Elamur Groupantssources-bibliography-parse
The groupant beside tools/sources/src/bib.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 | approximant |
| ast | partial | ../syntax.json |
| cost | symbolic source derived | invariant-and-cost |
| descriptant | present | descriptant |
| invariant | partial | invariant-and-cost |
| tests | bounded evidence and explicit gaps | 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 |
delimiter_state | delimiter_state | partial source derived model | delimiter_state |
concatenation_tail | concatenation_tail | partial source derived model | concatenation_tail |
normalization_fiber | normalization_fiber | partial source derived model | normalization_fiber |
truncated_value | truncated_value | partial source derived model | truncated_value |
Model text
descriptant
sources-bibliography-parse
Byte state (i,depth,fields,out) recognizes quoted/braced/bare values, consumes # tails and clean, then inserts lowercased field names.
Epoch 2026-10-31
Prospective full-source read preceded edits. Model separates source-derived effects from bounded actual-helper evidence. Bibliography host identity old code admitted foreign authorities; a failing fixture motivated the narrow HTTP(S) exact-authority repair. Other production bytes were retained; added test modules are cfg(test). Topic run now uses the same canonical ID guard for JSONL admission; an old-code failing fixture confirms foreign matched entries remain only in abstract listings. Namespace compilation is representation evidence, not independent runtime coverage. Value workflow and bench transport remain unexecuted. No live data, network, model, DB or benchmark invocation was used.
approximant
Forward model
Byte state (i,depth,fields,out) recognizes quoted/braced/bare values, consumes # tails and clean, then inserts lowercased field names.
Domain and assumptions: ASCII delimiters scanned as bytes; normalization uses chars/NFC. Duplicate fields overwrite. Tolerant truncation is not grammar validation.
This source-derived relation abstracts allocation, formatting and external failures; no executable percentage surrogate is asserted. Categorical decisions/effects have no supported error metric yet. Future budget caps are 1%, 2%, 5%, 10% only.
catalog
Typed model catalog
Subject: exact manifest source SHA-256 and optional byte scope. Domain/assumptions: forward and invariant companion. Observation is one completed helper transition or the stated IO prefix; external interference is excluded for finite fixtures. All models are partial source-derived specifications; the complement is an inner set of excluded behaviors, the reverse a compatible-antecedent relation. Evidence: tests.md; unsampled branches are not proven.
behavior_complement
clean removes all braces and performs no IO. Recognized # tail cannot become next field.
Operator and approximation: the stated constraint on the source transition, with all unmentioned state held fixed; not a total solver or uniqueness claim. Falsifiable oracle: compare actual helper outputs/effect order on a bounded input satisfying the domain. Current executed subset and unsupported observations are listed in tests.md.
reverse_reconstruction
Equivalent whitespace, field-name case, braces and earlier duplicate fields share output; exact source not recoverable.
Operator and approximation: the stated constraint on the source transition, with all unmentioned state held fixed; not a total solver or uniqueness claim. Falsifiable oracle: compare actual helper outputs/effect order on a bounded input satisfying the domain. Current executed subset and unsupported observations are listed in tests.md.
delimiter_state
Quoted closes at depth zero; braced closes at depth zero after opening increment. Escape byte skip applies only to quoted values.
Operator and approximation: the stated constraint on the source transition, with all unmentioned state held fixed; not a total solver or uniqueness claim. Falsifiable oracle: compare actual helper outputs/effect order on a bounded input satisfying the domain. Current executed subset and unsupported observations are listed in tests.md.
concatenation_tail
Horizontal whitespace then # appends quoted/alphanumeric token; bare month macro text retained rather than expanded.
Operator and approximation: the stated constraint on the source transition, with all unmentioned state held fixed; not a total solver or uniqueness claim. Falsifiable oracle: compare actual helper outputs/effect order on a bounded input satisfying the domain. Current executed subset and unsupported observations are listed in tests.md.
normalization_fiber
NFC accent conversion, brace stripping, whitespace split/join identifies clean('{A} B') with clean('A B').
Operator and approximation: the stated constraint on the source transition, with all unmentioned state held fixed; not a total solver or uniqueness claim. Falsifiable oracle: compare actual helper outputs/effect order on a bounded input satisfying the domain. Current executed subset and unsupported observations are listed in tests.md.
truncated_value
Unclosed braced/quoted value extends to available end. Existing tests cover listed fragments only, not all malformed Unicode boundaries.
Operator and approximation: the stated constraint on the source transition, with all unmentioned state held fixed; not a total solver or uniqueness claim. Falsifiable oracle: compare actual helper outputs/effect order on a bounded input satisfying the domain. Current executed subset and unsupported observations are listed in tests.md.
tests
Checks and limits
Exact target-qualified native names for this source family:
- bib::tests::latex_accents_become_unicode
- bib::tests::concatenated_month_keeps_following_fields
- bib::tests::malformed_snippets_do_not_panic
- bib::ids_tests::ids_to_manifest
- bib::ids_tests::slug_words_keep_accented_capitals
- bib::ids_tests::unknown_id_and_bad_topic_are_red
- bib::wave33_tests::ids_missing_preserves_output_and_license_boundary
- bib::wave33_host_tests::anthology_id_requires_exact_http_host
- bib::wave33_topic_tests::foreign_matching_entry_stays_in_abstracts_not_manifest
Run each selected exact name twice debug and twice release, one test thread, offline/locked dependencies. External wave33 sources receipts contain numeric statuses/raw logs and sorted outcome comparisons. Scoped children inherit only relevant evidence: parse/IDs directly exercised; admission/check/merge actual helpers exercised; subprocess actual helper exercised. bench_transport has source-derived evidence only, never an executed benchmark. Library namespace is compile/representation evidence.
Compiled source controls change bibliography license threshold, sources size guard and value status guard; targeted assertions must reject, then exact backup restoration and positive rerun. Finite controls are not comprehensive fault coverage. See evidence.md at source root for receipts summary.
invariant-and-cost
Invariants and costs
ASCII delimiters scanned as bytes; normalization uses chars/NFC. Duplicate fields overwrite. Tolerant truncation is not grammar validation.
O(B) stored text; repeated suffix collections may be quadratic; BTreeMap adds log field count.
Bounds are symbolic, conditional on stated domain and successful allocation. No unmeasured speedup or universal finite-test guarantee.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/sources/src/bib.rs.groupant/parse.groupant/