ReviewDesk
← Back

ReviewDesk guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026

Why does ReviewDesk render missing evidence as 'not recorded' — and never invent it?

Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing

The short answer. Because a gap can be fixed and a fabrication cannot be un-read. When a decision trace lacks an element — a factor, a notice flag, a review offer — ReviewDesk's packs render it as 'not recorded' rather than inferring the plausible value. The rule protects everyone: the reviewer trusts the pack's positives because its negatives are honest, the business gets a located work-list instead of false comfort, and the record never claims a duty was discharged when it was merely usual.

What is the 'not recorded' principle?

The 'not recorded' principle is ReviewDesk's evidence discipline: projections of decision records — the three-regime packs, exports and metrics — display exactly what the underlying records contain, with absent elements rendered explicitly as 'not recorded'. The engine never synthesises factors, notice status, review offers or metrics from patterns, defaults or sample data.

What problem does the principle solve?

The most dangerous moment in any evidence system is the gap quietly filled. A trace arrives without its notice flag; a helpful projection assumes notice is standard and renders it given. The pack now certifies a fact nobody recorded — and if that decision is ever examined, the organisation discovers its evidence system has been writing fiction at exactly the points where its process failed. The gaps and the failures coincide, because both come from the same broken step; auto-filling converts a locatable process fault into an undetectable evidentiary one. 'Not recorded' keeps the fault visible, priced and fixable.

The same logic governs ReviewDesk's own public surfaces. The landing coverage gauge reads the workspace's real answered-over-total and shows an empty state until cases exist — no sample figure, however clearly labelled, because labelled fictions have a way of losing their labels as they travel. Metrics, packs and pages all obey one rule: the number shown is the number the records hold.

Why is a gap better than a plausible value?

Three reasons, in ascending order of importance. Operationally, a gap is a work order with an address: this workflow, this field, this period — engineering and process fixes need exactly that specificity, and a rendered-plausible value deletes the ticket. Legally, a gap is a fact about your records that counsel can advise on; a fabricated value is a misstatement that changes the character of every document touching it. And epistemically, the pack's credibility is indivisible: a reviewer who catches one inferred value must discount every value, including the thousands that were real. Honesty about the one per cent is what purchases belief in the ninety-nine.

There is a fourth reason, internal and underrated: gaps keep the organisation's self-knowledge true. Dashboards that round their own ignorance upward produce leadership that believes the programme is finished. A pack with a page of 'not recorded' entries starts the right argument — why does this field not exist, whose pipeline owns it — in the first month, while the argument is cheap.

Where does the principle apply beyond the packs?

Everywhere ReviewDesk reports. The vault export shows the file as kept, including the phone outcome nobody logged — visible as an absence a quarterly test will find, per the vault guide. The decision history leaves a missing reviewer note impossible rather than reconstructed, because closure requires the note itself. The risk assessment draft carries its evidence gaps into the document rather than around them, per the assessment guide. Even these guides obey a version of it: they describe what the product's own pages and code say it does, and mark summaries as summaries, rather than decorating the product with capabilities it does not claim.

Apply it inside your own operation as a house rule for every artefact a sceptical reader might one day see. The test for any number or claim in a report: could we produce the record this came from, this afternoon, without preparation? If the honest answer involves the words 'we would reconstruct', the artefact should carry the gap instead. Reconstruction is how organisations talk themselves into histories they did not live.

How should teams respond to a page of gaps?

As a backlog, triaged by consequence. Gaps in fields that discharge duties — notice flags, review offers, factor lists on adverse decisions — outrank cosmetic gaps. Assign each a cause before assigning a fix: never-instrumented (build it), instrumented-but-failing (fix the pipeline), recorded-elsewhere (join it properly, at decision time, not by nightly inference). The cause taxonomy matters because the tempting wrong fix — backfill from the elsewhere system — reintroduces inference wearing an integration's clothes. If the value was not in the record when the pack was sealed, it does not enter that pack later; the corrected process fills tomorrow's packs, and the sealed ones keep their honest gaps.

Celebrate the burn-down visibly. The second pack, with half the gaps closed by process fixes, is the single most persuasive artefact a programme produces — for leadership, for counsel, for diligence. It proves the system detects and repairs itself, which is the actual definition of a controlled process. A first pack with no gaps, from a young programme, proves only that someone was afraid of gaps.

What does the principle forbid, precisely?

Inference of duty-discharging facts from defaults, patterns or surrounding records. Synthesis of factors from model identity. Backdating or backfilling sealed records. Sample or illustrative figures in any surface that reports the programme's own performance. Smoothing of metrics across periods to hide a bad month. And the softest, commonest form: narrative summaries that upgrade the records' certainty — the report that says 'notices were given' of a period whose traces say nothing either way. Each is a small fiction; the principle's claim is that small fictions in evidence systems are load-bearing, and fail at scale under load.

What it does not forbid is interpretation. The risk assessment may argue from the records; the reviewer note may weigh them; leadership may draw conclusions and own them as conclusions. The line is between judgement exercised openly on stated facts — the essence of the whole programme — and facts manufactured quietly under judgement's clothing. The line is bright, and the packs are built to keep it visible.

How does this make ReviewDesk different as a vendor?

It makes the product refusable to flattery, including its own. A vendor whose engine invents completeness is selling deployers a liability formatted as reassurance; a vendor whose packs show gaps is selling the less comfortable and more valuable thing — a true picture, early, while truth is cheap. Deployers evaluating evidence tooling should test this property directly, as the vendor guide suggests: file a deliberately incomplete trace and read the pack. What the pack does with your gap is what it will do with your programme's reputation on the worst day. Choose accordingly.

The principle is also why these guides never quote ReviewDesk capabilities beyond its own pages: Pro at $39 and Team at $99, the free workflow's exact contents, the engine's three packs with the EU's preparatory label — stated as the product states them, because a guide series about evidence honesty that decorated its own subject would refute itself before the first FAQ. The rule is the brand. Keep it, and the packs keep their value.

Your five actions for the 'not recorded' principle

  1. Render every absent element as 'not recorded' — in packs, exports, metrics and internal reports alike.
  2. Treat each gap as a located work order: assign its cause (never built, failing, or unjoined) before its fix.
  3. Never backfill sealed records; fix the process prospectively and let the next pack show the improvement.
  4. Apply the afternoon test to every reported number: producible from records today, or shown as a gap.
  5. Test any evidence vendor with a deliberately incomplete record and judge them by what the output does with the gap.

Worked example: the diligence pack with eleven honest gaps

A fictional vendor — ClearMark Screening, diligence time — produces its first three-regime pack for a prospective enterprise customer. The pack is strong on decisions and factors, and shows eleven 'not recorded' entries: notice flags missing for one legacy workflow whose instrumentation predates the trace rollout. The customer's reviewers do not penalise the gaps. They penalise nothing, in fact — they spend the meeting on the remediation dates attached to the gap list, and later report that the gaps were the reason the pack was believed: a supplier whose evidence shows its own unfinished edges is a supplier whose finished claims can be taken at face value.

The counterfactual writes itself: a spotless pack, assembled by inference, discovered as inference during the customer's own test filing — a standard diligence step the prospect runs on every vendor. ClearMark's eleven gaps cost it one roadmap conversation. The spotless fiction would have cost it the contract, and very nearly did cost a competitor exactly that, two quarters earlier, in the same buyer's process.

Frequently asked questions

What does 'not recorded' mean in a pack?

The underlying decision record does not contain that element — a factor, a notice flag, a review offer — so the pack displays the absence explicitly rather than inferring a plausible value. The gap is shown as a gap, with the record's integrity intact.

Why not infer values that are almost always true?

Because the exceptions are where examinations live: the missing flag and the failed process step coincide. Inference converts a fixable process fault into an undetectable misstatement, and one discovered inference discounts every true value in the pack.

Can gaps in a sealed pack be backfilled later?

No. Sealed packs keep their gaps; the fix is prospective — repair the recording process so future packs are complete. Backfilling sealed evidence would destroy the chain's meaning, per the append-only discipline.

Does the principle apply to metrics as well?

Yes. ReviewDesk's own coverage gauge reports real answered-over-total from the workspace's cases and shows an honest empty state when none exist. Sample figures in performance surfaces are forbidden by the same rule.

Is this guide legal advice?

No. ReviewDesk is templates and workflow tooling, not legal advice. The principle is ReviewDesk's evidence discipline, described as built; consult counsel for your evidence obligations.

Keep reading: How do you integrate decision traces through the ReviewDesk workspace API? · Who should own an ADMT programme — owner, reviewers and record-keepers? · What are the key ADMT deadlines — Illinois now, Colorado and California 2027, the EU in December 2027? · What changed with Colorado SB 26-189 — and what does it keep from the earlier AI Act approach? · What is the California ADMT compliance timeline — from the 2026 regulations to the 2028 assessment? · All ReviewDesk guides · ReviewDesk home · Try the free desk · Free letter generator · Free notice builder · Pricing

ReviewDesk is built by Ingenious AI. Templates and workflow tools — not legal advice.