FBI Image Capturing RFI: Which NARA FADGI Requirement Is Named

Date: August 30, 2026 · Author: Dmitrii Zatona  · Updated: September 29, 2026

TL;DR

• In July 2026 the FBI sought market input on a platform to scan and process its paper records at 100,000 images on an average day, and named one image-quality standard in one phrase (Sections 1 and 2).
• The phrase asks for “the temporary and permanent NARA FADGI digitization requirements,” and no resolution, bit depth, format or star level appears in the notice, its RFI or its thirty-two answered questions (Section 2).
• For permanent paper records and photographic prints, 36 CFR part 1236 Subpart E already fixes FADGI three-star parameters, formats, metadata and quality control, but only for reflective digitization, a limit the RFI does not state (Section 3).
• For temporary records the phrase has three readings: a literal one with no FADGI object behind it, a shorthand for two regulatory regimes, and a deliberate FADGI-class requirement whose star level and material class are unselected (Section 3).
• The requested automated quality checks, corrections, sensitive-data detection and OCR carry no metric, bound or threshold in the notice, and the regulations and standards reviewed supply none for their automated forms (Sections 4 and 8).
• The documents do not show which reading the FBI intended, and the analysis draws no conclusion about intent, vendors, or any later acquisition (Section 8).

Digitizing federal records to NARA’s requirements means different things for permanent and for temporary records, and a July 2026 FBI market-research notice asks for both in one phrase. The notice seeks a platform to capture and process the Bureau’s paper records: scanning, recognition, metadata, workflow and sensitive-data detection, at a stated 100,000 images on an average day. Across the notice description, a ten-page RFI and thirty-two answered questions, FADGI is the only image-quality standard named. No resolution, bit depth, color mode, file format standard or star level appears in any of the three documents.

The image-quality side of the requirement therefore rests on one compound phrase whose halves behave differently. For permanent records, a regulation chose the parameters years ago and attached a scope limit the notice never mentions. For temporary records, the phrase admits three readings that lead to three different objects, and no reviewed document says which was meant. This article takes the phrase apart into the six links that would make it checkable and shows which links the published documents fill. A market-research notice is entitled to leave choices open, and the subject here is what each open choice involves.

1. What was published

The notice is FBI-IMD-IC , “RFI for FBI Image Capturing,” which the FBI’s Finance and Facilities Division, Procurement Section, posted to SAM.gov on July 9, 2026. SAM classifies it as a Sources Sought notice. The attached ten-page document calls itself a Request for Information and cites FAR 15.201(e) in its disclaimer. The requiring organization named in the description is the Information Management Division (IMD), Technology Innovation Section. The recorded place of performance is a city only, Winchester.

The SAM record holds three versions: the July 9 original with responses due July 20, a July 14 revision that added a Q&A workbook with thirty-two answered questions, and a July 16 revision that moved the deadline to July 27. No version records a reason for a revision. The notice archived automatically on August 11, 2026. Its related field is empty, and the record shows no award data, NAICS code, product service code or set-aside. The point of contact is Brandon James, [email protected], listed on the PDF’s last page as “Contracting Point of Contact.” He is a procurement-section contact, and nothing below treats him as a program contact.

Two documents supply background, and the notice connects itself to neither of them. The FBI’s FY 2027 Budget Request to Congress  (March 2026, p. 88) describes the Central Records Complex in Winchester, Virginia, as a 256,000-square-foot facility with “the capacity to house approximately 1.7 billion pages of records.” Its robotic storage system manages more than 361,000 records storage bins. The Q&A names Sentinel as the “FBI’s Enterprise-level case management system” and as the downstream system for the captured documents and metadata (Q3). The FBI’s National Press Office announced  on July 31, 2012 that Sentinel had been deployed to all employees on July 1, 2012.

As of September 29, 2026, SAM.gov public search shows no follow-on solicitation or related notice. The searches covered the FBI-IMD and FBI-IMD-IC identifiers, the product name Captiva, the term FADGI, and image-capture phrasing filtered to FBI offices. That negation is limited to what SAM public search shows, because a next step can proceed through channels it does not capture. The notice is closed market research, and the timing of any next-step solicitation is not public.

This is the seventh teardown in a series; the previous one, DARPA COOP BAA: What Anchors “If and Only If the Device Physics Is Correct”, asked what anchors a biconditional guarantee about device physics.

2. What the requirement says

The document sets its own register in its introduction, which calls the list that follows “preliminary requirements envisioned for the Image Capture Suite and Document Processing Modernization project.” Its disclaimer states that responses are not offers and that the Government is under no obligation to award a contract. The capability text that follows is organized into groups, each a list of short bullets:

The table shows a full capture pipeline described by function, with every quality attribute stated as a capability and none as a number. Four passages carry the analysis. The first is the only place an image-quality standard is named (PDF p. 4, verbatim, with the bullet’s own title):

“Flexible Digitization Platform Supporting Simultaneous FADGI and NonFADGI Workflows: The system must be able to handle the capturing and processing of textual documents and photographs, meeting the temporary and permanent NARA FADGI digitization requirements and simultaneously running a non-NARA FADGI workflow”

The second opens the same group, four bullets above it:

“Automated Quality Checks: Alerts and corrections for blurred, skewed, or incomplete captures.”

Its neighbors extend the same idea. One asks for “manual and automated flagging options” to replace low-quality images during quality control. Another lists enhancement tools for brightness, contrast, noise reduction, de-speckling, blank-page detection, cropping “and other image clean up features.”

The third is the metadata bullet, “Consistent Schema Application: Uniform metadata schemas for efficient organization,” which names no schema. Asked whether schemas would be standardized across the enterprise or kept per business area (Q7), the later FBI Q&A answers that the Bureau is “striving for FADGI compliance,” which requires capturing descriptive, technical and administrative metadata. The metadata question is routed back to the same single reference.

The fourth carries the only “shall” among the capability bullets. Elsewhere in those bullets obligations are phrased as “must” or “should,” and the PDF’s submission instructions use “shall” separately:

“Automated Sensitive Data Detection: The system shall utilize Artificial Intelligence and Machine Learning to identify potentially classified or sensitive information, replacing or supplementing basic regular expression (regex) ‘dirty word’ searches with context-aware analysis to improve accuracy and reduce manual review.”

The Q&A fills in what that capability is meant to recognize: classification levels, dissemination controls, compartments, CUI categories and PII, “displayed in banner lines, portion markings, and classification blocks” (Q28). Its output is advisory: “Advisory flags. All AI-detected sensitive data would undergo human review.” (Q29). Asked whether acceptance thresholds would be set, the Q&A says they “have not yet been established” and “may be defined during requirements refinement and acceptance testing” (Q30).

One more fact about the corpus is bounded by a case-insensitive search of all three documents: the SAM description, the full text of the PDF, and every cell of the Q&A workbook. Image parameters are absent, since dpi, ppi, pixels, bit depth, bitonal, grayscale, compression, lossless, sRGB and color mode never occur. Format standards are absent too. JPEG, JPEG 2000, JP2 and PDF/A never occur, and TIFF appears once, lowercase, in the informal export list. The search also finds no EBTS, ANSI/NIST-ITL, FISWG, FIPS, RMF, ATO, NIST 800-53, CJIS, Section 508, METS, Dublin Core, EXIF or XMP. Provenance, chain of custody, integrity, retention, disposition, records schedule, hash, checksum and verification are absent as well. “FADGI” appears in one PDF bullet and one Q&A answer, with no star level and no edition. Every quantitative image-quality parameter the requirement would need therefore reaches it, if at all, through that single phrase.

3. What makes the FADGI phrase checkable

A requirement to meet NARA FADGI requirements becomes testable only when six links are filled, and different authorities fill them. The chain below is an inference from 36 CFR part 1236 and the FADGI guidelines, and no reviewed document draws it in this form.

1. The record’s category, temporary or permanent, which the applicable records schedule determines. Unscheduled records must be treated as permanent until a schedule is approved, per NARA’s FAQ on digitizing temporary records ; the reviewed documents do not say whether any are in this workload.
2. The applicable normative object, meaning which regulation or guideline governs that category.
3. The scope gate, meaning whether the material and the capture method fall inside that object’s coverage.
4. The parameters: resolution, bit depth, color, tone, sharpening, noise, formats, metadata elements.
5. The assessment method: what is measured, on what, how often, against which target and tolerance.
6. The attester: who asserts conformance, and who checks the assertion.

The diagram’s branches are an inference from the RFI phrase, Subparts D and E of 36 CFR part 1236 and the FADGI guidelines, not something the notice or the regulation defines. The solid branch is the one regulation has filled; the dashed branches are the readings no published document selects.

3.1 The permanent branch

On the permanent side the chain is filled, mostly by regulation rather than by the notice. 36 CFR part 1236, Subpart E , the subpart on digitizing permanent federal records, came from a final rule at 88 FR 28410  that took effect on June 5, 2023. It does not incorporate FADGI as a moving reference. It writes a fixed parameter set into the Code of Federal Regulations and names the source. Section 1236.50(b) bases the performance parameters on FADGI three-star aim points and tolerance ranges, and § 1236.41 describes the FADGI guidelines as the basis for many of the part’s technical parameters, which equate to the FADGI three-star level.

Link 4 is therefore closed by regulation. For modern textual paper, Table 1 of § 1236.50 sets resolution at 294 ppi or more, bit depth at 8 or 16 and sharpening below 1.1 max modulation, along with further measured parameters and their tolerances. Table 2 raises resolution to 392 ppi or more for photographic prints and paper records with fine detail. Section 1236.48 lists the formats: TIFF 6.0, JPEG 2000 Part 1 and PNG 1.2 for both tables, and PDF/A for paper records only. In its provisions for photographic prints, § 1236.48(c)(3) forbids transcoding and upsampling anywhere in the workflow. Section 1236.54 lists mandatory metadata elements in the regulation’s own vocabulary.

Link 5 is closed to a range. Quality assurance works at the device level against a target: a reference target is scanned at the beginning of each workday, and auto-correction settings that may cause non-conformance are turned off (§ 1236.46(c)). The method is FADGI’s Digital Image Conformance Evaluation  (DICE) or a documented alternative (§ 1236.46(b)). Per-file checking covers all image files for openability, format, compression and recorded attributes, with visual inspection of a sample (§ 1236.46(d)). When 1% or more of the examined records fail, the agency must find the source of the errors, correct or re-digitize, and reinspect (§ 1236.46(e)).

Link 6 might be expected to involve an outside certifier, and none appears in the reviewed corpus of guidelines and regulation. The FADGI Technical Guidelines, Third Edition  (May 2023) treat conformance as a property of a digitizing program with four components (§ 2.2). Section 2.3 states that validating or recommending targets and software is outside FADGI’s mission. Under the regulation, conformance is the agency’s own assertion, validated by staff independent of those who performed the quality-control inspections (§ 1236.56(b)).

Link 3 is where the notice and the regulation part company. Subpart E covers permanent paper records, photographic prints included, digitized by reflective techniques (§ 1236.40(b)). Transmissive originals such as negatives, transparencies, roll film and micrographic or radiographic materials fall outside it, and so do dynamic media; for those, the regulation directs agencies to contact NARA (§ 1236.40(c) and (d)). The RFI does not state that gate. Its object statement, “textual documents and photographs,” specifies neither medium nor capture technique. The Q&A describes a mixed workload of mostly textual images, with material “often intermingled with photographs, receipts, books, newspapers, magazines, pamphlets, brochures, maps, blueprints, oversized material” (Q5).

What stays open on this branch is narrow. Material has to be assigned to one of the two parameter tables. The agency chooses DICE or a documented alternative. And it chooses between the default visual sample, ten records or 10% of each batch, whichever is larger, and a statistically valid sampling plan of its own (§ 1236.46(d)(2)).

3.2 The temporary branch

On the temporary side the chain forks, and the fork is the subject of this analysis. One gate conditions the first two readings. Subpart D  applies its standards when an agency digitizes temporary records “in order to use the digitized records in place of the source records” (§ 1236.30(a)). The RFI does not say whether digitized temporary records would replace their sources, and no reviewed document settles it. Every statement below about Subpart D’s standards therefore holds only under that use. Where the use applies, § 1236.30(a)(3) also requires managing the digitized records under Subparts A through C, a records-management layer separate from the five standards. The third reading sits outside the gate, because a voluntary FADGI-class requirement neither depends on it nor is settled by it.

Read literally. Subpart D came from a final rule at 84 FR 14265, effective May 10, 2019, and was amended at 89 FR 46803 , effective August 28, 2024. Its standards are short enough to list in full, and their completeness is the point. Section 1236.32 requires an agency digitizing temporary records to (a) capture all the information in the source records, (b) include every page or part of them, (c) make sure the digitized records can serve every purpose the sources served, attesting to transactions and activities included, (d) protect the digitized versions against unauthorized deletion, addition or alteration, and (e) keep them locatable, retrievable, accessible and usable for the whole retention period.

All five are functional standards. As codified in the CFR annual edition of July 1, 2025, Subpart D contains no reference to FADGI, no star level, no resolution figure, no format list, no metadata schema and no quantitative image parameter. Validation is required and documented. The agency may set up its own validation process or use a third-party one, and NARA may review the documentation (§ 1236.34). Read against that text, the temporary half of the phrase, NARA FADGI digitization requirements for temporary records, names an object that does not exist, and the marked outcome of this reading is null. NARA’s FAQ points the same way: NARA does not recommend the permanent standards “as a default for all temporary records.”

Read as shorthand. The phrase distributes across its halves: NARA’s requirements for temporary records under Subpart D, plus the FADGI-derived requirements for permanent records under Subpart E. It is then a compact way to name both regimes. For a builder, the consequence is that one platform must run two workflows specified in different currencies. One is quantitative, parameterized and tied to a method. The other is functional and defined by the agency. The FADGI parameters still do not reach the temporary side.

Read as intent. An agency may exceed Subpart D. The same NARA FAQ says agencies may apply the permanent standards to temporary records, and an organization running one pipeline over mixed holdings has an operational reason to want one quality regime. On this reading the parameters exist and the sentence is deliberate. What remains open is what the permanent branch has closed: the star level, the material class and the assessment method. FADGI defines four star levels and “does not recommend digitization to less than a three-star level when possible” (§ 2.1). It also cautions against four-star imaging for all but the most sophisticated programs, and the Third Edition defines fifteen material classes.

Three readings lead to three different objects, and the published documents do not choose among them. No reviewed regulation or guideline defines FADGI requirements for temporary records. Neither the notice description, the RFI nor the Q&A says which reading was intended. The null belongs to the literal reading only, as its marked outcome. It is not a finding about what the FBI meant, and none is drawn here.

4. Four choices the documents leave open

The unresolved referent is the first of four open choices, and the other three sit in the capability bullets around it. Each is a place where the notice asks for a function and the reviewed documents supply no measure for it. A later contract would have to settle each one, and the settlement changes what gets built.

4.1 The referent itself is unresolved

A respondent has to pick one of the three readings to build around, and the pick changes the engineering answer. Under the literal reading, the temporary workflow needs a documented validation process and nothing quantitative about image quality, and even that only where Subpart D’s gate applies. Under the shorthand reading it needs the same, alongside a Subpart E pipeline for permanent records. Under the intent reading it needs FADGI-class capture on both sides, with a star level and material classes still to be selected. Those are different equipment profiles, different quality-control regimes and different unit costs at 100,000 images a day.

The Q&A had chances to settle the question. The metadata question (Q7) was answered by invoking FADGI compliance in general, and one question (Q23) used the phrase “FADGI validation.” No answer in the workbook names a subpart, a star level or an edition.

4.2 The boundary of automated quality checks is not set

The bullet names its subject plainly: automated detection of blurred, skewed or incomplete captures on production images. The three defects do not stand equally against the permanent regime’s own checks. For completeness, § 1236.46(f)(2) already requires automated and visual verification that 100% of the source’s informational content was captured. For skew, § 1236.46(d)(3)(v) requires a visual check that images are not skewed. Both apply on the permanent branch only, and neither says anything about blur.

Neither provision supplies what an automated production-image detector would need, namely a metric for each defect, a threshold and a checking frequency. The RFI sets none of these. It also does not say how its capability relates to the file-attribute checks that § 1236.46(d) requires for all image files, a provision that prescribes coverage and says nothing about automation.

The standards reviewed here measure quality thoroughly at the device level and not at all per production image. ISO 19264-1:2021  analyzes imaging-system quality from a single capture of a specified test target, and DICE works the same way. FADGI’s own image-level check is sampled visual inspection. In the corpus reviewed, meaning the FADGI Third Edition, 36 CFR part 1236 in its July 1, 2025 edition and ISO 19264-1:2021, quality is standardized at the device-and-target level. File attributes are checked in full, completeness verification is prescribed, and humans inspect a sample, with skew among their criteria. None of these is an automated method for scoring an arbitrary production image against the three defects without a target in the frame. For the automation the RFI asks for, the corpus offers no standard method or threshold to inherit.

4.3 Corrections have no stated bound

The same bullet asks for corrections, and the enhancement bullet next to it lists brightness, contrast, noise reduction, de-speckling, blank-page detection and cropping. No passage in the RFI, the description or the Q&A limits how far a captured image may be altered before it is stored. The regulation and the guidelines do set limits, and the limits differ in status. That difference decides what an acceptance test can enforce:

• Regulatory requirements (Subpart E): sharpening below 1.1 max modulation (§ 1236.50); no transcoding or upsampling in the provisions for photographic prints (§ 1236.48(c)(3)); auto-correction off during target evaluation, with production settings matching the evaluated ones (§ 1236.46(c)); embedded technical metadata left intact (§ 1236.54(b)(4)); visual rejection of over-sharpening, clipping, and images improperly cropped, flipped, inverted or skewed (§ 1236.46(d)).
• Guideline recommendations (FADGI Third Edition): scan the whole object without cropping, leaving a small visible border (§ 6.3), a recommendation and no prohibition; avoid relying on automatic correction features (§ 6.2); rotation correction allowed (§ 2.4.12); and where corrected and uncorrected files are both kept, the uncorrected ones are the archival masters.

The guidelines also accept minor adjustments after capture (§ 6.1), and the regulation separately requires failed records to be corrected or re-digitized (§ 1236.46(e)). The RFI asks for corrections and enhancements and states no bound of either kind. The documents leave open which kind a platform would be expected to enforce.

4.4 The capability list’s one “shall” has no yardstick

The sensitive-data detection bullet is the only “shall” among the capability bullets, and the Q&A states that its performance thresholds are not yet set (Q30). The engineering question is what a contract would have to invent if it wanted thresholds. The corpus reviewed for it is 36 CFR part 1236 in its July 1, 2025 edition, the FADGI Third Edition, NIST AI 100-1  (AI RMF 1.0, January 2023) and NIST SP 800-122  (April 2010). In that corpus, no federal standard sets quantitative performance thresholds for automated detection of sensitive information or PII in digitized records.

AI 100-1 is a voluntary framework whose Measure function leaves the choice of metrics and acceptable risk levels to the organization, and it gives no numeric detection threshold for any application. SP 800-122 defines PII and confidentiality impact levels and states no detection-accuracy requirement. The regulation requires recording access restrictions, Privacy Act and FOIA exemptions among them, as metadata carried over from the source records (§ 1236.54(c)(2)). It prescribes no method for finding such content. Any threshold would have to come from later requirements or acceptance criteria, which is where Q30 places it.

The adjacent capability has the same shape. Asked for minimum OCR, ICR and HTR accuracy levels, the Q&A names ten languages and no figure (Q27). FADGI § 6.7 asks for images good enough for OCR to reach a specified accuracy level, and it names neither the level nor a way to measure it. Section 1236.40(e) makes OCR optional in the permanent subpart. The corpus searched here covers that regulation, that guideline, the ISO catalog and NIST publications. In it, as of August 29, 2026, no federal regulation or international standard defines a normative measurement method or minimum value for OCR, ICR or HTR accuracy. The systematic U.S. benchmark record is more than thirty years old (NISTIR 4990 , 1993).

The diagram’s order of steps is an inference from the RFI’s capability groups and the Q&A, not something the notice defines, since the notice lists capabilities without sequencing them. The dashed steps are the open choices of this section plus the unnamed metadata schema of Section 2.

5. Scale

The notice’s own numbers come first. Asked for daily volumes, the Q&A answers “100,000 images daily (average) / 500,000 images daily (peak)” (Q4, repeated in Q25). The deployment is a main scanning location, the DocLab, plus two remote secured locations and 100 concurrent users (Q25). The fleet runs to roughly a hundred devices across eight models, some 80 of them Ricoh fi-7700 units, alongside IBML, Opex Falcon+, Epson, BancTec and Contex scanners (Q16, Q25). A quarter to a half of the documents carry handwriting, often intermingled with typed text (Q6), and the current OCR process covers over 70 languages (Q27).

The facility figure, sourced in Section 1, is capacity for approximately 1.7 billion pages at the Central Records Complex. One policy date sits behind federal digitization programs generally. OMB and NARA’s M-23-07  (December 23, 2022) states that starting on July 1, 2024, agencies “will be required to digitize permanent records created in analog formats before transfer to NARA.” It followed M-19-21  (June 28, 2019), which set the original end date for NARA’s acceptance of paper. The same budget request records, in an unrelated part of the Bureau’s records work, a FOIPA workload of almost 11,000 pending requests and near 500 pending litigations, together over 10 million pages awaiting processing (pp. 81–82). That figure is published FBI context only, and neither the RFI nor this analysis connects the two.

One comparison has to be refused explicitly, because it is the natural one to reach for. Subpart E’s visual sampling rule, a minimum of ten records or 10% of each batch, cannot be set against the 500,000-image peak to yield a workload figure. The units differ, since the rule counts records per batch and the peak counts images per day. Neither the batch size nor the ratio of images to records appears in any reviewed document, and the regulation also permits a statistically valid sampling plan in place of the default. Any number produced by combining them would describe the assumptions, and none of it would describe the program.

For calibration on what visual checking costs and finds, one peer-reviewed measurement was located, and it measures a different program. Chapman and Leonard studied quality control with visual checks of every scan in a large-scale digitization of archival manuscripts (Library Hi Tech 31(3), 2013 ). They report that “one error was discovered for every 223 scans reviewed (0.4 percent of scans).” One critical error turned up per 700 scans, or 0.1 percent, and quality control took 15 percent of project time. The figure comes from a 2013 academic-library project and measures nothing in this environment. The searches run for this analysis in August 2026 located no publication reporting automated-QA failure rates at volumes of the order the RFI states.

6. Mandates, notices and awards

What follows is the dated public record around this notice, in three groups that the documents themselves do not connect. The notice’s related field is empty, and the award records name products and dates without saying anything about where a requirement came from. No causal statement is made below, and none is available from these sources.

The mandates and the regulations. M-19-21 is dated June 28, 2019. Subpart D of 36 CFR part 1236 took effect on May 10, 2019 under final rule 84 FR 14265, and direct final rule 89 FR 46803 amended it effective August 28, 2024. M-23-07 is dated December 23, 2022. Subpart E took effect on June 5, 2023 under final rule 88 FR 28410, and its preamble applies it retroactively “to digitized permanent records that have not been transferred to the National Archives.” The FADGI Technical Guidelines, Third Edition, date from May 2023.

The notice family. A SAM search on FBI-IMD returns five notices, all inactive as of September 29, 2026. Three are recent Sources Sought notices under that prefix: FBI-IMD-ECM , “FBI Enterprise Content Manager Solution,” April 16, 2025; FBI-IMD-RT , “IMD Redaction Tool,” November 13, 2025, revised December 3, 2025; and this one, July 2026. The ECM description opens in near-identical language, with the same division and section conducting market research and seeking available sources. None of the three is linked in SAM to a posted solicitation. Listing them is a statement about SAM records and says nothing about a program.

The award record. FPDS data via USAspending, awarding sub-tier the FBI, was retrieved on September 29, 2026. The office prefix 15F067 matches the contracting office that issued the RFI. Twelve awards mention Captiva across FY2016 to FY2026. Among them are 15F06722F0002009 ($1,160,427.14, “PURCHASE OF NEW CAPTIVA LICENSES FOR THE CRC,” 2022 to 2024) and 15F06726F0000184 (“CAPTIVA RENEWAL,” to January 9, 2027). Alongside them is 15F06724F0000117 (“OPENTEXT MAINTENANCE,” $1,890,000.00, to December 29, 2026). Twenty-one awards mention IBML, including 15F06725P0000744 (software maintenance for IBML 8300/8400 scanners, to February 12, 2027), 15F06724P0001307 (maintenance for three IBML IT6 scanners “FOR DOCLAB”) and, most recently, 15F06726P0000936 (“IBML FUSION 8300”). One Opex award appears: 15F06724F0002139 (“PURCHASE OF TWO OPEX FALCON SCANNERS,” $192,995.36). Ricoh, formerly Fujitsu, appears through fi-series scanner purchases and maintenance, such as 15F06725F0001681 (“NEW FUJITSU 7700 SCANNERS,” 2025) and 15F06724F0001669 (maintenance to September 18, 2027).

The dated record shows those mandates, those notices and those awards, and no reviewed document connects them to each other. Two limits apply to the award figures. FPDS keyword search is not exhaustive, so no total for the capture environment is computed or quoted here. The horizons above are what the award records state, and they say nothing about what will be renewed.

7. Who would build it

What is requested. The RFI is open to manufacturers “with a documented proven history of successful integration” of image capture and document processing applications “on classified networks.” A reseller, integrator or prime may respond on the manufacturer’s behalf or jointly with it (Q22). Initial submissions are limited to eight pages. A potential Industry Day requires a Facility Clearance up to Top Secret, a GSA or NASA SEWP prime contract, active SAM registration, no active debarments and U.S. ownership. It may take a “Show Me, Don’t Tell Me” demonstration approach, and vendors must demonstrate with “customer provided data samples.”

The deployment constraints are firm. The system “must operate within a secret enclave without direct access to the internet,” and cloud functionality “must support hybrid architectures and must not require direct internet connectivity from classified networks.” The Q&A adds that the solution “must function completely within the Secret environment without requiring external cloud-based AI services” (Q19). Cloud or hybrid features may be demonstrated as optional enhancements, must be FedRAMP High authorized, and “must not impact the solution’s core capabilities if unavailable.”

One place where the RFI and its Q&A read differently is shown here with both texts rather than resolved. The PDF’s submission instructions state:

“The proposed software must also hold FedRAMP High certification to ensure compliance with federal security standards.”

The later FBI Q&A states (Q21):

“The FedRAMP High requirement applies only to cloud-hosted components and services. Software deployed exclusively within the on-premises classified enclave is not required to hold FedRAMP High authorization.”

Both are FBI-authored texts on the same notice, and the second is the later clarification.

What is documented. The environment the platform would enter is heterogeneous in the FBI’s own description (Q15): “OpenText Captiva Suite (ScanPlus, Completion, RescanPlus) Opex Software (CertainScan, Edit, Transform, and Monitor),” then “Ricoh (PaperStream Capture), IBML Capture Suite, and custom-coded applications.” The Q&A leaves explicitly open whether that environment would be replaced or joined. Asked whether the FBI seeks one platform to replace all current applications or a solution that coexists with some of them, the Q&A answers “Either” (Q24). It gives the same answer to whether the FBI wants an integrated solution with third-party components or everything from one manufacturer (Q23). No claim of consolidation is made here, because the notice makes none.

The award record backs those product names at different granularities, and the difference matters. For Captiva, the evidence is at the level of the exact product: award descriptions name Captiva licenses and Captiva maintenance, one of them tied to the CRC. For IBML, it is maintenance and license purchases on named scanner series, one award naming the DocLab, and a 2026 order for a Fusion 8300 under a hardware and perpetual-license product code. For Opex, the documented award buys two Falcon scanners and says nothing of the CertainScan software suite the Q&A names. For Ricoh, the awards are fi-series scanner purchases and maintenance, and PaperStream appears in one 2016 line item bundled with a scanner purchase. The named software stack is therefore documented by award chains only in part. Some of it appears at product level, and some only as the hardware it runs on.

What is unproven. Everything about transition remains unproven. No reviewed document states what would be migrated, in what order or on what schedule, and no total contract value for the capture environment is computed here. As of September 29, 2026, no reviewed public record shows which vendors responded or what any of them proposed.

8. What does not exist as a standard

The survey below states, for each reviewed document, what it fixes and what it does not, bounded to the version named in its row.

The table shows a corpus that is detailed about devices, formats and metadata for one class of records and silent about measuring individual production images. Four bounded negations follow, over the documents reviewed here: 36 CFR part 1236, both subparts, in the CFR annual edition of July 1, 2025; the FADGI Technical Guidelines, Third Edition (May 2023); NARA’s FAQ on digitizing temporary records; ISO 19264-1:2021; NIST AI 100-1 and SP 800-122.

First, no regulation or guideline among them defines FADGI digitization requirements for temporary federal records, and no document of the notice says which of the three readings in Section 3 was intended. Second, none defines an automated, targetless metric, threshold or checking frequency for scoring an arbitrary production image against blur, skew or incompleteness. What they standardize is device-and-target measurement, full checking of file attributes, prescribed completeness verification (§ 1236.46(f)(2)) and a visual skew check (§ 1236.46(d)(3)(v)) on the permanent branch, and sampled human inspection. Third, none sets quantitative thresholds for automated detection of sensitive information or PII, and none defines a measurement method or minimum value for OCR, ICR or HTR accuracy. Fourth, no external certifier of FADGI conformance appears in them. Conformance is asserted by the digitizing organization and validated by agency staff independent of the quality-control staff.

The regulation also holds something the gaps should not hide. For permanent paper records digitized reflectively, an agency does not have to design an image-quality regime, because parameters, formats, metadata elements, inspection coverage and the corrective threshold are written into the CFR with the assessment method named. The open choices in this notice sit on the other side of the phrase: which records the requirement is about, and which of three referents its temporary half names.

The bullet is one line in a document that labels its own requirements preliminary, and market research is supposed to leave choices open. What is unusual is the level at which the openness sits. Most underspecified requirements name an object and leave its parameters unselected. This one names a compound object whose permanent half arrives with parameters already selected by regulation and whose temporary half resolves three ways, depending on a reading no published text fixes. A respondent has to decide which requirement was named before deciding how to meet it. This article does not claim to know which reading the FBI intended, and it draws no conclusion about any vendor or about the acquisition that may follow. I work on verifiable evidence chains and provenance records in an adjacent area, and that work is separate from this analysis. More write-ups are on the engineering blog.

If a digitization program has to show which records standard its images meet and how each quality check was performed, designing that evidence trail is contract work I take on.