Questions
The things firms actually ask.
Including the ones with unflattering answers. Every question here has been asked by a real person considering this, and the answers are the ones we give in the room.
What it is · Safety and security · Ethics and the Bar · Cost and billing · Your data · Getting started · Limits and gaps
What it is
What is Lexicera?
Lexicera is case management for a small litigation firm, built on a file that reads itself. Intake with conflict and limitations checks, the document file, deadlines computed from court rules, discovery and productions, drafting, a trial notebook, and time and billing all live in one place. What separates it from the other systems that put those things in one place is what happens to a document when it arrives: it is read once and turned into individual facts, each carrying the exact quote and page behind it. Every later answer, chronology, contradiction check and first draft is built from those facts rather than by reading the file again, which is what makes them cheap enough to use on an ordinary Tuesday.
Is this case management software or an AI tool?
Both, and that combination is the whole argument. An AI tool you run alongside your practice software gives you a second place to look and a second thing to keep in step. Practice software with AI bolted on gives you a summarizer next to a document list. Here the reading is underneath everything: the same facts that answer a question also build the chronology, flag the contradiction, feed the first draft, and carry the citation into it. It is intended to be the system the firm runs on, not an addition to one.
Who is it for?
Small litigation firms, in Florida state and federal practice. The deadline engine ships with a pilot subset of the Federal Rules of Civil Procedure and the Florida Rules of Civil Procedure, counted under FRCP 6(a) and Fla. R. Jud. Admin. 2.514, and the data says so about itself. A firm outside those jurisdictions would get everything else and no computed dates until its own rules were encoded. A firm that holds client money in trust needs a trust accounting system alongside this one, because there is not one here.
What does it replace?
The shared drive, the matter list, the spreadsheet of dates, and most of the work you do by hand between them. In practice that means: documents scattered across an inbox and a drive, the chronology you build the week before a hearing, the deadline calculation done with the rules open beside you, the privilege log typed at the end of a production, and the Friday afternoon spent reconstructing your own week from memory. If your firm runs on a practice-management system for matters and documents and time, this is meant to be that system.
What does it not replace?
Your accounting system, your trust account, your card processor, your e-filing portal, your case-law research service, and your contact database. There is no IOLTA ledger and no three-way reconciliation; there is no card payment of any kind; nothing here files to a court or scrapes a court portal; citations are resolved against a free public opinion database for existence and accuracy only, which is not research and never tells you whether a case is still good law; and parties are recorded on the matter they belong to rather than in a firm-wide address book. It also does not replace a client-facing portal today: the groundwork for one is built, but there are no screens a client can reach yet.
Can I try it with one matter?
Yes, and that is the test we would rather you ran than a demo. Pick one closed case with real volume in it, and we run it end to end and show you what came out — the reading, the findings, the chronology. Before anything is uploaded we agree the handling in writing: what we may look at, that it is used only to show you the product, and that it is returned or destroyed when you say so. That first run is free, and it is still fully metered: the charges post and a founder credit posts beside them, so at the end you are looking at a real ledger of what the work cost rather than at a blank one. Above the free run there is a $350 pilot — one closed case of up to about 5,000 pages, the same end-to-end run, with the searchable copies of your scans handed back. There is no self-service sign-up; a firm is provisioned by a command we run, so starting is a conversation rather than a form and a credit card.
Safety and security
Where are my files?
Your documents are in cloud object storage, in a bucket belonging to your firm alone, as ciphertext. Each file is encrypted on our server before it is sent, with a key unique to that file, which is itself wrapped by a key belonging only to your firm; the storage vendor holds bytes it cannot read. Everything around the documents — matters, parties, facts, deadlines, the audit trail — is in a PostgreSQL database on a single server in Atlanta. No plaintext document is kept on that server, but be precise about what that means: to lift the text out of a file, the server decrypts it to a scratch file first. That copy is readable only by the service account, it is deleted as soon as extraction finishes or fails, and the whole scratch directory is emptied hourly by the operating system so a crash cannot leave plaintext sitting there. Scanned pages are rasterised in the same place and removed the same way. The one plaintext written deliberately and kept is an export bundle, produced by hand at a root shell, because handing you your files in the clear is the entire point of that command.
Who can read my files?
The people at your firm you have given logins to. One firm can never see another's data, and that is enforced at three levels: a storage bucket of its own per firm, PostgreSQL row-level security the application role cannot bypass, and firm predicates in the application code itself. The test suite asserts that a query with no firm context returns zero rows — not that it should, that it does. One gap in the first level, stated because you would want to know: the buckets are separate but the server reaches all of them with a single storage credential today. Per-firm scoped keys are due before a second firm is onboarded, and until then the bucket boundary is real while the credential boundary is not. Underneath those three, every document is sealed under a key of its own that only your firm's master key unwraps, and a document you upload is encrypted as it streams through the server, so what reaches storage is ciphertext and nothing else. There are two database roles and the difference between them is the honest part: the application connects as a role with no power to bypass row-level security, so nothing served over the web can leave your firm's boundary, while the provisioning role that can bypass it is reachable from no page at all — only from a shell on the server. There is no screen anywhere in this platform that shows the operator a client document; what he can see is queue depth, spend, and storage growth. Every read of a privileged document is written to an audit trail that a database trigger prevents anyone from editing or deleting, us included — and that trigger is unconditional, so the role that can bypass row-level security cannot edit the trail either. That is the whole of the boundary, at its real size: the operator runs that server and holds the master keys, so what this platform tells you about him is what he is shown and what gets recorded, rather than an assurance nothing in the code could keep.
Is anything stored unencrypted?
Document contents, no — with one transient exception. The files themselves are encrypted before they leave our server and are decrypted on the way to someone entitled to them, and briefly into a scratch file while their text is being extracted; that copy is deleted when extraction ends and the directory is swept hourly regardless. But the text extracted from a document page by page, the facts drawn out of it, matter titles, party names, filenames and the audit trail are ordinary rows in the database — they have to be, because that is what search, conflict checking and citation depend on. Those rows are protected by the database's own access control and row-level security rather than by a second key. Anyone assessing this should assume that a full compromise of the server is a compromise of the searchable record, not only of the metadata.
Do you have my master key?
Yes, today, and you should understand why before you ask us to change it. Your firm's master key is a file on the platform server, readable by root and by the application user; the server cannot decrypt a document for you without it. The first item on your onboarding checklist is to take your own copy of that key off the server and prove a restore actually works with it — it is an attestation, because nothing in the data can detect it, and it is the one step on that list you cannot recover from skipping. The direction we intend is an appliance in your own office holding your key sealed behind a passphrase only you have — see the answer on keeping a copy in your own office. That box is designed and not built, and note what it would and would not do: it would give you an independent copy, not take ours away.
What happens if Lexicera disappears?
You take an export and you are gone, and nothing in the bundle needs this software to read. Every document in the clear, plus the fact ledger, both timelines, the parties and the audit trail as CSV a paralegal can open, plus a SHA-256 manifest so whoever receives it can prove nothing changed on the way. The honest caveat is that today an export is a command we run for you on request, not a button you can press yourself — so if "Lexicera disappears" means the person rather than the company, ask for the bundle while asking is still possible, and keep the escrowed master key that the onboarding checklist asks you for. The appliance — a box in your own office, answered separately here — is the intended full answer to this question, and it is not built yet.
What happens if you are breached?
It depends on what was taken, and the honest version has a soft spot in it. A stolen storage credential yields ciphertext and no keys — but today that one credential reaches every firm's bucket, not just yours, because per-firm scoped keys are planned and not yet in place. A stolen database is the searchable record and the metadata, not the documents. A full compromise of the server is the bad case and we will not dress it up: the master keys live there, so documents that server can serve are documents an attacker with root can decrypt. What exists against all of that is a hardened single-purpose box, a separate master key per firm so one firm's decryption is not every firm's, and an append-only audit trail that records reads and cannot be rewritten to hide them. What does not exist yet is a written incident-response commitment with times in it; there is one operator, and you would hear it from him.
How many copies of my data are there, and are restores tested?
In effect, today, there is one copy of your documents, and here is the precise state behind that sentence. The second copy (immutable, at a different vendor, under object lock) and the third (deep archive) are built and their round trips verified — but the schedules that would run them nightly are deliberately off until the first real client document is uploaded, which is a hard gate written down in the project's own issue tracker precisely so it cannot quietly slide. So the honest present-tense answer is still one copy, with the other two one command away rather than a build away. The only restore proved so far is the per-firm key restore that onboarding makes you perform. If you are reading this and it still says one copy, ask before you send us anything.
Ethics and the Bar
Is using this consistent with Florida Bar Opinion 24-1?
It is built to be, and the concrete answer is more useful to you than an assurance would be. Everything the model produces that has legal consequence stops in a review queue until an attorney approves or rejects it, with a name and a time on the record. The engagement letter drafter is instructed to disclose AI use and the review that goes with it. AI and storage costs appear as their own invoice lines and are never folded into a fee for time. Time entries are proposed from timestamps and bill nothing until you accept them. Whether your own use of it is consistent with the opinion remains your judgment — supervision, confidentiality and candour are yours and cannot be delegated to software — but the compliant path here is the default one rather than the one you have to remember.
Is my client's data used to train AI models?
No. There is no training pipeline anywhere in this system — nothing collects, exports or retains your material for model improvement, and doing it would mean building something that does not exist. Documents go to a commercial API under terms that do not permit training on what is sent; they are read and the answer comes back. This is the one answer here whose proof is partly a third party's contract rather than a file in our own tree, and it is worth saying that plainly rather than letting it read as a guarantee we can enforce alone.
Does the AI make decisions?
No. Nothing with legal consequence is auto-approved, and in the places that matter most that is a constraint in the database rather than a code path someone could route around. A deadline cannot be calendared unless a human confirmed it. A client update cannot leave draft status without a named approver. A suggested time entry cannot bill until you accept it. A simulation is classified as internal decision support at the row level and can never become approved work product. Editing an approved draft revokes the approval, by trigger. Those are four CHECK constraints and a trigger, not five settings. Be precise about the one that is not: approving a draft in the first place is recorded by the application, which stamps the approver and the time, rather than refused by the schema if the approver is missing. The review queue is the only route to it and every approval carries a name — but it is code holding that, not the database.
Do I have to tell my client?
Whether disclosure is required in a given matter is your judgment under the Rules, not ours — but the product assumes you will and makes it the default. The engagement letter drafter is instructed to include two disclosures: that the firm uses AI in the work and that an attorney reviews the work product, and — where you pass costs through — that AI and storage are billed through and on what basis. Be clear about what that guarantee is: it is an instruction to the drafter, not a post-generation check that the sentences are present. What actually stands behind it is that you read and approve the letter before it goes out, like every other draft this system produces.
Who is responsible for what the AI produces?
You are. The lawyer who signs it owns it, and nothing in this system is a second opinion, a supervisor, or a substitute for reading the thing. What the product does is make supervision possible rather than nominal: every answer carries the page it came from so you can check it in two clicks, every legal citation in a draft is reported and resolved against a public opinion database — existence and accuracy, never whether the case is still good law — approvals are recorded against a person and a timestamp, and the queue exists so that approving is a deliberate act. A tool that made you feel supervised without making you check anything would be worse than no tool.
Cost and billing
What does it cost?
Three lines, and you can see all three separately. $199 a month for the firm, which includes 3 people and 100 GB of storage; $59 a month for each person after that, and your clients in the client portal are free and unlimited. AI, charged per operation — the model's own cost for that operation times your firm's multiplier, so these are what you are charged and not what it costs us: about $0.27 to read 100 pages, once ever, and about five cents for a question asked of the file afterwards. Storage past the included 100 GB is a tier, not a per-gigabyte rate. The per-page ingestion fee applies to every document that arrives in a matter, whichever way it arrived — uploaded, emailed in, or brought in by a bulk import — at the one published rate, never multiplied and the same under either processing cadence. It is charged once, as each file is read in. A recognized duplicate is free. Audio and video pay the per-hour transcription rate instead of a page rate, and a file that is not paginated — a Word document, an email, an image — counts as one page. What varies with the size of your cases stays on its own line rather than being averaged into a seat price, so you can always see which part is expensive.
How is AI usage billed?
Per operation, as it happens, drawn from a prepaid balance. Every call records three figures: the raw cost the model vendor charged, the platform cost you are charged (raw times your firm's multiplier), and what is billable to a client. The expensive operation is reading a document, and it happens once per document and never again — everything afterwards works from the facts that reading produced, which is why questions, chronologies and comparisons cost cents. The published per-operation figures are estimates computed from representative documents at current model prices, and every page that publishes them marks them that way; your documents will vary, and your ledger shows the real figure for every call the moment it happens.
Can I see what each action cost?
Yes, itemized down to the individual call. The cost ledger lists every AI action with when it ran, what it was in words a lawyer uses rather than a task code, which case it was on, who asked for it, which document it was about, and what it cost. Narrow it by period, case, person or kind of work and export the result as a spreadsheet. Work with no matter behind it — a question about the software rather than about a case — is labelled firm overhead and is never attached to a client. It is the same ledger we bill from; there is no second set of numbers.
What happens if I run out of prepaid balance?
AI pauses, and nothing else does. Your documents, your search, your downloads, the deadline engine, the calendar and everything already extracted keep working exactly as they were; a firm that runs out of credit does not lose access to its own files. Because everything draws against a balance you funded first, nothing in this system can run up a bill you did not agree to. Once the balance falls below half of the last thirty days of AI use, the morning briefing raises it and says how many days it covers at that rate; at zero it becomes a critical signal saying which features have paused. Running out is something you see coming rather than something you discover.
Can I bill AI costs to my clients?
Yes, if you decide to and your engagement letter says so. Three choices: absorb the cost as firm overhead, recover it at cost, or recover it with the firm multiplier. Whatever you choose, recovered costs appear as their own invoice lines and are never folded into a fee for time — a client is entitled to see which is which, and folding a machine cost into an hourly rate is the thing the Bar opinion is most direct about. One honest gap: there is no settings screen for that choice yet. You tell us and we set it on the firm record, and your onboarding checklist carries it as a decision to be made before the first invoice rather than as a switch to flip.
What does storage cost?
Storage comes in tiers, and the first 100 GB are included in the platform fee. Above that: Practice, 500 GB for $79 a month; Firm, 2 TB for $199 a month. Larger than that and the honest answer is a conversation rather than a table, so ask. Tiers rather than a per-gigabyte rate on purpose: a firm billed by the gigabyte hesitates before adding a document, and a case file with hesitation in it is worth less than a complete one. Going over your allowance charges you nothing and blocks nothing — we come and talk to you about closing out finished matters. The two one-time fees — $75 to close a case out to cold storage and $50 for an export bundle — are flat and are multiplied by nothing.
What does a pilot cost, and what happens during one?
$350 for one closed case of up to about 5,000 pages, above a free first run. The order is deliberate. First we agree the handling in writing. Then you hand over the case as it sits — a directory of scans, however it came off the scanner — and it is brought in overnight: encrypted on the way in, duplicates skipped, every page read, findings and a chronology built, and searchable copies of the scans handed back to you to keep. The free run is fully metered even though you are not charged for it: the work posts as charges and a founder credit posts against them, both visible, so the last thing we do together is open your cost ledger and read what it actually cost. Bringing a case in after that is $0.02 a page — about $100 for a case that size — and it is multiplied by nothing.
What happens to my costs when a case closes?
Nothing, today, and that is a gap rather than a policy. The design is that closing a matter out moves its documents to cold storage, which frees the space against your tier allowance, for a one-time $75 charge. Two things stand between that and you: the cold storage leg has no destination configured yet, so archiving is switched off, and there is no screen that marks a matter closed in the first place, so the close-out path cannot be reached from the interface even though the machinery behind it is built and tested. Until both are finished, a finished case occupies the same allowance as an active one.
Your data
Can I export everything, and in what format?
Yes, everything, in formats that need nothing from us to read. An export bundle is every document in the clear under its own filename, plus the fact ledger, both timelines, the parties and the audit trail as CSV, plus a manifest of SHA-256 hashes so whoever receives it can prove nothing changed in transit. There is no proprietary container and no step that requires this software. An export that only another Lexicera install could open would not be an exit path, which is the whole point of the format choice.
Can I leave?
Yes, and the software is deliberately built so that leaving is a file transfer rather than a negotiation. Take the export bundle — every document in the clear plus the ledger around it as CSV plus a hash manifest — and it is complete and readable without us. The one friction worth naming is that the export is a command we run on request rather than a button you press yourself, at the flat $50 fee per bundle. Whatever commercial terms sit between us live in your agreement with us and not in the software, and nothing in the software is designed to make going somewhere else expensive.
What happens to a matter when the case closes?
The design is that only the documents move. Closing out archives the file blobs to cold storage and leaves the facts, both timelines, the parties, the deadlines and the audit trail live — so an archived matter is still searchable, still answers questions, and still appears in every conflict check you run afterwards. Opening an archived document again is a restore measured in hours rather than a click, and the hot copy is removed only once the cold copy has been read back and found to exist at exactly the expected byte count, on top of the transfer tool's own integrity check. The manifest describing the archive is written, read back and hash-verified; the individual documents are checked by length rather than re-hashed, because their seal is a hash of the plaintext and cannot be computed from the encrypted copy without pulling every byte back and decrypting it — that seal is checked again when you export. What is true today, plainly: this is built and tested and not available, because no cold storage destination is configured and no screen marks a matter closed. Until then a closed case simply stays where it is, intact.
How long do you keep my data?
Forever, by default. Nothing expires, nothing is aged out, and there is no retention job anywhere in this system — a document uploaded in year one is still there in year nine unless somebody deliberately removes it. Deletion is explicit: you ask, and it is done by hand, because there is no delete-everything button in the interface today. One thing survives any deletion by design: the audit trail is append-only at the database level, so removing a document removes the document and not the record that it existed and who read it.
Can I have a copy of everything in my own office?
Not yet. This is planned and it is not built, and the distinction matters enough to lead with it. The design is a small appliance — a computer and disks in a case, sitting in your office — that pulls your encrypted files and your archived case bundles on a schedule, holds your master key sealed behind a passphrase you alone have, and runs a restore drill every month and mails you the result. It would never hold credentials that could write to the platform, so a stolen box is a stolen box rather than a way into your file. It is intended as a fourth copy rather than a replacement for the cloud ones: the cloud copies protect you against a fire in your office, and the box protects you against us. Nothing of it exists today beyond the design.
Getting started
What do I need to do first?
Six things, and the product tracks them for you on an onboarding checklist that reads the actual data rather than asking you to tick boxes. Escrow the master key off the server and prove a restore works with it. Have an attorney read and confirm the court rulesets and limitations rules, because until that happens every computed date is treated as soft and shown faint. Set billing rates. Invite the rest of the firm. Decide whether AI and storage costs are absorbed or recovered, before the first invoice. Then upload one real matter and watch it process end to end. An incomplete checklist raises a health signal rather than sitting quietly in a document nobody finishes.
How long does onboarding take?
Provisioning the firm itself is a command that takes minutes; the checklist is what takes real time, and most of it is decisions rather than data entry. Confirming the rulesets is an attorney reading encoded rules against the real ones — an afternoon, not a week. Escrowing the key and proving the restore is an hour with somebody who knows where your safe is. We have onboarded one firm, so any confident number here would be invented; what we can tell you honestly is the list of steps and that none of them is a data migration project.
Do you migrate my existing files?
Documents, yes, and it is manual. Files go in through the same upload box you would use afterwards — yours or ours doing the uploading — and the reading pass treats them exactly like anything that arrives later, so a back catalog becomes searchable and citable the same way. What does not exist is an importer for another system: your matter list, contacts, calendar entries and historic time are re-keyed, or handed to us as a spreadsheet and loaded directly into the database on request. Email can be forwarded straight into a matter once mail routing is configured, and that piece is built but not live yet.
Do you train my staff?
Yes, in the literal sense that there is one of us and he will sit with your people. There is also a guide inside the product covering every part of it, step by step, generated from the same registry as this page so it cannot drift out of date with what the software actually does. And the chat inside a matter answers questions about the software as well as about the case, from that same source — so "how do I lock a production" is answerable in the place where you are stuck, and asking it is billed as firm overhead rather than to the client whose matter you happened to be in.
Limits and gaps
What will it not do?
It will not file anything, calendar a deadline, or send anything to a client without an attorney approving it, and those are constraints in the database rather than settings you could switch off. It will never tell you a case is still good law — citations are checked for existence and accuracy, not treatment. It records what a document says, not whether that is true. A simulated jury is a machine imagining people and is labelled that way everywhere it appears. And it will not hold your files hostage: the export bundle is complete, standard, and available on request.
Can I pay by card, or set up automatic billing?
No. There is no card on file, no self-service billing screen and no automatic renewal. The prepaid balance is topped up by hand against a check or a transfer, and posting it is a person running a command. Everything posted appears on the ledger with the balance after it, so the record is complete — but the payment itself happens outside this software.
Do you do trust accounting or IOLTA?
No, and if you hold client funds this cannot be your only system. There is no trust ledger and no three-way reconciliation. Under Florida Bar Chapter 5 that is not an optional module for a firm holding money in trust, and building half of it would be worse than not having it at all — a trust ledger you half trust is a bar complaint waiting for an audit. What this does do is bill for time and costs, assemble one invoice per client, and record what was paid against it: everything is earned when it is billed, and no client funds are held at any point. Retainers, advances and credit balances are not supported, because there is nowhere in this system for money that is not yet earned to sit.
Is there a firm-wide contact book?
No. Parties are recorded on the matter they belong to. Conflict checking does search across every matter the firm has ever had, by name similarity rather than exact match, so the part that protects you works — but there is no single place to look up opposing counsel's address, and no contact that exists independently of a case.
Does it check my citations against case law?
It checks that they exist and that they are accurate, and it is careful about the difference between that and research. Every case citation in a draft is resolved against CourtListener, the free public database of American opinions: a citation that resolves links to the opinion, and one that does not is flagged unciteable, in red, on the cite check and in Chambers — the check cannot silently pass one. Taking it out of the draft is your call; the product's job is making sure you saw it. What it is NOT is Westlaw or Lexis — there is no treatment analysis, no Shepard's signal, and it will never tell you a case is still good law. A citation that resolves can still have been overruled last year. That judgment stays with you. When CourtListener is unreachable the citation is marked not-checked rather than quietly passed, because a confident "verified" that was never actually checked is the worst thing a tool like this could produce.
Will it file or calendar anything for me?
No to both. Nothing is ever auto-calendared: the deadline engine proposes dates and shows the rule, the citation and the counting behind each one, and an attorney confirms the ones they agree with. An unconfirmed date never reaches a calendar, and no amount of clicking will make it, because a database constraint rejects the write. Nothing files to a court either — there is no e-filing integration, and nothing here logs into a court portal on your behalf, deliberately: a scraper driven by a lawyer's own court credentials is a bad idea on its own terms. Florida matters are tracked instead by forwarding the portal's service emails into the matter, once mail routing is switched on.
Can my clients log in and see their case?
Not yet. The credential type, the per-matter access table and the login function are built and tested; the screens are not. There is no address a client can visit today and no way to invite one from the interface. The design is deliberately narrow — access is one row per matter, with no way to express "all matters", so the worst a mistake could do is grant the wrong single case — and status updates are drafted and reviewed like any other AI output before they could reach anybody. Until the screens exist, a client is kept informed the way they are now.
Are Bates numbers stamped onto the pages?
Yes. When a locked production is assembled, every page is separated, stamped bottom-right with its own Bates number and any confidentiality legend, and written out both as one PDF per number and as the complete produced document — alongside the DAT and OPT load files a review platform imports and a manifest recording the SHA-256 of the bytes as produced, stamping included. That is a flat $75 per production, whatever its size. What is still NOT done is redactions: those are held as coordinates on the page and nothing here burns them into pixels, so a production containing a redacted document is refused rather than produced. Delivering a file whose redactions exist only as numbers in a table would hand over the very material somebody marked for withholding, and refusing is the honest half of shipping the rest.
Can it transcribe a deposition recording?
Yes, and the transcript is searchable and citable like any other document — each segment is filed as a page with its start and end time beside it, so a citation resolves to a position in the recording. It runs on our own server rather than at a transcription vendor, and it costs $6 per hour of recording, measured from the file itself. The limits matter more than the capability: this is automatic speech recognition, so it is a search and drafting aid and never a substitute for the court reporter's certified record. It is English-only. It does not label speakers — telling two voices apart is not something it can do honestly, and a transcript that attributes testimony to the wrong witness is worse than one that attributes it to nobody. A recording with no speech in it produces no transcript and no charge. Video also has its key frames pulled out as an index of where the picture changed; nothing watches or describes the picture itself.
Which court rules are actually encoded?
A pilot subset of the Federal Rules of Civil Procedure and the Florida Rules of Civil Procedure, and the rule data describes itself that way rather than implying completeness. Counting follows FRCP 6(a) and Fla. R. Jud. Admin. 2.514. What is not encoded is stated on the page that produces the dates rather than left to be discovered: local clerk holidays, standing orders and tolling are not modelled. A ruleset your own firm has not confirmed produces soft, unverified dates only, shown faint everywhere they appear.