The AI RFP Studio for sealed, defensible sourcing

Prep Studio, “Draft with AI” in the buyer console, is the drafting half of the studio. AI-assisted intake that drafts the scope, requirements matrix, weighted criteria and pricing schedule, reads its own draft as a hostile bidder would, and hands every artifact to you for approval. Built and verified against recorded model runs; live-model drafting switches on with the model key.

The work splits into three lanes:

Prep Studio’s definition of done is a draft that passes the publish gate on the first try. Everything that decides correctness — gates, deadlines and lateness, sealing and reveal, audit writes, access control and state changes — is code, never a model.

Each pricing submission is envelope-encrypted with its own data key, wrapped by a key the database never holds. The buyer sees that a bid exists — never its contents — until the reveal.

A scheduled job opens every envelope at once after the deadline plus grace, re-checking the effective deadline at run time so an extension can never trigger an early reveal.

After publish a deadline can only move later, every change is historized with a reason, and every vendor is notified at once. Lateness is frozen at submission time.

Nobody can open a sealed bid early — not the buyer, not an admin, not a database dump — and no vendor can be given a different deadline: there is no per-vendor deadline anywhere in the schema.

Evaluators attest no conflict of interest before scoring, see only their own scores until lock, and outliers are flagged at lock time. Blind means each evaluator scores alone: until the evaluation locks, no evaluator sees another evaluator’s scores, so no one’s first number anchors the rest.

AI never scores, ranks, or recommends an award.

That is a structural rule, not a setting. Scores are human acts, made under conflict-of-interest and variance controls, and the database rejects any score from an evaluator who has not attested.

The award has a gate of its own. An award needs a locked evaluation, a written justification, every vendor question answered, every abnormal-bid flag acknowledged, and an approver who is not the author.

Publishing is a database check, not a checklist: contact details, Q&A cutoff spacing, minimum response window, criteria weights that sum to 100 and a pricing schedule are all enforced before an RFP can go out, and it warns when fewer than three vendors are lined up to invite.

The rules keep holding after an RFP goes out. Vendors ask privately; answers publish to everyone, anonymized. Post-publish changes go out only as versioned addenda every vendor must acknowledge, and an addendum inside the notice window forces a deadline extension or a written override.

One write path into the audit log, no updates, no deletes, each event hash-chained to the one before.

No one — vendor or buyer — reads a file directly. Every read is logged and served through a signed link that expires in 60 seconds.

Uploads are stamped with the server clock and a server-computed checksum, scanned, then frozen. Late uploads are accepted only with a written reason and flagged as late.

When an award is questioned, there is one place to look. One export compiles the timeline, addenda and acknowledgments, published Q&A, submission metadata, locked scores, attestations and approvals into a single hashed document.

Vendors work on a separate host with its own sessions; a vendor identity can never become a buyer user.

Deny-by-default row security, forced on every table. A cross-workspace isolation test runs on every merge: every vendor-visible table × a foreign workspace = zero rows.

Single-use invitation links that only an explicit click consumes (corporate link scanners cannot burn them), plus step-up re-authentication before sensitive actions.

Questions arrive privately; answers publish to everyone, anonymized. There is no side channel to favor anyone through.

One subscription covers Pactwise Source and Pactwise Contracts.

• Compare the plans on the pricing page.
• Walk through a sealed RFP from package to award on the Pactwise Source overview.
• See how Source fits public agencies, K-12 school districts, healthcare sourcing teams and mid-market teams.
• Compare an RFP run by email and spreadsheet with Source, or look up terms like sealed bid and blind evaluation in the glossary.
• Manage the agreements that follow with Pactwise Contracts.
• Read what is model-driven and what is deterministic in the agent architecture docs.
• Find answers to account and billing questions in the Help Center.