The choices in the chronolexography implementation that are still open, and the parts that need investigation or a large change.
This page lists what is still open from implementing chronolexography, one row per question. Some rows lie outside chronolexography itself, in the Awb, the engine, storage or the PoC, because that is where the work ran into them. The RFCs link here instead of keeping their own lists. When a question is settled, its row goes.
A question is listed when:
Missing functionality that only needs building is not listed. It gets built when a case needs it. A limitation of the proof of concept is not listed either, unless it affects how chronolexography works.
Open questions kept elsewhere:
The impact column says how far an answer reaches:
| Question | Why open | Impact | RFC |
|---|---|---|---|
REDUCTIE, projection, source.projections and combine: ONLY exist for the cell, not for the law. Should they leave the law format and move to a profile of their own? Today only cel.yaml says which answers a cell gives. A profile would also settle whether picking one record needs its own operations beside combine: ONLY, such as taking the latest record (LATEST) or the first that matches (FIND). | Choice; breaking schema change | High | 048 |
| A cell can receive a copy of another cell’s decision. It keeps the decision value as it came in. It cannot check that value against the other cell’s chronicle, so it trusts the copy, as a court trusts the documents a party submits. Should a receiving cell be able to check a copy at its source? | Choice; investigation | Foundational | 053 |
The cell’s API only appends to a chronicle. A copy kept outside the cell, such as the JSON Lines file or the browser of the PoC, can still be edited. Cell::open and Cell::restore check what they read back: unique ids, dates that do not go back, the recording actor, and that a reference points to an earlier record. A changed value that keeps those rules is taken as it is. A hash chain over the records would make a changed, removed or inserted record visible. Only a head kept outside the chronicle, or signed, would also stop someone who rewrites the whole chain or cuts off its end. That is a property of how a cell stores its chronicle, apart from the model, and not part of the PoC. Should a cell protect its stored chronicle this way, and who keeps or signs the head? | Choice; not part of the PoC | Low | none |
| An entry carries no format version, so a change in the form of an entry leaves a chronicle stored before it unreadable. Since PR #1920 the cell refuses a chronicle that holds a decision taken in turn as an entry of its own. Like the integrity of a stored chronicle, this belongs to storage and not to the PoC, which starts every session with an empty chronicle. Should an entry say its form, so that a cell can read an older one without rewriting it? | Choice; not part of the PoC | Low | 046 |
A chronolexogram (an entry in the chronicle) has one date, recorded_on. The cell uses it as the day of receipt (Awb 4:13, 6:9) and as the day a decision is made known (Awb 6:8). Another cell may record something late, or give it an earlier date. That changes what a replay of a decision sees. Does a chronolexogram need a second date for the moment that counts in law? | Choice; breaking change to the chronicle | Foundational | 046, 048 |
A Cell in packages/cel holds one chronicle. The position paper allows more: “Een chronolexocel kan één of meerdere chronolexokronieken beheren”. Each chronicle (chronolexokroniek) has a time axis of its own, and “die ordening bestaat niet over kronieken heen”. The cell falls short of the paper here, and several chronicles per cell have to be built. Two things are still to be decided. What bounds a chronicle: the rulework of the establishing article, the procedure, a register, or a list in cel.yaml? A record that refers to another could follow its target, but an article of the Awb, such as the objection or the suspension, belongs to no single procedure. And how does a decision read a record from another chronicle of its own cell, when nothing orders the two? The leave decision of OBS De Klimop (Leerplichtwet 1969, article 13a) reads the school’s establishment, a fact that refers to no application. The decision and its replay need a point in that other chronicle to read up to. The calendar date is the only measure the chronicles share, so that point depends on which date a record carries. This question is settled together with the second date above. | Choice; large refactor | High | 046, 048 |
| To explain a decision, the cell takes it again on its original day (a replay). That assumes every source can say what it knew on a given day. A source that keeps only its current state needs a facade that stores each value with its date. How that works for real registers needs investigation. | Investigation | Foundational | 046 |
The corpus keeps only its current state: a corrected version keeps its valid_from and loses its old content. An entry names the versions it was taken under (law_versions), so a replay can say which rulework changed since, but it still runs under the corpus as it is now. Should a replay run under the recorded version? That needs a store of every content by its hash, and an answer to what a replay under an outdated version explains. Part of law_versions repeats the corpus: which version was in force on the day of an entry follows from valid_from. A version with retroactive effect, added later, changes what that derivation gives today. Retroactive effect is lawful in itself. If the corpus is not to change what it held on a past day, it needs a second time axis beside valid_from: the day a version became known. Every version has a publication_date, which might serve as that axis. If it does, an entry was taken under the version with the latest valid_from among those whose valid_from and publication_date both fall on or before its recorded_on. The entry then need not keep valid_from. It would still keep the $id of each rulework it read, so that Cell::law_since keeps working without a replay. It would also keep each hash, which shows that a version was changed in place after the entry. With two axes a correction could become a new version with the same valid_from and a later publication_date. The rule above would then take the latest publication_date among versions with the same valid_from, and the corpus, which now names and finds a version by its valid_from alone, would need to tell the two apart. The hash catches a change that skipped that rule. Whether publication_date can serve is unchecked. In some versions it lies years after valid_from (the Awb version of 1994 says 2026, so the rule above would hide it from every entry recorded before 2026), and the day of publication is not the day the corpus added a version. Also open is whether the derivation belongs in the engine or in the cell. Last, the engine finds hooks in the newest version of a rulework such as the Awb, even on a day that version was not in force. Which hooks fire then depends on a version that neither law_versions nor a derivation by date names. The engine notes on the trace and the receipt that it read another version (DeclarationsFromOtherVersion), but the entry does not keep that note. | Choice; investigation | High | 046 |
| Who fixes the name of an answer? A binding asks a cell for an answer by name, and the cell’s catalogue has to give the answer under that name. Who sets the name, and how it is versioned, is open. The position paper names Consumer Driven Contracts as a research question here. | Choice; investigation | Medium | 048 |
The engine asks the cell that a term names wherever the policy that binds the term runs. RFC-052 asked it only when another law reached the term through source.regulation. Should the engine route by competence everywhere, or only where another law reads the term? | Choice | Low | 048 |
| The engine builds the parameters of a binding before it asks a cell. If the binding cannot fill a parameter, the decision fails as soon as there is a cell to ask, even a cell that turns out not to give the answer. Should a missing parameter make the binding silent instead? | Choice | Medium | 048 |
Which instance does the law mean? The law names a category, and the engine asks the cell of that category in the call. The cell does not pass its own identity to the law. So the engine does not find a policy scoped by gemeente_code, and cannot check competence by place of residence. See also #1029. | Choice | High | 048 |
| A term that reads a fact repeats the list of authorities competent for that fact, and the two lists can drift apart. To derive the term’s list from the fact, the engine would need to know which fact a term reads. Should it? | Choice | Low | 051 |
| The cell takes the inputs of a decision only from the deciding article. A finding that the law places in another regulation must therefore sit on the deciding article and be passed on as a parameter. The rulework then departs from the law. | Large refactor | High | 046 |
| When a policy fills an open term, the policy’s value counts. So the authority cannot depart from the policy in a single case, although Awb 4:84 allows that. A policy can let the authority judge certain cases itself, with an open term of its own, but only the cases the policy describes. How should the authority record a departure that the policy does not foresee? | Choice | High | 045, 048 |
| The version history says what each version adds to the schema. It does not say what changed in the engine and the cell at the same time, although a rulework author notices those changes too. Nothing checks that the newest version has a row. Should each version get release notes, split into schema and behavior, with a CI check, filled in back to v0.8.0? | Choice | Low | none |
A reduction filters rows with a FOREACH, once per question. A filter on a source would let a reduction ask only for the rows it needs, for example a where on an input, or a lambda passed to a reduction it calls. Since schema v0.8.17 source.projections has such a where, by the values its roots hold, and since v0.8.19 a reduction reads recorded facts through aggregates too. A where on values other than those of the root, or a lambda, is for a later change. | Choice | Low | 048 |
| The Awb does not provide for an applicant to withdraw an application before the decision; only case law recognizes that. Nor does it provide for withdrawing a favorable decision at the holder’s request, except for subsidies (Awb 4:48–4:50). Article 6:21 is about withdrawing an objection or appeal, not an application. No effect in the Awb therefore leaves out a decision that was withdrawn on request, such as granted leave that the holder wants moved to other days. We do not fill this in with a rule of our own. | Not in the law; left open on purpose | Medium | 048 |
| One question to several cells of a category. The law names a category, such as the college of any municipality, and more than one cell of it may hold the answer. How that works in practice, and which cell is asked, is open. Until then the engine asks one cell and merges no answers. If the call holds more than one cell that the term names, the decision fails. | Choice; investigation | Medium | 048 |
A step of a reduction says which output of a law each property of the state takes, through input and source.regulation. That reference could move into the law itself, for example as an entry on the produces item that points at the article, and the step would disappear from the reduction. The alternative is that the reduction says per produces item what it does to the state. Which of the two holds once more cases use a projection is open. | Choice | Medium | |
| The actions of a step take only a bare reference as their value: an input, a field of the record or a property of the state. Whether a reduction needs operations of its own there is to be found out. Until a case needs one, the rule stays strict, so what a record does stays in the law. | Choice | Low | |
Should a reduction and a regelwerk be written in the same YAML language? They share the expression language: operations, types, references, input with source, the engine, the load checks and the trace, and steps now take inputs as articles do. Yet a REDUCTIE takes the shape of a law and refuses most of it: no competent_authority, hooks, overrides, produces, open terms or implements, and text and legal_basis mean nothing there. That suggests a different kind of document. Syntax that only a reduction uses makes the question stronger: $state, $record, the #<KIND> suffix on a step’s record, projection.per and source.projections. Since v0.8.19 each article of a REDUCTIE is an aggregate or a read model, and an RFC that makes the reduction a document type of its own can start from those two forms. One direction is a shared schema for the building blocks with two roots: a regelwerk with articles, text and a layer, and a reduction with a projection, an execution and the answers it gives. That goes together with renaming regulatory_layer, for which a name such as data or data operations is in view. It changes the format and needs an RFC. | Choice | High | |
The engine plays every aggregate of a kind on every question that reads them (source.projections), from the first record of the chronicle. That is fine for the PoC. A real cell with many decisions would need to keep the last state of each aggregate as a snapshot, and then the version of the reduction in force on the calculation date may differ from the one that played it. A snapshot changes how fast a row is read, never what it holds, so this is a question for the engine and the cell. How a cell keeps and invalidates snapshots is open. | Investigation | Medium | |
The court’s cell can decide on an appeal (BEROEP), but not on a request for a provisional measure (Awb 8:81): the schema knows AANVRAAG, BEZWAAR and BEROEP as kinds of submission. So the PoC records the receipt of the voorzieningenrechter’s ruling at the municipality without playing the ruling in the court’s cell, and a test builds that receipt as Cell::record_receipt would. Should the schema get a submission kind for a request (verzoek), or should a request for a provisional measure be recorded under an existing kind such as BEROEP? | Choice; schema change | Medium | 053 |
| The reduction plays in the version in force on the calculation date. A new version that points a step at another article reads the whole history again in the new way. That is defensible while a reduction only couples outputs to properties, and it belongs with the question of which date and which version apply. | Choice | Low | |
| Article 2, first paragraph, of the Regeling gehandicaptenparkeerkaart issues no card without a medical examination, and article 3, first paragraph, has the GGD or an expert the municipality appoints carry it out. As the law reads, the college asks the GGD for the examination and the GGD then carries it out; Awb 3:2, under which the authority gathers the knowledge it needs about the facts, supports that reading. The PoC does not model the request: the GGD records its advice without one from the municipality, and the advice also proves the examination. A possible next step records the request in the municipality’s chronicle, after which the GGD records its advice on that request and refers to it. No choice has been made. | Choice | Low | 051, 052 |
The engine no longer reads from decided_per_case_by who decides. RFC-031 describes the field as the name of the authority that judges the norm in each case. The proposal is that the field only marks that the norm is judged per case. The term’s competent_authority then says who judges. Without one, the authority that carries out the article judges (Competent Authority). This still has to be aligned with RFC-031. | Choice | Low | 031, 048 |
From the law and the policy, the engine derives which open terms a decision may need and who answers each one (decision_terms). The cell adds the parameters from the application and the date of the decision. The process decides when and how it asks the authority. Should the engine also say which of a policy’s own open terms a case needs, without running the decision? | Choice | Medium | 048 |
A reduction, or the policy it calls, can pick one record with combine: ONLY. It fails when two records qualify. A correction recorded on the same day then blocks the decision, and the chronicle cannot undo it, because entries are never removed. Should a later record replace an earlier one? A reduction with a projection (schema v0.8.14, experimental) can already let a later record set a property again, so the last one counts, and a fact that says an earlier one was wrong from the start can revoke it (packages/cel/bdd/features/corrections.feature); ONLY itself is unchanged. | Choice | Foundational | 048 |
The fictitious parking ordinance of Wilgendaal lets the college revoke a business parking permit, at the holder’s request or when the company no longer works in the area (article 5, after article 6 of the Delft model). The cell can take such a decision without an application (Cell::decide_ex_officio), but the Wilgendaal case does not use it yet. The Awb of the PoC revokes only a decision reversed on objection (7:11, herroeping) and one annulled by the court (8:72), and the Awb has no general article on revoking a decision, so a revoked permit would still count. The same holds for a company that moves out of an area: its permits count until they expire. Where does it belong that a revoked decision no longer holds: in the ordinance that grants the power, read by the municipality’s own reduction, or in a general rule beside the Awb? The synthetic law of the cell’s BDD suite tries the first: its withdrawal is a decision taken ex officio that declares revokes. | Choice | High | 048, 050 |
The Kamer van Koophandel records its registrations as a fact of a fictitious policy of its own, not of the Handelsregisterwet 2007. That law is not in the regelrecht corpus. A source may hold its own version, and one under the same $id here would shadow it, because the regelrecht corpus takes precedence over a source. The number of people working in a company is not in the law either; an order in council may add it to the register (article 17, first paragraph, under a), and that order, the Handelsregisterbesluit 2008, is not in the corpus. Should the Handelsregisterwet move into the central corpus, with the facts the Chamber records as requires on its own articles? | Choice; investigation | Medium | 051 |
Only policy may bind an open term to a cell’s answer (source.reduction in UITVOERINGSBELEID). The implementing decision of Delft that the Wilgendaal case follows lays down further rules under the ordinance, so as a legal instrument it is not policy. The schema has no layer for further rules of the college, so the Wilgendaal version is a GEMEENTELIJKE_VERORDENING, and a separate fictitious policy of the college binds its terms to the answers of the cells. The implementing decision carries a gemeente_code, as the ordinance does; the policy carries none. A cell passes no code to the engine, so the engine lets a rulework of the same municipality fill the terms of the ordinance whatever the scope. Should further rules get a layer of their own, and should the scope come from the cell? | Choice; schema change | Medium | 048 |
| The Delft model also caps the number of permits per area across all companies, with a waiting list and priorities, and lots among equal applications (articles 5 and 6 of its implementing decision). The Wilgendaal case leaves that out and caps permits per company only. A cap across applicants makes one decision depend on the applications of others, and on their order, which a replay has to reproduce; drawing lots does not replay at all. How a cell decides on a cap shared by several applicants is open. | Choice; investigation | Medium | 048 |
| A decision can rest on a finding of a person instead of a calculation. The fictitious parking ordinance of Wilgendaal has a hardship clause (article 6, after article 13 of the Parkeerverordening Deventer 2026). Under it, the college decides per case whether the maximum leads to unacceptable consequences. The engine asks for that finding only for a company at the maximum. The cell records the finding among the fields of the decision, as a given beside what the Kamer van Koophandel answered. The chronicle does not record why the college decided so. The grounds of the objection stay in the objection, and the weighing itself is recorded nowhere. Yet Awb 3:46, and on objection 7:12, requires the decision to rest on sound reasons. A replay of the decision takes the recorded finding. Without it the engine refuses, because the term is required and nothing fills it. Should the chronicle keep the reasons with the finding? And should a replay under a changed rule reuse the finding or ask for it again? | Choice; investigation | High | 046, 050 |
| Could the college depart from the maximum without a hardship clause? Without one, the refusal on the maximum is a bound power in a regulation below formal law. ABRvS 2 February 2022, ECLI:NL:RVS:2022:285 (Harderwijk), gives the test of proportionality under Awb 3:4, second paragraph, for decisions with discretion. It leaves bound powers open (considerations 7.1 and 7.2), so the answer does not follow from it. If the college can, a cell needs a way to record a departure from a bound power that no open term foresees, as the row on Awb 4:84 asks for a policy. | Choice; investigation | Medium | 045, 050 |
Scenarios get the data a law reads from outside through the DataSourceRegistry. The caller registers rows, and the engine finds a value by the name of the input and the parameters of the article. The demo builds those rows from corpus/demo/bindings.yaml, which also says per input whether a missing row means unknown, null or zero (RFC-036). Should a scenario build a chronicle instead, with a fixture cell that gives the answers the test needs? The engine would then fetch outside data in one way only, by asking a cell, and the registry and the bindings could go. Today a plain law input cannot ask a cell at all: only policy binds a term to an answer. Which cell answers such an input, and under what name, is open; names of inputs clash between laws. A fixture cell can also answer what no real cell gives. Its answers would have to be checked against the catalogue in cel.yaml of the cell it stands in for, and each real cell still needs scenarios that run through its chronicle and its reduction. | Choice; large refactor | High | 048 |
A source on an input is an object of ten optional keys, and the key that is present says what kind of source it is. A series of dependencies rules in the schema forbids the combinations that make no sense; the Source struct of the law model holds only Options. Five kinds came with chronolexography: decisions, article, facts, reduction and projections; since schema v0.8.17 projections replaces decisions, and since v0.8.19 facts as well. Each had to fit beside the keys already there, which is where most of the exclusion rules come from. Two forms do not say what they are. source: {} is a value the caller supplies (see the row above), and {output: x} without regulation is an output of another article of the same law. Should source get a required kind, as in source: {kind: reduction, ref: aantal_geldige_bedrijfsparkeervergunningen, parameters: {kvk_nummer: $kvk_nummer}}? The kinds would be data, regulation (without ref for the law itself), article, reduction and projections, with decisions and facts kept only for files on the versions that allowed them. In the law model a tagged enum would then make an invalid combination impossible to write. The change breaks every rulework, about 1,090 sources in some 170 files in this repository, and corpus repositories elsewhere have to migrate with it. | Choice; breaking schema change | High | 048 |
Awb 4:2, first paragraph, under c, requires an application to indicate the decision it asks for. The requests on the article that establishes the application names that decision, yet the PoC still asks for it as a parameter of 4:2. The starting point is that the engine fills the parameter from requests itself. Where no article names requests, as for an application made directly under the Awb, the applicant fills it in. How the Awb rulework links the parameter to requests is open. First, a new element like produces.moment names the parameter; the cell then fills it, not the engine. Second, the parameter becomes an input whose source points to that requests. The engine resolves that source, and the input cannot fall back to what the applicant enters. Third, a context variable such as referencedate adds a name to the law format that no rulework declares. | Choice; schema change | Medium | 045 |
A decision taken ex officio (Cell::decide_ex_officio) rests on no submission. Replaying a decision starts from the submission it was taken on, so the cell cannot explain such a decision or give its value in a copy, and an objection against it cannot reach the decision value. Should a replay take the fields the decision holds as supplied, as the cell did when it took the decision? | Choice | Medium | 046, 053 |
| When the court decides the case itself (Awb 8:72, third paragraph, under b), its ruling holds the decision it puts in place. The receiving cell records the receipt, not that decision. In the PoC a step of the reduction reads the decision in place from the receipt, so leave the court grants counts as granted leave, and a card it grants makes the cell silent on a valid card, also after that card has expired, so from then on the college judges. What that decision does with another, such as revoking a grant of the authority, is an effect on a record the receipt does not refer to, so that part is not expressed. Should the receiving cell record the decision in place as a record of its own, with its effects, or should a receipt be able to revoke more than the record it refers to? | Choice | Medium | 053 |
A court may annul a decision in part (Awb 8:72, first paragraph). revokes takes a whole record. A partial annulment is then a revocation of the whole decision and a record of the part that stays, such as the new decision the authority takes on the court’s order. The language has no partial revocation, and nothing yet asks for one. | Choice | Low | 053 |
A step calls a rulework with source.regulation, and that rulework may not be the one whose article asked the cell for the answer: the engine is still resolving it, and refuses. The PoC therefore keeps what one decision means in a policy of its own beside the policy that binds the term. Should the engine allow it, or is the separate rulework the better place anyway? | Choice | Low | 048 |
Cell::explain_answer replays a decision and runs the reduction again, while the trace of the decision now holds the same run under the answer. The PoC builds its account of each answer on it, by how many times the decision asked the cell that answer, which the nested trace does not number. Should it go, and should the PoC read the nested run instead? | Choice | Low | 048 |
A step reads $record.<field> only as the law of the record’s article declares it. A decision holds every parameter of its article, so a step reads a date the article declares itself, such as besluit_datum on article 2 of the parking regulation of Wilgendaal. A field that only a hook or a stage of the procedure declares is in the record as well, but a step cannot read it. So the PoC cannot end a provisional measure under Awb 8:85 lid 2 onder a, because the period for appeal runs from the stage BEKENDMAKING. Should the loader take the procedure into account, as the cell does when it derives what a decision holds? | Choice | Medium | none |
| When a reduction loads, the loader checks its steps against the newest version it holds of each law they name, and the play checks them against the versions in force on the calculation date. A step calls the law in the version in force on the day of the record, so one version of a reduction may meet several versions of a law. After a version that renumbers the article or adds a required parameter that the output reads, the loader and the play refuse the reduction, even where every record predates that version. The loader does so as soon as that version is loaded, before it is in force, so whether the reduction loads depends on whether it arrives before or after that version. Should the loader check each version of the law in force while the reduction version is? | Choice | Medium | none |
| A step reads only an output that reads no input and no open term of its article. An override or a hook from another regulation can still change that output when the step calls it. Should the loader refuse a step whose output another regulation overrides? | Investigation | Low | none |
In order of impact, as of 27 October 2026. A question that waits for a case to need it comes last.
policy_first and the fallback to a finding are removed; a departure in a single case is the Awb 4:84 question. Working out what is asked of whom moves out of the cell. In progress; it changes RFC-048 §2.After that: release notes for the schema. The Awb questions on 4:1 with 4:5 and on the motivation requirements belong there too (open questions of the Awb procedure). Later: one question to several cells of a category, and a profile of its own for what serves a cell.
An exploration into transparent, executable legislation, and one of the three projects in the starting selection of the Nederlandse Digitale Dienst.
GitHub repository
How it works
Demo walkthrough
Stay informed
Roadmap (Dutch)
Documentation
Research
Basiswerk recht (Dutch)
Nederlandse Digitale Dienst
Ministry of Economic Affairs and Climate Policy