A compliance matrix that quotes the answer to every shall, and is ready again the morning after an amendment
Every shall answered and cited before red team, so no bid is lost on compliance.
See one case, screen by screen ↓A compliance matrix lists every requirement in a solicitation, its source paragraph and the part of the proposal that answers it, so the team can show nothing is missed. RFP Compliance Matrix has eight agents build and keep it: they shred the package into every shall and must, map each one to its Section M factor and Section L instruction, quote the draft passage that answers it and replay every amendment. The capture manager approves each matrix version.
How to choose AI solutions for back-office and enterprise work →
Updated
A government solicitation runs to about a thousand pages across Sections C, L and M, a dozen attachments and every amendment the contracting office issues, with about 400 shall, must and will statements. A proposal coordinator shreds it by hand into a spreadsheet, maps the rows to the Section M factors and then hopes the volume leads answer them. Compound sentences hide two requirements, and Government “will” statements get mixed up with the offeror’s.
The matrix then says “done” when the draft does not answer. Section owners mark rows complete and nobody re-reads the draft against every row before red team. When an amendment lands two weeks before the due date it adds requirements, cuts page limits and moves dates, and the shred starts again. A proposal that misses a material requirement can be found unacceptable, and when the Government awards without discussions there is no chance to fix it.
Estimated split for one large solicitation with one late amendment, by hand and with the solution.
Eight agents shred the solicitation, map every requirement to Section M, re-check the draft and replay each amendment; the capture manager approves the matrix.
Kestrel & Vane Federal Services is bidding W912-26-R-0041, Base Operations Support Services at Fort Halvorsen: about $186M, 412 requirements across 14 documents, proposal due Tue 10 Nov 2026 at 14:00 ET. Amendment 0003 was posted at 16:40 yesterday. Here is how capture manager Dan Ruiz works with the agents the next morning, screen by screen, in the working solution.
The agents replayed Amendment 0003 overnight and re-read the whole draft against all 412 requirements. Dan’s screen shows 371 addressed, each one quoted from the draft, 41 to close before red team (9 missing, 32 partial) and red team in 7 days. His next step is one bar: approve matrix v4. Below it, five agents report what they did between 16:52 and 06:58, each with its sources.
Amendment tracker: “Amendment 0003 adds 6 requirements, changes 2 and removes 1. It also cuts Volume II to 45 pages and moves the due date to 10 November.”
Dan types the question he would ask his proposal manager: “Are we at risk of being found non-compliant?” The compliance checker answers with its sources: the amendment is not acknowledged, Volume II is 3 pages over and those pages would not be read, and 9 requirements have no answer. It offers the next step as a button, Close the gaps.
“The Government intends to award without discussions, so there is no second chance to fix them. All three are fixable before red team.”
The crosswalk puts Section L’s instruction (L.4.2) beside Section M’s criterion (M.2.2) for Factor 2, Technical Approach, 35% of the non-price score: 196 requirements, 171 addressed, 19 partial, 6 missing. Dan opens C.6.4.2-c, annual backflow testing. The factor mapper put it in 2.2, Utilities operations, 91% sure. The compliance checker, which had no part in that mapping, re-read the draft and found no passage that answers it.
Compliance checker: “The outline maps this to 2.2.3 and the section owner marked it done, but 2.2.3 only covers cross-connection surveys. Nothing on annual testing or records.”
Dan asks the agents to draft it. The gap planner writes a start from Attachment J-2, table 6.4, and the work order system’s standard job BF-01: “We test all 1,406 backflow prevention devices each year on a rolling schedule, and keep each test record for three years in the work order system under standard job BF-01.” One button sends it to Ji-woo Park, who owns section 2.2.
“The owner edits and saves it into the volume; I re-check it then.” The gap planner never writes into a volume.
The tracker compared Amendment 0003 with the approved matrix, paragraph by paragraph. Six new requirements, none answered in draft v0.9 and four of them in Factor 2. The Volume II limit is cut from 50 to 45 pages while the draft is at 48. One paragraph is removed, freeing 0.3 page in 2.1, and the due date moves to 10 November. Each change shows where it lands in the draft: “Vol II 2.3 · draft says 2 hours · Missing”.
“You approve the changes, not the whole matrix again.”
The trace shows what was done while the team slept: saw Amendment 0003 on the opportunity feed at 16:52, read the SF 30 and 11 pages, compared every paragraph with matrix v3, mapped the 6 new requirements to Section M factors (92 to 99% sure), re-read draft v0.9, re-checked the format against the new limit, found a conflict and drafted question 4, and moved the plan to the new due date, with red team staying on 21 Oct.
The agents check the opportunity feed every 30 minutes and start a replay within an hour of a new amendment.
Matrix v4 is v3 with 6 added, 2 changed and 1 removed: 412 requirements, every change cited to its paragraph, no mapping below 90%. Dan picks reason code AP-01, “Matches the solicitation as written”, and ticks that he has read the cited passages. The 9 open gaps are not part of this approval; they stay on the gaps list. The acknowledgement itself waits for Helen Okafor, who signs the SF 30 for the company.
“Signed as Dan Ruiz, Capture Manager · Wed 14 Oct 2026 · recorded in the audit trail.”
Ranked by risk, pass-or-fail first and then Section M weight, the list opens with the unsigned acknowledgement, which goes to Helen Okafor to sign. Next is A3-01: report any cybersecurity event in the installation’s facility control systems within 1 hour. The checker is 96% sure the draft does not answer it. The gap planner proposes the place (Volume II, 2.2), the owner (Ji-woo Park), a three-step plan and a starting text from cyber policy CP-7 and the Halden Air Base monthly report. Dan can assign it, edit it, or say it is not ours to answer.
“The owner writes it into the volume. The checker re-reads the section when it is saved.”
Volume II is 48 pages against the new 45-page limit, so pages 46 to 48, which hold 2.5.3 to 2.5.5, environmental and safety, would not be read. The format checker proposes 3.0 pages of cuts: repeated workload tables in 2.1 (1.4 pages, already in Attachment J-2), the corporate history opening 2.1 (1.1, also in Volume I) and a second photo in 2.2 (0.5). Every other Section L.3 check is met, from fonts and margins to no price outside Volume V.
“These cuts remove no answer to any requirement.” Marcus Lindqvist decides what is cut.
The new C.7.4.3 asks for snow removal to start within 30 minutes, but C.7.4.1 still defines the snow season as 1 November to 31 March, and Attachment J-2 records 3 October snow events in the last five years. The conflict finder drafts a neutral question for the contracting officer and, when asked, explains how the team would answer it either way. Questions close on Mon 26 Oct.
The conflict finder never sends a question. Dan decides whether to ask, and Helen Okafor sends it.
The same bid, seen by four of the people who carry it — what amendment week looked like, and what it looks like now.
Eight agents shred the solicitation, map every requirement to Section M, re-check the draft and replay each amendment; the capture manager approves the matrix.
Reads every document in the package, finds each shall, must and will, and splits compound sentences into one requirement each, cited to its paragraph.
Maps each requirement to its Section M factor and sub-factor, its Section L instruction and the outline section that answers it, with a confidence.
Watches the opportunity feed, compares each amendment with the approved matrix, proposes the new version and moves the plan to any new date.
Re-reads the draft volumes on each save and marks a requirement addressed only when it can quote the answering passage with its page.
Checks page limits, fonts, margins, file names and sizes and keeps price out of the other volumes, then proposes cuts that remove no answer.
Finds contradictions across sections, attachments and amendments, explains both ways to answer, and drafts a neutral question for the contracting officer.
For each requirement that is missing or partial, proposes where to answer it, who should write it and a starting text drawn only from approved company sources.
Builds each colour team reviewer’s packet from the matrix, folds comments back onto their rows and gives a labelled forecast of likely deficiencies and weaknesses.
Approves each matrix version. The agents propose; a named person decides.
Every screen has a place to ask or instruct the agents. Answers cite the solicitation and the draft, and anything they change goes to a person first.
Are we at risk of being found non-compliant?
Today, yes, in three places. The amendment is not acknowledged; Volume II is 3 pages over, and those pages would not be read; and 9 requirements have no answer. The Government intends to award without discussions, so there is no second chance to fix them. All three are fixable before red team.
Which ones were marked done but are not answered?
Three. C.6.4.2-c, backflow testing: 2.2.3 covers surveys, not tests. C.8.2.5-b, 24-hour vehicle recovery: 2.4.2 is daytime only. L.4.1.2-d, signed letters of commitment: five résumés, no letters.
Which new requirements land in Factor 2?
Four of the six: A3-01 cyber monitoring in 2.2, A3-02 the 30-minute snow start in 2.3, A3-03 cooling plant spares in 2.1 and A3-04 the PFAS inventory in 2.5. Factor 2 carries 35% of the non-price score. A3-05 sits in Factor 1 under 1.3, and A3-06 is the acknowledgement itself.
What should the red team look at first?
Factor 2, where the forecast sees 2 deficiencies: A3-01 and A3-02, both new response-time requirements. Then the 6 open pink team comments, mostly response times in 2.2 and 2.3. I have put both at the top of each Factor 2 packet.
13 screens from the working solution, on its sample data. Pick one to see it large.
How the agents read the solicitation and when a person must step in are settings. A change takes effect only after a second person approves it.
| Setting | Default | Choose from |
|---|---|---|
| Words that make a requirement | shall · must · will (offeror only) · is required to · is responsible for · at a minimum | Add your own words; compound sentences are split into one requirement each |
| “The Government will …” | Context only | Kept as context, not as the offeror’s requirement |
| Accept a mapping without a person at | 90% | Any value from 80% to 99%; below it the proposal manager settles the mapping |
| Read a scanned page without a person at | 85% | Any value from 70% to 99%; below it a person confirms the page |
| Amendment replay | Always on | Re-read, re-map and re-check on every new amendment |
| Independent check | Always on | The checker must quote the draft for every “Addressed” |
| New amendment posted | Replay within 1 hour | The capture manager and contracts are told |
| Amendment not acknowledged | Alert 5 days before due | Contracts and the capture manager are alerted |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample solicitation · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (a 2025 RFP benchmark survey of 1,544 respondents, run by an RFP software vendor with APMP — 153 RFPs a year at about 25 writing hours each). Regulation named as context: FAR 15.204-5, 15.206, 15.304 and 52.215-1 on Sections L and M, amendments, acknowledgement and award without discussions. People, companies, solicitations and installations named on this page are fictional — characters and sample data in the working solution.
A compliance matrix lists every requirement in a solicitation, its source paragraph and the part of the proposal that answers it, so the team can show nothing is missed. RFP Compliance Matrix has eight agents build and keep it: they shred the package into every shall and must, map each one to its Section M factor and Section L instruction, quote the draft passage that answers it and replay every amendment. The capture manager approves each matrix version.
The requirement extractor reads Sections A to M, the attachments and every amendment, and finds each sentence with a requirement word such as shall, must or will (offeror only). Compound sentences become one row each, “the Government will” statements are kept as context, and every row cites its paragraph. A scanned page it cannot read with confidence goes to a person.
Yes. The factor mapper puts each requirement under its Section M factor and sub-factor with the factor’s weight, beside the Section L instruction and the outline section that answers it. Mappings below 90% confidence go to the proposal manager to settle.
A separate agent, the compliance checker, re-reads the draft volumes each time they are saved. It marks a requirement addressed only when it can quote the answering passage with its page, and it never checks its own mapping. Rows an author marked done but the draft does not answer go back to Missing with the reason.
The amendment tracker starts a replay within an hour of the notice on the opportunity feed. It compares the amendment with the approved matrix paragraph by paragraph, re-maps what changed, re-checks the draft and the page limits, moves the plan to any new date and briefs the capture manager. The capture manager approves the changes as a new matrix version, and contracts signs the acknowledgement.
No. The agents never write into a volume, acknowledge an amendment, sign for the company or send a question. The gap planner proposes a starting text from named company sources for the owner to edit, and the conflict finder drafts questions that the capture manager decides on and contracts sends.
Yes. Nothing an agent prepares takes effect until a person decides, with a reason code, and no one approves what they drafted. The capture manager approves each matrix version, packet and question, the contracts manager signs for the company, volume leads accept their gaps, and every step is on a sealed audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — a past solicitation with its amendments, your draft volumes and your approved content — and it goes live once every quality gate has passed. We will walk you through it on one of your own solicitations first.
Questionnaires back in days, not weeks, and every answer matches what customers were told.
1-min demoRFP Response AutomationHigher win rate, no promise you can't keep, and the bid in on time.
1-min demoDue Diligence Questionnaire (DDQ) ResponseEvery DDQ answered from approved sources, and no investor told two different stories.
1-min demoWe'll run RFP Compliance Matrix on one of your own past solicitations, its amendments and your draft volumes.