important retained observation: network-looking helpers are not all pure (DNS inside github_host and books build); selection deliberately excludes those effects. Mixed-format manifest order is not deterministic, and error paths do not imply rollback.
No network/DNS/download, real catalog/corpus/model/teacher or database execution. Effect paths are source-derived; finite helper tests are not universal verification. No executable percentage surrogate or speed claim; future caps require metric/domain/reference and remain 1/2/5/10%.
Epoch notes
- Comments repointed to consolidated docs, 2026-11-04.
approximant
Forward relation
run creates directory, replays state, returns early for terminal token; otherwise initializes state, checks STOP/page limit, forms encoded URL, handles bounded consecutive 503 responses, parses OAI error/token, writes compressed page before durable state append.
Model observations include returns, retained state and ordered partial effects; on external failures the successful-prefix relation replaces a total-function claim.
Composition
run creates directory, replays state, returns early for terminal token; otherwise initializes state, checks STOP/page limit, forms encoded URL, handles bounded consecutive 503 responses, parses OAI error/token, writes compressed page before durable state append.
Scoped children isolate the declared contiguous source intervals. Their operations compose only where the source calls them; costs of overlapping parent/child descriptions are not added twice. Remaining effect paths retain explicit gaps.
catalog
behavior_complement
Subject: mathlib-oai, source SHA-256 bad99249391507309f3ac75c270b7c802490e1b6ae7f6ffb54460d8790404846; finite transition observation.
Terminal empty state token and STOP prevent requests. >20 consecutive 503 cannot continue sleeping; other HTTP status errors abort without checkpoint of that failed response. Source-derived network bounds.
Domain: stated source predicates, bounded inputs and available resources; arithmetic assumptions and external effect gaps as in the invariant. Operator is the stated relation/projection; complement is a source-derived inner exclusion, reverse retains compatible antecedents. Other views are source-derived partial constraints, not inverse laws.
Evidence: tests.md identifies executed helpers and unexecuted paths. No universal guarantee follows from the finite fixtures.
reverse_reconstruction
Subject: mathlib-oai, source SHA-256 bad99249391507309f3ac75c270b7c802490e1b6ae7f6ffb54460d8790404846; finite transition observation.
Last-state tuple admits many earlier lines, ignored short/header rows and unparseable counts mapped zero. Gzip page plus state constrains body/token at commit but cannot determine all request timing or discarded failed responses.
Domain: stated source predicates, bounded inputs and available resources; arithmetic assumptions and external effect gaps as in the invariant. Operator is the stated relation/projection; complement is a source-derived inner exclusion, reverse retains compatible antecedents. Other views are source-derived partial constraints, not inverse laws.
Evidence: tests.md identifies executed helpers and unexecuted paths. No universal guarantee follows from the finite fixtures.
checkpoint_order
Subject: mathlib-oai, source SHA-256 bad99249391507309f3ac75c270b7c802490e1b6ae7f6ffb54460d8790404846; finite transition observation.
Increment page/count, write gzip, finish+sync, append+sync state. State does not advance on page write failure; page may exist on append failure. Re-running can overwrite that uncheckpointed page.
Domain: stated source predicates, bounded inputs and available resources; arithmetic assumptions and external effect gaps as in the invariant. Operator is the stated relation/projection; complement is a source-derived inner exclusion, reverse retains compatible antecedents. Other views are source-derived partial constraints, not inverse laws.
Evidence: tests.md identifies executed helpers and unexecuted paths. No universal guarantee follows from the finite fixtures.
token_selection
Subject: mathlib-oai, source SHA-256 bad99249391507309f3ac75c270b7c802490e1b6ae7f6ffb54460d8790404846; finite transition observation.
resumption_token selects rfind textual tag, trims inner text, XML-unescapes with raw fallback. Missing, self-closing and explicit empty are distinct Option representations but all terminal after successful page.
Domain: stated source predicates, bounded inputs and available resources; arithmetic assumptions and external effect gaps as in the invariant. Operator is the stated relation/projection; complement is a source-derived inner exclusion, reverse retains compatible antecedents. Other views are source-derived partial constraints, not inverse laws.
Evidence: tests.md identifies executed helpers and unexecuted paths. No universal guarantee follows from the finite fixtures.
retry_partition
Subject: mathlib-oai, source SHA-256 bad99249391507309f3ac75c270b7c802490e1b6ae7f6ffb54460d8790404846; finite transition observation.
503 increases consecutive busy and waits clamp(retry_after or 30,5,600); successful text resets busy before OAI error parsing. NoRecordsMatch appends page+1 count zero; other OAI errors stop.
Domain: stated source predicates, bounded inputs and available resources; arithmetic assumptions and external effect gaps as in the invariant. Operator is the stated relation/projection; complement is a source-derived inner exclusion, reverse retains compatible antecedents. Other views are source-derived partial constraints, not inverse laws.
Evidence: tests.md identifies executed helpers and unexecuted paths. No universal guarantee follows from the finite fixtures.
record_count_observation
The checkpoint record count is body.matches('<record>').count(), a literal textual observation, not XML namespace-aware record counting. Tags with attributes or namespace prefixes are not counted; matching text in comments can contribute. It constrains the counter without promising semantic record cardinality.
Subject and source identity: manifest.json. Domain: scoped forward/invariant; relation is source-derived and partial. Evidence: tests.md distinguishes actual checks from unexecuted effect paths; finite fixtures do not prove all inputs.
tests
Bounded evidence
Selected actual source checks:
- mathlib::oai::tests::tokens_and_errors
- mathlib::oai::wave33::state_resume_and_stop_without_requests
These checks are file-level evidence; fragment branches not reached by these names remain source-derived.
External wave33 mathlib receipt retains native-commands.json, numeric statuses, raw debug/release logs, sorted exact outcomes, mutation logs and original/restored hashes. Seventeen named checks pass twice per profile; six production behavior mutants reject in both profiles, with restored positives. Source syntax is regenerated twice after formatting and compared byte-for-byte. Structural validation tests bindings, not Rust semantics.
No network/DNS/download, real catalog/corpus/model/teacher or database execution. Effect paths are source-derived; finite helper tests are not universal verification. No executable percentage surrogate or speed claim; future caps require metric/domain/reference and remain 1/2/5/10%.
invariant-and-cost
Invariant and cost
Invariant
Resume uses last valid >=3-cell nonheader line's page/token and cumulative parsed record counts. noRecordsMatch appends terminal state without gzip page. Successful page commit stores body before token checkpoint; crash can leave page ahead of state.
Cost
last_state O(state bytes); token/error substring scans O(body), gzip O(body) and full body memory. At most 20 waits for consecutive 503 before 21st abort; each clamped 5..600 seconds. Successful pagination itself unbounded.
Domain and gaps
No network/DNS/download, real catalog/corpus/model/teacher or database execution. Effect paths are source-derived; finite helper tests are not universal verification. No executable percentage surrogate or speed claim; future caps require metric/domain/reference and remain 1/2/5/10%.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/mathlib/src/oai.rs.groupant/