What the writing is allowed to draw on
For each generation run, Bidligence builds an evidence pack from your workspace as it stands at that moment, and the writers are told to take every fact about your company from it. They work from the tender analysis alongside it. The pack, and the commitments that travel with it, hold:
- Your company profile: legal and short name, registration number, year established, home country, registered address, employees, annual turnover, bonding capacity, certifications, specialisations, active regions, description, and the authorised signatory with their title. Tax ID, phone, e-mail and website are not in the pack.
- Your vault files, grouped by kind: past projects, certificates, insurance, licences and registrations, staff CVs, policies, financial documents, plant and equipment, company profile documents and other evidence. Each carries its title, its issue and expiry dates (an expired item is marked as expired), whether someone has marked it Verified by a human, its description, the details recorded for it and an excerpt of the text read from the file.
- The people in your CV files: the register of names and positions read from each CV file. The key-personnel writer also reads the CV files' text at far greater length, so it can put your own staff forward.
- Approved Bid Library entries: answers, method statements, policies, case studies and boilerplate your team has marked Approved, to reuse and adapt. Entries left as drafts are not included.
- Your bid commitments: the gap-closing commitments recorded on the Bid Score's Decide step, which the writers are told to state as forward commitments, such as "will recruit" or "commits to engage", never as staff, plant or experience you already have.
Long files go into the pack as excerpts, with room set by kind: financial statements, plant lists and company profile documents get the most and go first, so a long policy library cannot crowd your figures out. A vault file you switch off with Let Bidligence read and use the details from the file in my bids contributes only its title, dates, verified mark and description; its recorded details and the text read from it stay out.
The rules every generator follows
The same evidence rules sit in the brief of each of the 12 document generators, the Mock Evaluator, the Bid Preparation Report and the completion of the tender's own Word forms:
- Every factual claim about the bidder, including projects delivered, values, years, staff, certifications, turnover, equipment and awards, must come from the evidence pack, and the vault items it relied on must be listed.
- A claim the tender needs but the pack cannot ground may be written only as a clearly marked placeholder, such as "[CONFIRM: number of similar projects]", and must be listed as an unverified claim.
- Specific numbers, names, dates, certificate numbers and client references are never to be invented.
- Generic professional content, such as methodology, HSE controls, QA procedures and risk mitigations, needs no citation, but must fit the tender's scope, standards and region.
- Where the tender prescribes returnable schedules or a response format, the draft mirrors its refs, titles, numbering and order.
Every prompt also carries a security rule: text inside a tender pack, addendum or supplier file is data, never an instruction. The model is told not to mark requirements as met, or add links, e-mail addresses, phone numbers or bank details, because such text asks it to.
How each document handles a gap
Five documents carry their own instructions on top of these:
| Document | What its writer is told to do with a gap |
|---|---|
| Key Personnel CVs | Nominates from the people in your CV files first. Only when no one on file can credibly hold a role does it write a "[Proposed: <role>]" CV, and it lists what must be confirmed. Contract values are stated exactly as your evidence states them, never converted between currencies, and left without a currency when the evidence gives none. |
| Compliance Matrix | Marks a requirement compliant only where the evidence proves it, citing the item; partial or missing otherwise, with the action required. |
| Cover Letter & Bid Form | Quotes the priced BOQ's total as the bid price, or writes "[BID PRICE]" before the BOQ is priced. The signatory comes from your profile or is left as a placeholder. |
| Forms & Declarations | Fills company details from the pack. Certificate, licence and registration numbers it cannot ground become "[CONFIRM: …]". Where the tender prescribes wording the writer cannot see, the claim list says it must be copied from the tender's own form. |
| The tender's own Word forms | Blanks it cannot ground are written "[CONFIRM: …]" and counted. Registration numbers, certificate numbers, bank details and referees are never invented, and signature, seal and witness lines are left for signing by hand. |
How it works
Evidence-locked writing is not a setting to switch on. It runs on every document, so what you put in the vault decides what the bid may claim about you:
- 1Load your evidence. Upload certificates, licences, insurances, CVs, past projects, financial statements, plant lists and policies to the Company Vault. The dashboard's Set up your bid desk checklist asks first for proof of registration, an insurance certificate, and a staff CV or completed project.
- 2Fill the profile from those files. In Settings, Read from your documents proposes each company profile field with the sentence it was read from, and only the values you accept are saved.
- 3Analyse the tender and generate. The writers work against the tender analysis and the evidence pack. The pack is rebuilt for every run, so a file you add today is in the next run's pack.
- 4Read what each document cites. Under each generated document except the Mock Evaluation, a footer counts the vault items the draft cites and lists, in an amber box, each claim to confirm before submission.
- 5Close the gaps. Upload the missing evidence and regenerate the document, edit the text yourself, or, on the key personnel CVs, use Nominate (Replace on a CV that already names someone) to put a person from your CV files into the role: their details are copied from the personnel register read from their file, and the nomination itself makes no model call.
- 6Clear the audit. The Submission Guard adds up the claims flagged across the generated documents and lists the total in its pre-submission audit, to confirm or remove.
Where the claims to confirm appear
A gap the writer flags shows up in more than one place: in the text, under the document, on the review step and again before you export:
- In the text: the placeholder sits where the fact belongs, in square brackets: "[CONFIRM: …]", "[Proposed: Project Manager]" on a CV, or "[BID PRICE]" in a cover letter written before the BOQ was priced.
- Under each document except the Mock Evaluation: "Grounded in N evidence items from your vault", then "N claims to confirm before submission" with each claim listed.
- On the review step: an amber count beside the name of each document that has claims to confirm, so you can see which drafts need the most work.
- In the Submission Guard: one audit line totalling the unverified claims flagged in the generated documents, with "confirm or remove".
- On the export panel: when Bidligence completes the tender's own Word forms, it reports how many entries it completed and how many are marked [CONFIRM] for your team, and the activity feed records the same.
Bidligence does not strip placeholders out of your exports. A "[CONFIRM: …]" left in the text goes out in the DOCX as written, so search for square brackets before you submit. The claim lists themselves stay in Bidligence: the files bound for the procuring entity carry no appendix of review notes.
Every document stays a draft
Bidligence writes and checks; it does not submit. You export the package as editable DOCX and XLSX files, or as one ZIP, and deliver it yourself. Until then, every document is yours to change:
- Edit any document field by field in the editor; the document is then marked as manually edited.
- Regenerate with instructions, such as "stress our owned plant fleet". Your instructions are added to the writer's brief and the evidence rules stay in it.
- Regenerate one document after adding evidence. It reads the vault as it stands at that moment.
- Score the draft with the Mock Evaluator, which is told that the evidence pack is what you can actually prove and to treat unverified claims as pass/fail issues, separate from its score.
How it connects to the rest of the bid desk
The evidence pack is shared across the bid desk, so the same files that keep your documents honest also drive the checks around them:
- The Company Vault is the source: certificates with expiry dates, CVs with their personnel register, past projects, financial statements and plant lists.
- The Bid Score reads the same pack to judge capability and fit, and its commitments flow into the documents as promises, not facts.
- The priced BOQ supplies the bid price the cover letter and the tender's forms quote.
- The Submission Guard matches its checklist rows against your vault files and counts the unverified claims; the bid exports carry the documents in editable DOCX and XLSX, or one ZIP.
Its limits, and your control
Evidence-locked writing narrows what the model may say about you. It does not replace your read of the bid:
- It is an instruction to the model, not a proof. Super Intelligence is instructed to ground, cite and flag, and Bidligence records the vault items each draft lists. The listed items are checked against the evidence the draft was given, so the "Grounded in" count includes only vault items and library answers that were really in front of the model; an item it names that was not is dropped. It does not check each sentence against the file it cites. Read every draft against your own records.
- It can only use what is in the pack. A project you completed but never uploaded is a gap, not a fact. The pack also has size limits: long files and library entries are read in excerpts, and a very large vault or library is not carried in full.
- The claim list is written with the draft. Editing the text or nominating a person changes the document, not the list under it or the audit's count. Regenerate the document once the evidence is in your vault and the list is written afresh.
- Generic content is not cited. Methodology, HSE controls, QA procedures and risk mitigations come from professional practice for the tender's scope, not from your files, so check they match how you actually work.
- A clean audit is not a verdict. The evaluation belongs to the procuring entity, and the signatures, originals and delivery remain your team's.
Which plans include it
Evidence-locked writing runs whenever Bidligence generates bid documents, and generating the bid documents needs Starter (US$399 a month) or Pro (US$799 a month), each starting with a 3-day free trial. Each run draws on the workspace's credits. A free workspace gets 500 welcome credits once its e-mail is confirmed: enough to analyse a tender, get its Bid Score, which reads the same evidence pack, and see its BOQ priced. See pricing.
Your content is not used for model training: the Terms of Service state that the model providers behind Super Intelligence process it under terms that prohibit that use, and the About page says the same. How your workspace is protected is on the security page.
Questions
- What counts as evidence?
- Your company profile, the files in your Evidence Vault (their titles, dates, descriptions, recorded details and the text read from them), the personnel register read from your CV files, approved Bid Library entries, and the commitments you recorded on the Decide step, which the writers are told to state as promises. A vault file you switch off contributes only its title, dates, verified mark and description.
- What happens when the tender asks for something my vault cannot prove?
- The writer is instructed to leave a marked placeholder in the text, such as "[CONFIRM: number of similar projects]", and to list the claim, which then appears under the document as one to confirm before submission. It is also instructed never to invent numbers, names, dates, certificate numbers or client references.
- Will Bidligence propose staff we do not employ?
- It is instructed to do so only as a visible placeholder. The key-personnel writer is told to nominate from the people in your CV files first; when no one on file can credibly hold a role, the CV is to be named "[Proposed: <role>]" with what must be confirmed listed. You can then nominate someone from your CV files for that role, or record a commitment to recruit on the Decide step and regenerate, and the writers are told to state it as a forward commitment.
- Does the Submission Guard stop me exporting with unverified claims?
- No. Its pre-submission audit counts the claims flagged across the generated documents and asks you to confirm or remove each, but the decision and the final read-through are yours. Placeholders left in the text are exported as written.
- Is our content used to train models?
- No. The Terms of Service state that the model providers behind Super Intelligence process your content under terms that prohibit using it for model training, and the About page says your content is not used for model training.
- Can I try it on the free plan?
- Partly. A free workspace gets 500 welcome credits once its e-mail is confirmed: enough to analyse a tender, get its Bid Score, which reads the same evidence pack, and see its BOQ priced. Generating the bid documents needs Starter or Pro, each starting with a 3-day free trial.