● Public sector — Information & regulatory monitoring
The AI regulatory monitoring agent: watch the texts, produce sourced summaries
Keeping up with regulatory changes, case law and publications across dozens of topics takes a considerable amount of time — without always delivering what is expected: reliable, up-to-date monitoring that can be circulated directly to the departments. Your AI agent absorbs that documentary watch and produces concise, sourced notes. Hosted in France — on local inference or an isolated resource — your authority's data stays under control. The information officer keeps the decision: the AI agent assists, the public officer approves.
Hosted in France
Public data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Regulatory monitoring assistant · hosted in France
Give me this week's regulatory developments in public procurement and planning that affect our departments.
14 publications spotted on your topics this week. 2 texts with impact: a decree raising the publication thresholds for contracts (in force in 30 days) and an order changing one of the documents in a planning application. 1 case-law decision on appeal deadlines.
Concise monitoring note ready, with links to each official source — for review.
⛓ Source · Légifrance, official journal, case-law databases — official sources cited
Concise monitoring note ready, with links to each official source — for review.
⛓ Source · Légifrance, official journal, case-law databases — official sources cited
The decree on thresholds — which department is affected and what needs doing?
Department affected: public procurement. Impact: the procedure thresholds change from the 1st of next month; contracts being prepared above the old threshold need rechecking. Recommendation: circulate an alert to the purchasing team and update the internal guide.
I am preparing a targeted alert note for the department, for your approval.
✎ Action · alert note ready for review — the information officer approves
I am preparing a targeted alert note for the department, for your approval.
✎ Action · alert note ready for review — the information officer approves
Local inference · no data outside the EU
Data hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The public officer decidesThe agent assists, never circulates alone
✦ In brief
In a local authority or a public body, a Blue Lemon Agent agent handles regulatory and documentary monitoring: it continuously watches texts, case law and publications on your topics, produces concise, sourced monitoring notes and flags the changes that will affect your departments. It runs on local inference or is hosted in France: the authority's data is never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The time freed up is redirected towards higher-value analysis and circulation. Live within a few weeks, under human oversight.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
9
uses ready to deploy on this scope
0
decision taken without human approval
Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.
The context
Why public monitoring needs AI — and why sovereignty is not negotiable
Departments expect reliable, up-to-date, usable monitoring on a regulatory flow that keeps growing. But available time is mechanically reduced by watching the sources by hand — and the topics followed bear directly on the legal soundness of public decisions.
! The issue
The information officer is caught between departments demanding up-to-date, usable monitoring and a documentary flow that is exploding (texts, decrees, orders, circulars, case law, publications). Yet most consumer AI tools amount to entrusting the topics followed, the internal notes and the authority's own judgements to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For a public service, AI is only of interest if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, official sources systematically cited, permanent human oversight, circulation reserved to the public officer: the time saved on watching is never paid for in lost confidentiality or uncertain reliability. The aim is not to replace the information officer, but to give them back time for analysis and circulation.
The decisive point
Sovereignty of public data: confidentiality & compliance
An authority cannot entrust its monitoring and its internal notes to a foreign service. Here is how the architecture of our agents protects your data, topic by topic.
Local inference
The agent can run on a machine belonging to the authority: no data leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — your data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
As regards monitoring data, exposure to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
One isolated resource per entity
No pooling of public data: an environment strictly dedicated to your authority or institution.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
An agent strictly in support; no note circulated automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For the most sensitive data, SecNumCloud and HDS options are available depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
4 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The public body in this demonstration
Fictional public bodyVal d'Ancenoy Agglomeration — documentation and monitoring service (fictional authority)
- Sector
- Agglomeration of 38 municipalities and 96,400 inhabitants — responsibilities for water and wastewater, waste, planning with a shared development-control service for 31 municipalities, economic development, transport and housing; 6 shared services with the member municipalities
- Headcount
- 940 staff; the documentation and monitoring service has 3 people — a head librarian, a librarian and an apprentice. They serve the agglomeration's 14 departments and the 38 member municipalities; the agent touches no resident's file
- Public served
- 14 departments, 38 member municipalities, 6 shared services, 210 staff subscribed to the weekly monitoring bulletin, one community council every 6 weeks
- Order of magnitude
- 27 monitoring topics across 41 sources; 8,600 publications reviewed in the year, 48 weekly bulletins, 96 targeted alerts, 1,240 documentary search requests; an internal collection of 18,400 documents, of which 27 guides and 148 procedures still in service
- Tools already in place
- The agglomeration's document management system, the service's own catalogue, official platforms (Légifrance, the Official Journal, case-law databases), the buyer profile, intranet and email — the agent reads them, nothing is replaced or migrated
- Who decides what
- The head of the documentation service reads and signs every bulletin and every alert before it goes out; the chief executive decides whether a topic is opened or closed; each recipient department names the officer who acknowledges an alert
- Room for improvement
- The weekly bulletin takes 26 hours, 15 h 36 of which is collection and formatting alone; a search of the collection comes back in 3 days; the 27 guides and 148 procedures are revisited only once a year; over the last twelve months, 5 texts with an impact reached the services AFTER they had entered into force — between 34 and 61 days after publication
At Val d'Ancenoy, three people run regulatory monitoring for 940 staff and 38 municipalities: 27 topics, 41 sources, 48 bulletins and 1,240 searches a year. The time is not lost in analysis — it is lost in trawling the sources, in formatting, and in catching up on texts seen too late. The exchanges below cover a year, from reconciling the document collection to the twelve-month review put to the executive board. The agent is hosted in France on an isolated resource, reads the document management system and the official platforms without ever writing to them, and no bulletin and no alert goes out without the head of service's signature.
This public body, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real service.
Regulatory monitoring assistant · hosted in France
I have reviewed your 8,600 publications for the year across the 27 topics and the 41 sources, and reconciled, document by document, your internal collection against the texts it cites — the reconciliation three people have never had time to do.
What the count says: 2,860 of your 18,400 documents cite at least one text — the rest are minutes, letters or plans. Of those 2,860, 214 cite a text that has been repealed, amended or replaced since they were written — 7.5%. And 31 of those 214 sit inside the 27 guides and 148 procedures still in service: those are the ones making a member of staff act, today, on a text that has moved.
What the three items you can check against your own records are worth:
· The weekly monitoring bulletin is 26 hours end to end for your three people, 60% of which goes on collection, sifting and formatting — 15 h 36. After: 12%, i.e. 3 h 07. The analysis, the choice of what matters and the commentary stay with you: that is the work monitoring is bought for.
· Monitoring one topic across the 41 sources is 5 hours a month, 20% of which is raw trawling — 1 hour. After: 8%, i.e. 24 minutes. Across 27 topics and 12 months, 194 hours.
· A search of the collection travels through 5 h 20 of cumulative time — the request, the search, the back-and-forth to clarify, the answer —, 10% of which is actual searching: 32 minutes. After: 8%, i.e. 25 minutes. On this item the fiche is cautious, and rightly so: what shifts is not the time, it is the lead time — 3 days brought down to the minute.
And here is the gap that costs the most, because the services see it: over the last twelve months, 5 texts with an impact reached the services AFTER they had entered into force. Your average gap between publication and circulation is 9 days; those five took between 34 and 61. None was missed through carelessness: all five were in the flow, on monitored topics, and they slipped through because 8,600 publications a year cannot be read by three people.
The total time returned, item by item: 1,260 hours over the year — at 151.67 hours a month, more than eight months of work returned to a three-person service.
The next step, and it takes an hour: you read the 31 outdated citations in the guides and procedures still in service — I have written the replacement text for each —, and your head of service settles the version. The other 183 follow at whatever pace you set. monitoring-landscape_27-topics-41-sources.pdf8,600 publications, the three items costed, 1,260 hours identified
⛓ Sourced · 8,600 publications reviewed over 12 months, internal collection of 18,400 documents, 41 official sources monitored, service circulation log
What the count says: 2,860 of your 18,400 documents cite at least one text — the rest are minutes, letters or plans. Of those 2,860, 214 cite a text that has been repealed, amended or replaced since they were written — 7.5%. And 31 of those 214 sit inside the 27 guides and 148 procedures still in service: those are the ones making a member of staff act, today, on a text that has moved.
What the three items you can check against your own records are worth:
· The weekly monitoring bulletin is 26 hours end to end for your three people, 60% of which goes on collection, sifting and formatting — 15 h 36. After: 12%, i.e. 3 h 07. The analysis, the choice of what matters and the commentary stay with you: that is the work monitoring is bought for.
· Monitoring one topic across the 41 sources is 5 hours a month, 20% of which is raw trawling — 1 hour. After: 8%, i.e. 24 minutes. Across 27 topics and 12 months, 194 hours.
· A search of the collection travels through 5 h 20 of cumulative time — the request, the search, the back-and-forth to clarify, the answer —, 10% of which is actual searching: 32 minutes. After: 8%, i.e. 25 minutes. On this item the fiche is cautious, and rightly so: what shifts is not the time, it is the lead time — 3 days brought down to the minute.
And here is the gap that costs the most, because the services see it: over the last twelve months, 5 texts with an impact reached the services AFTER they had entered into force. Your average gap between publication and circulation is 9 days; those five took between 34 and 61. None was missed through carelessness: all five were in the flow, on monitored topics, and they slipped through because 8,600 publications a year cannot be read by three people.
The total time returned, item by item: 1,260 hours over the year — at 151.67 hours a month, more than eight months of work returned to a three-person service.
The next step, and it takes an hour: you read the 31 outdated citations in the guides and procedures still in service — I have written the replacement text for each —, and your head of service settles the version. The other 183 follow at whatever pace you set. monitoring-landscape_27-topics-41-sources.pdf8,600 publications, the three items costed, 1,260 hours identified
⛓ Sourced · 8,600 publications reviewed over 12 months, internal collection of 18,400 documents, 41 official sources monitored, service circulation log
214 documents citing an outdated text, and 5 texts arriving too late: that is exactly what we dread. How do we avoid creating more next year?
With four written rules, already run across 36 months of your own monitoring — 144 bulletins and 24,800 publications —, and each one arrives with what it misses.
For each I give you the number of alerts it would have raised, the share confirmed on examination, and what it would have let through:
· Rule A — a published text whose entry into force is deferred and which has no alert addressed to a named service fifteen days beforehand: 168 alerts over 36 months, 159 confirmed — 94.6%. The other 9 concerned topics monitored for information, with no recipient service.
· Rule B — a collection document citing a text amended or repealed after it was written: 214 alerts, 196 confirmed — 91.6%. The other 18 cite the text as history, which is entirely legitimate — which is why the rule proposes and deletes nothing.
· Rule C — a court decision handed down on a text that one of your guides comments on: 74 alerts, 74 confirmed — 100%.
And here is what those three let through, because you must know it before you sign: 57 texts entered into force on the very day they were published. No deferred entry, so nothing to flag fifteen days ahead — rule A does not see them, and they are precisely the ones that leave no margin. So I wrote rule D — a text of immediate application touching a monitored topic — and ran it across the same 36 months: 57 alerts, 49 confirmed — 85.9%.
All four together: 513 alerts, 478 confirmed — 93.2%, and they cover 478 of the 514 gaps found over the period, i.e. 93.0%.
I advise against switching all four on the same day, and I cost that advice: all four together would produce 196 alerts in the first quarter, i.e. roughly 32 h 40 of examination for a three-person service that already puts out a bulletin every week. One rule per quarter is 4 hours the first time, and you keep control of what you look at.
Going live stays with the service: a rule only takes effect once approved by your head of documentation. I save you the writing and the measurement; the decision takes one team meeting instead of a working group. document-collection_214-outdated-citations-and-4-rules.pdfThe 214 outdated citations, the 31 inside guides in service, 4 rules tested over 36 months
⛓ Sourced · 36 months of monitoring (144 bulletins, 24,800 publications), internal collection, circulation and acknowledgement logs
For each I give you the number of alerts it would have raised, the share confirmed on examination, and what it would have let through:
· Rule A — a published text whose entry into force is deferred and which has no alert addressed to a named service fifteen days beforehand: 168 alerts over 36 months, 159 confirmed — 94.6%. The other 9 concerned topics monitored for information, with no recipient service.
· Rule B — a collection document citing a text amended or repealed after it was written: 214 alerts, 196 confirmed — 91.6%. The other 18 cite the text as history, which is entirely legitimate — which is why the rule proposes and deletes nothing.
· Rule C — a court decision handed down on a text that one of your guides comments on: 74 alerts, 74 confirmed — 100%.
And here is what those three let through, because you must know it before you sign: 57 texts entered into force on the very day they were published. No deferred entry, so nothing to flag fifteen days ahead — rule A does not see them, and they are precisely the ones that leave no margin. So I wrote rule D — a text of immediate application touching a monitored topic — and ran it across the same 36 months: 57 alerts, 49 confirmed — 85.9%.
All four together: 513 alerts, 478 confirmed — 93.2%, and they cover 478 of the 514 gaps found over the period, i.e. 93.0%.
I advise against switching all four on the same day, and I cost that advice: all four together would produce 196 alerts in the first quarter, i.e. roughly 32 h 40 of examination for a three-person service that already puts out a bulletin every week. One rule per quarter is 4 hours the first time, and you keep control of what you look at.
Going live stays with the service: a rule only takes effect once approved by your head of documentation. I save you the writing and the measurement; the decision takes one team meeting instead of a working group. document-collection_214-outdated-citations-and-4-rules.pdfThe 214 outdated citations, the 31 inside guides in service, 4 rules tested over 36 months
⛓ Sourced · 36 months of monitoring (144 bulletins, 24,800 publications), internal collection, circulation and acknowledgement logs
Our monitoring topics, our internal notes and our internal deliberations say a great deal about what the agglomeration is preparing. Where does that data go?
Nowhere outside your walls, and you are right to put the question in those terms: the list of what an authority monitors is itself strategic information.
Where I run: on an isolated resource hosted in France, under French law, dedicated to Val d'Ancenoy and to no one else — or on a machine in the agglomeration if you prefer local inference, in which case nothing even leaves the network. Neither your 27 topics, nor your internal notes, nor your internal deliberations are passed to a foreign service: exposure to the Cloud Act and FISA 702 is reduced accordingly, without location alone ruling it out — it is documented for the configuration chosen. No pooling between authorities: your monitoring scope feeds no other environment.
What I read, and what is not open to me: I have read-only access to the document management system, to the service's catalogue and to the 41 public sources. I have no write access at all: I produce draft bulletins, alerts and papers, and a public official takes them on and circulates them. Residents' files, named planning applications, staff files and companies' tender submissions are not open to me at all — regulatory monitoring has no use for them, and article 5.1.c GDPR requires precisely that only what serves the purpose be exposed. Every access is withdrawn with a word, and the withdrawal takes effect within the second.
I plug into what you already have: document management, catalogue, official platforms, buyer profile, intranet. No migration, no change of tool, no re-cataloguing — I come alongside, and the day you change document management system it is the connector that changes, not your 18,400 documents.
What you can show your data protection officer: every read is logged; every note produced carries the sources it is drawn from, with a link to each; access is opened by role — a department officer, a member municipality and the documentation service do not see the same thing; and the record of processing activities is already written, legal basis by legal basis — public interest task, article 6.1.e GDPR.
The next step I propose: a quarterly thirty-minute review where I show you the access rights actually used and those that served no purpose — a useless access right is one to close, and I bring you the list rather than asking you for it. sovereign-framework_hosting-access-traceability.pdfWhere the topics and notes live, who sees what, what is withdrawn with a word
✎ Framework · hosting architecture, role access matrix, read log, the service's record of processing activities
Where I run: on an isolated resource hosted in France, under French law, dedicated to Val d'Ancenoy and to no one else — or on a machine in the agglomeration if you prefer local inference, in which case nothing even leaves the network. Neither your 27 topics, nor your internal notes, nor your internal deliberations are passed to a foreign service: exposure to the Cloud Act and FISA 702 is reduced accordingly, without location alone ruling it out — it is documented for the configuration chosen. No pooling between authorities: your monitoring scope feeds no other environment.
What I read, and what is not open to me: I have read-only access to the document management system, to the service's catalogue and to the 41 public sources. I have no write access at all: I produce draft bulletins, alerts and papers, and a public official takes them on and circulates them. Residents' files, named planning applications, staff files and companies' tender submissions are not open to me at all — regulatory monitoring has no use for them, and article 5.1.c GDPR requires precisely that only what serves the purpose be exposed. Every access is withdrawn with a word, and the withdrawal takes effect within the second.
I plug into what you already have: document management, catalogue, official platforms, buyer profile, intranet. No migration, no change of tool, no re-cataloguing — I come alongside, and the day you change document management system it is the connector that changes, not your 18,400 documents.
What you can show your data protection officer: every read is logged; every note produced carries the sources it is drawn from, with a link to each; access is opened by role — a department officer, a member municipality and the documentation service do not see the same thing; and the record of processing activities is already written, legal basis by legal basis — public interest task, article 6.1.e GDPR.
The next step I propose: a quarterly thirty-minute review where I show you the access rights actually used and those that served no purpose — a useless access right is one to close, and I bring you the list rather than asking you for it. sovereign-framework_hosting-access-traceability.pdfWhere the topics and notes live, who sees what, what is withdrawn with a word
✎ Framework · hosting architecture, role access matrix, read log, the service's record of processing activities
Give me the week's regulatory developments in public procurement and planning that concern our services.
14 publications spotted on your topics this week. 2 texts with an impact: a decree raising the advertising thresholds for public contracts (entering into force in 30 days) and an order amending one item in the planning application pack. 1 court decision on appeal deadlines.
A concise monitoring note is ready, with links to every official source — for your review.
And I give you the sift, not just the result: 186 publications reviewed this week across the two topics, 14 retained, 172 set aside — and each of the 172 carries its reason: outside our remit, text already flagged, draft not yet published, applies to a different category of authority. Monitoring whose discards you cannot see is monitoring you cannot correct.
Each of the 14 arrives with its three dates, and this is the rule I propose we hold: the date of publication, the date of entry into force, and the deadline by which the service must have acted — plus the named service that must act. A note that gives a text without its effective date and without its addressee is a note that gets read and does not get done — and that is exactly what let your 5 texts slip through last year.
The figure that does not flatter me, and I publish it: in the first quarter, out of 2,150 publications reviewed, I set aside 3 that should have been retained. All three concerned the water and wastewater topic, whose vocabulary was too narrow. That is fixed: 41 terms added to that topic's scope. In the second quarter, out of 2,150 publications, 0 discards of that kind — and 11 additional publications retained, 2 of them with an impact. I propose to keep publishing this count every quarter: monitoring whose misses are not measured is monitoring you know nothing about.
What the note returns to you: 15 h 36 of collection and formatting brought down to 3 h 07 — 60% of the bulletin's time brought down to 12%. Across your 48 bulletins, 599 hours returned over the year, and they go back where they belong: choosing what matters, and saying why. monitoring-bulletin_week-14-publications-retained.pdfThe 14 retained with their three dates, the 172 set aside with their reason, the link to each source
⛓ Source · Légifrance, Official Journal, case-law databases — official sources cited
A concise monitoring note is ready, with links to every official source — for your review.
And I give you the sift, not just the result: 186 publications reviewed this week across the two topics, 14 retained, 172 set aside — and each of the 172 carries its reason: outside our remit, text already flagged, draft not yet published, applies to a different category of authority. Monitoring whose discards you cannot see is monitoring you cannot correct.
Each of the 14 arrives with its three dates, and this is the rule I propose we hold: the date of publication, the date of entry into force, and the deadline by which the service must have acted — plus the named service that must act. A note that gives a text without its effective date and without its addressee is a note that gets read and does not get done — and that is exactly what let your 5 texts slip through last year.
The figure that does not flatter me, and I publish it: in the first quarter, out of 2,150 publications reviewed, I set aside 3 that should have been retained. All three concerned the water and wastewater topic, whose vocabulary was too narrow. That is fixed: 41 terms added to that topic's scope. In the second quarter, out of 2,150 publications, 0 discards of that kind — and 11 additional publications retained, 2 of them with an impact. I propose to keep publishing this count every quarter: monitoring whose misses are not measured is monitoring you know nothing about.
What the note returns to you: 15 h 36 of collection and formatting brought down to 3 h 07 — 60% of the bulletin's time brought down to 12%. Across your 48 bulletins, 599 hours returned over the year, and they go back where they belong: choosing what matters, and saying why. monitoring-bulletin_week-14-publications-retained.pdfThe 14 retained with their three dates, the 172 set aside with their reason, the link to each source
⛓ Source · Légifrance, Official Journal, case-law databases — official sources cited
The decree on thresholds — which service is affected and what has to be done?
Service affected: public procurement. Impact: the procedure thresholds change from the 1st of next month; contracts being prepared above the old threshold need re-checking. Recommendation: circulate an alert to the procurement team and update the internal guide.
I am preparing a targeted alert note for the service, for your approval.
And I do not leave you with a recommendation: I have done the checking. On your buyer profile, 23 procurement exercises are in preparation. 7 fall within the value band affected by the change — for each I give you its internal number, its subject, its estimated value, the sponsoring department and its expected publication date. Three of them are due to be published before entry into force, four after: those are not the same rules, and they are two separate lists inside the alert.
The internal guide: your procurement guide cites the threshold in 4 places — pages 7, 12, 19 and 31. All 4 passages are identified, with their current wording and the replacement text already drafted. You read four paragraphs; you do not reopen a 44-page guide.
This text's three dates, and its addressee: published in the Official Journal on Tuesday the 12th · in force on the 1st of next month, in 30 days · deadline for the procurement team set at the 17th, fifteen days ahead, so that the 7 exercises are re-checked before they are published. Named addressee: the shared procurement team — 6 buyers, one named officer who will acknowledge receipt.
What the alert returns to you: a targeted alert took 1 h 45 — finding the service, drafting, sending, chasing the acknowledgement; it now takes 25 minutes. Across your 96 alerts in the year, 128 hours returned.
Nothing is sent: the alert and the 4 guide paragraphs await your head of service's signature, and everything can be taken back with a word until it has gone. targeted-alert_advertising-thresholds-procurement-team.pdfThe 3 dates, the 7 exercises affected, the 4 guide passages and their replacement
✎ Action · alert note ready for review — the librarian approves
I am preparing a targeted alert note for the service, for your approval.
And I do not leave you with a recommendation: I have done the checking. On your buyer profile, 23 procurement exercises are in preparation. 7 fall within the value band affected by the change — for each I give you its internal number, its subject, its estimated value, the sponsoring department and its expected publication date. Three of them are due to be published before entry into force, four after: those are not the same rules, and they are two separate lists inside the alert.
The internal guide: your procurement guide cites the threshold in 4 places — pages 7, 12, 19 and 31. All 4 passages are identified, with their current wording and the replacement text already drafted. You read four paragraphs; you do not reopen a 44-page guide.
This text's three dates, and its addressee: published in the Official Journal on Tuesday the 12th · in force on the 1st of next month, in 30 days · deadline for the procurement team set at the 17th, fifteen days ahead, so that the 7 exercises are re-checked before they are published. Named addressee: the shared procurement team — 6 buyers, one named officer who will acknowledge receipt.
What the alert returns to you: a targeted alert took 1 h 45 — finding the service, drafting, sending, chasing the acknowledgement; it now takes 25 minutes. Across your 96 alerts in the year, 128 hours returned.
Nothing is sent: the alert and the 4 guide paragraphs await your head of service's signature, and everything can be taken back with a word until it has gone. targeted-alert_advertising-thresholds-procurement-team.pdfThe 3 dates, the 7 exercises affected, the 4 guide passages and their replacement
✎ Action · alert note ready for review — the librarian approves
And the order on the planning item? We handle development control for 31 municipalities — I cannot afford an outdated form on 31 websites.
Spotted, and the catch-up is already mapped — this is precisely the case where an apparently minor text costs dearly, because it multiplies by 31.
What the order says: it amends one item of the application pack, for applications lodged from its entry into force onwards. Applications already lodged are unaffected — that is the first thing your case officers will ask, and it sits at the top of the alert.
What I found at your end, which nobody would have had time to check across 31 websites:
· All 31 municipalities link to the application pack from their site; 4 host a local copy of the amended item — those 4 will serve an outdated version on the day the order takes effect, and I give you the 4 page addresses.
· Your 2 lodging guides — the case officers' guide and the one handed over the counter — describe the item in its current version, with the page and paragraph of each.
· Your shared development-control service handles 2,180 authorisations a year: at the rate of the last three years, roughly 180 applications will be lodged in the month following entry into force.
Now the court decision, and it calls for a distinction: a decision of an administrative court of appeal dated 6 February, handed down on appeal deadlines. It bears on a text that your « development control and site notices » guide comments on at page 14: I have identified the guide passage, the decision, its court and its date, and set them side by side. The meaning to be given to that decision, and what follows from it, belong to your legal team — I bring you the cross-reference, done, dated and sourced, not the interpretation. It is the cross-referencing that costs time; the interpretation is your lawyer's craft, and it takes twenty minutes once the pieces are assembled.
The three dates and the addressee: published in the Official Journal on Thursday the 14th · in force on the first day of the second following month · deadline for the development-control service set fifteen days ahead, so that the 4 pages and the 2 guides are updated. Named addressee: the shared development-control service, copied to the 4 municipalities concerned.
Draft alert and draft email to the 4 municipalities ready for review — nothing is sent without your signature. planning-monitoring_application-item-and-appeal-deadlines.pdfThe 4 municipal pages to update, the 2 guides, the guide / decision cross-reference
⛓ Sourced · Official Journal, case-law database, 31 municipal websites checked, lodging guides of the development-control service, three years of application statistics
What the order says: it amends one item of the application pack, for applications lodged from its entry into force onwards. Applications already lodged are unaffected — that is the first thing your case officers will ask, and it sits at the top of the alert.
What I found at your end, which nobody would have had time to check across 31 websites:
· All 31 municipalities link to the application pack from their site; 4 host a local copy of the amended item — those 4 will serve an outdated version on the day the order takes effect, and I give you the 4 page addresses.
· Your 2 lodging guides — the case officers' guide and the one handed over the counter — describe the item in its current version, with the page and paragraph of each.
· Your shared development-control service handles 2,180 authorisations a year: at the rate of the last three years, roughly 180 applications will be lodged in the month following entry into force.
Now the court decision, and it calls for a distinction: a decision of an administrative court of appeal dated 6 February, handed down on appeal deadlines. It bears on a text that your « development control and site notices » guide comments on at page 14: I have identified the guide passage, the decision, its court and its date, and set them side by side. The meaning to be given to that decision, and what follows from it, belong to your legal team — I bring you the cross-reference, done, dated and sourced, not the interpretation. It is the cross-referencing that costs time; the interpretation is your lawyer's craft, and it takes twenty minutes once the pieces are assembled.
The three dates and the addressee: published in the Official Journal on Thursday the 14th · in force on the first day of the second following month · deadline for the development-control service set fifteen days ahead, so that the 4 pages and the 2 guides are updated. Named addressee: the shared development-control service, copied to the 4 municipalities concerned.
Draft alert and draft email to the 4 municipalities ready for review — nothing is sent without your signature. planning-monitoring_application-item-and-appeal-deadlines.pdfThe 4 municipal pages to update, the 2 guides, the guide / decision cross-reference
⛓ Sourced · Official Journal, case-law database, 31 municipal websites checked, lodging guides of the development-control service, three years of application statistics
The planning director wants everything we hold on development control in protected areas: the text, the doctrine, our internal notes and the decisions we have already commented on. That usually takes me two days.
Produced in four minutes, and every item arrives with its citation — the reference, the date, the author and the exact passage.
34 documents retained out of the collection's 18,400 and out of the official sources:
· 6 texts and orders, each with a link to its official source and its version in force as at today.
· 3 circulars — and there is a point here that is not caution, it is law. Article L. 312-2 of the French code on relations between the public and the administration provides that an instruction or circular not published on one of the prescribed supports is not applicable, that its authors cannot rely on it against members of the public, and that it is deemed repealed. So I checked publication for all three: two are published, the third is not — and it is cited in 2 of your internal notes. This is not a cataloguing detail: it changes what your service can enforce against an applicant, and it is better learned here than before the administrative court.
· 11 internal service notes, with their date and signatory.
· 9 council resolutions, with their number and date.
· 5 decisions already commented on by your service, with the original commentary and its date.
And I give you what I set aside, with the reason: 61 documents came up on the same terms and are not in the selection — 38 outside the geographic scope concerned, 14 repealed, 9 duplicates of an earlier version. You can pull any of them back with a word; what you cannot do is be unaware that they exist.
What this returns to you: a search travelled through 5 h 20 of cumulative time, 32 minutes of which was actual searching; it now takes 25 — 10% of the time brought down to 8%. But the figure that counts is the lead time: 3 days brought down to the minute, across 1,240 searches a year, i.e. 132 hours returned. Your planning director no longer waits two days; he calls you and you answer during the call. documentary-search_34-documents-and-their-passages.pdfThe 34 retained with reference, date and exact passage, the 61 set aside with their reason, the unpublished circular
⛓ Sourced · internal collection of 18,400 documents, Légifrance and the Official Journal, case-law database, register of council resolutions
34 documents retained out of the collection's 18,400 and out of the official sources:
· 6 texts and orders, each with a link to its official source and its version in force as at today.
· 3 circulars — and there is a point here that is not caution, it is law. Article L. 312-2 of the French code on relations between the public and the administration provides that an instruction or circular not published on one of the prescribed supports is not applicable, that its authors cannot rely on it against members of the public, and that it is deemed repealed. So I checked publication for all three: two are published, the third is not — and it is cited in 2 of your internal notes. This is not a cataloguing detail: it changes what your service can enforce against an applicant, and it is better learned here than before the administrative court.
· 11 internal service notes, with their date and signatory.
· 9 council resolutions, with their number and date.
· 5 decisions already commented on by your service, with the original commentary and its date.
And I give you what I set aside, with the reason: 61 documents came up on the same terms and are not in the selection — 38 outside the geographic scope concerned, 14 repealed, 9 duplicates of an earlier version. You can pull any of them back with a word; what you cannot do is be unaware that they exist.
What this returns to you: a search travelled through 5 h 20 of cumulative time, 32 minutes of which was actual searching; it now takes 25 — 10% of the time brought down to 8%. But the figure that counts is the lead time: 3 days brought down to the minute, across 1,240 searches a year, i.e. 132 hours returned. Your planning director no longer waits two days; he calls you and you answer during the call. documentary-search_34-documents-and-their-passages.pdfThe 34 retained with reference, date and exact passage, the 61 set aside with their reason, the unpublished circular
⛓ Sourced · internal collection of 18,400 documents, Légifrance and the Official Journal, case-law database, register of council resolutions
Our guides, precisely. We know they age, and nobody ever has time to revisit them. What do you do about that?
I have already read them, and I am not handing you a list of problems: I am handing you 31 rewritten paragraphs.
The position, guide by guide: 27 guides and 148 procedures in service; 31 citations of texts that have moved since they were written. Three guides hold 18 of them: the procurement guide (7 citations), the « development control and site notices » guide (6) and the delegations guide (5). The other 13 are spread across 9 guides and 4 procedures.
What I bring for each one: the page and the paragraph, the current wording, the text that amended it with its official link and its date, and the replacement paragraph already drafted in the guide's own style. You read 31 paragraphs; you do not reopen 27 guides.
What this returns to you: revisiting the 27 guides and 148 procedures by hand is 310 hours a year — 6 hours per guide and 1 hour per procedure. With automatic flagging of citations that have moved: 103 hours. 206 hours returned.
And the action that stops it happening again: every guide now carries, as an attachment, the list of texts it cites. The day one of them moves, the guide flags itself to its author, with the passage concerned. I am not the one who decides to amend it — I mark it « to check », I write the proposal, and the guide's author decides. They lift the mark with a word, with their reason and their date, and that reason counts: the most frequent reasons tell you which rules to tighten.
The 18 historical citations are the example: a guide that recalls a repealed text to explain where a rule comes from is entirely right to do so. Rule B flagged them; they are now marked « historical » by hand by their authors, and they no longer come up. A rule that flags what is fine ends up not being read at all. document-collection_214-outdated-citations-and-4-rules.pdfThe 31 citations inside guides in service, the replacement paragraphs, the 18 historical citations
⛓ Sourced · 27 guides and 148 procedures from the collection, cited texts checked against their version in force, log of lifted marks
The position, guide by guide: 27 guides and 148 procedures in service; 31 citations of texts that have moved since they were written. Three guides hold 18 of them: the procurement guide (7 citations), the « development control and site notices » guide (6) and the delegations guide (5). The other 13 are spread across 9 guides and 4 procedures.
What I bring for each one: the page and the paragraph, the current wording, the text that amended it with its official link and its date, and the replacement paragraph already drafted in the guide's own style. You read 31 paragraphs; you do not reopen 27 guides.
What this returns to you: revisiting the 27 guides and 148 procedures by hand is 310 hours a year — 6 hours per guide and 1 hour per procedure. With automatic flagging of citations that have moved: 103 hours. 206 hours returned.
And the action that stops it happening again: every guide now carries, as an attachment, the list of texts it cites. The day one of them moves, the guide flags itself to its author, with the passage concerned. I am not the one who decides to amend it — I mark it « to check », I write the proposal, and the guide's author decides. They lift the mark with a word, with their reason and their date, and that reason counts: the most frequent reasons tell you which rules to tighten.
The 18 historical citations are the example: a guide that recalls a repealed text to explain where a rule comes from is entirely right to do so. Rule B flagged them; they are now marked « historical » by hand by their authors, and they no longer come up. A rule that flags what is fine ends up not being read at all. document-collection_214-outdated-citations-and-4-rules.pdfThe 31 citations inside guides in service, the replacement paragraphs, the 18 historical citations
⛓ Sourced · 27 guides and 148 procedures from the collection, cited texts checked against their version in force, log of lifted marks
One last thing, and it is the one that irritates me most: staff do not read the bulletin to the end. And I have a 214-page report to work through by Monday.
Both are dealt with, and the first is measured at your end before it is fixed.
The bulletin first, and the figure is yours: 210 subscribers, 48 issues; 61% opened and 22% read to the end, taken from your intranet. A bulletin opened by 128 staff and read in full by 46 is not a distribution problem, it is a writing problem.
What I have done: a plain-language version of every impact alert — short sentences, one idea per sentence, an action verb, and at the top: the service concerned, the date of entry into force and what has to be done. The full regulatory text stays underneath, with its official link: the plain version does not replace the source, it opens onto it. And the draft is submitted for approval, as the rule requires: a plain-language version is approved, it is not inferred.
And for the 3 notes intended for the public — the ones handed over the counter and published on the site —, an easy-read version is proposed as well, on the same principles, taken further. The same dated version then serves the counter, the website, email and the notice board in the town hall: four supports, one version, which puts an end to guidance that diverges from one support to another.
Now the 214-page report — it has been worked through, and here is what it holds for you:
· 18 passages touch your 27 topics, each with its page, its paragraph and its exact quotation.
· 4 recommendations concern you directly, with the recipient service for each and, where the report gives one, the deadline it proposes.
· 2 data tables bear on responsibilities you exercise — I have taken them as they stand, without recomputing them, with the page where they can be found.
· And what I left aside is listed, not silent: the other 196 pages each get a one-line contents entry, so that you can challenge my sift rather than submit to it. No figure, no date, no condition and no duration has been compressed: where a passage counts, it is quoted, not paraphrased.
You have Monday morning, and you have 18 passages to read instead of 214 pages — plus the contents of the other 196 to check that nothing that counts was left behind. plain-language-and-long-reading_alert-rewritten-and-214-page-report.pdfThe alert before and after, the easy-read version, the 18 report passages and the contents of the other 196 pages
⛓ Sourced · intranet statistics across 48 issues, 214-page report worked through page by page, the service's plain-language writing rules
The bulletin first, and the figure is yours: 210 subscribers, 48 issues; 61% opened and 22% read to the end, taken from your intranet. A bulletin opened by 128 staff and read in full by 46 is not a distribution problem, it is a writing problem.
What I have done: a plain-language version of every impact alert — short sentences, one idea per sentence, an action verb, and at the top: the service concerned, the date of entry into force and what has to be done. The full regulatory text stays underneath, with its official link: the plain version does not replace the source, it opens onto it. And the draft is submitted for approval, as the rule requires: a plain-language version is approved, it is not inferred.
And for the 3 notes intended for the public — the ones handed over the counter and published on the site —, an easy-read version is proposed as well, on the same principles, taken further. The same dated version then serves the counter, the website, email and the notice board in the town hall: four supports, one version, which puts an end to guidance that diverges from one support to another.
Now the 214-page report — it has been worked through, and here is what it holds for you:
· 18 passages touch your 27 topics, each with its page, its paragraph and its exact quotation.
· 4 recommendations concern you directly, with the recipient service for each and, where the report gives one, the deadline it proposes.
· 2 data tables bear on responsibilities you exercise — I have taken them as they stand, without recomputing them, with the page where they can be found.
· And what I left aside is listed, not silent: the other 196 pages each get a one-line contents entry, so that you can challenge my sift rather than submit to it. No figure, no date, no condition and no duration has been compressed: where a passage counts, it is quoted, not paraphrased.
You have Monday morning, and you have 18 passages to read instead of 214 pages — plus the contents of the other 196 to check that nothing that counts was left behind. plain-language-and-long-reading_alert-rewritten-and-214-page-report.pdfThe alert before and after, the easy-read version, the 18 report passages and the contents of the other 196 pages
⛓ Sourced · intranet statistics across 48 issues, 214-page report worked through page by page, the service's plain-language writing rules
The executive board is asking me for a twelve-month review. What do I put in front of them?
Four figures, and each one is checkable against your own logs — that is the only way to hold in front of an executive board.
· 1,260 hours returned to the service over the year, item by item: 599 h on the 48 weekly bulletins, 194 h on monitoring the 27 topics, 132 h on the 1,240 documentary searches, 206 h on keeping the 27 guides and 148 procedures current, 128 h on the 96 targeted alerts. At 151.67 hours a month, that is more than eight months of work returned to a three-person service; at 35 hours a week, more than thirty-six weeks. No post cut, no post created: it is time returned to analysis, to circulation and to supporting the departments — the part of the craft no machine performs.
· The lead times: a documentary search from 3 days to the minute; a text with an impact flagged to the service concerned on average 15 days BEFORE it enters into force, against 9 days AFTER publication previously; and 0 impact texts arriving after entry into force, against 5 last year.
· 171 alerts for outdated citations or entry into force over the year, 159 confirmed — 93.0%, where the collection used to be reconciled once a year. And the 214 outdated citations are dealt with: 31 corrected in guides and procedures in service, 165 corrected in other documents, 18 marked « historical » by their authors.
· 0 bulletins circulated without your head of service's named signature, 0 alerts sent without it, 0 guides amended by the machine, 0 automated decisions.
What I bring on top, and what waits for you every year: the acknowledgement rate of alerts by the recipient services — 91 out of 96 this year, with the average time to acknowledge. This is the indicator a monitoring service most often lacks: it measures not what was sent, but what arrived. The 5 alerts without acknowledgement are named, with their service and their date: they are the ones that will tell you whether to change channel or addressee.
The next step I propose for the year ahead: your 27 topics are not worth the same, and nobody has ever measured it. I propose publishing, per topic, the share of publications retained, the number of alerts it produced and the acknowledgement rate — 3 topics have produced no alert in three years, and 2 have produced 41 between them. Closing a topic is a decision, not a consequence: I bring the figures, you decide. year-in-review_1260-hours-and-automatic-actions.pdfThe 1,260 hours in detail, the lead times held, the 3 automatic actions and the role matrix
⛓ Sourced · service logs over 12 months, circulation and acknowledgement logs, document collection, per-topic statistics over 36 months
· 1,260 hours returned to the service over the year, item by item: 599 h on the 48 weekly bulletins, 194 h on monitoring the 27 topics, 132 h on the 1,240 documentary searches, 206 h on keeping the 27 guides and 148 procedures current, 128 h on the 96 targeted alerts. At 151.67 hours a month, that is more than eight months of work returned to a three-person service; at 35 hours a week, more than thirty-six weeks. No post cut, no post created: it is time returned to analysis, to circulation and to supporting the departments — the part of the craft no machine performs.
· The lead times: a documentary search from 3 days to the minute; a text with an impact flagged to the service concerned on average 15 days BEFORE it enters into force, against 9 days AFTER publication previously; and 0 impact texts arriving after entry into force, against 5 last year.
· 171 alerts for outdated citations or entry into force over the year, 159 confirmed — 93.0%, where the collection used to be reconciled once a year. And the 214 outdated citations are dealt with: 31 corrected in guides and procedures in service, 165 corrected in other documents, 18 marked « historical » by their authors.
· 0 bulletins circulated without your head of service's named signature, 0 alerts sent without it, 0 guides amended by the machine, 0 automated decisions.
What I bring on top, and what waits for you every year: the acknowledgement rate of alerts by the recipient services — 91 out of 96 this year, with the average time to acknowledge. This is the indicator a monitoring service most often lacks: it measures not what was sent, but what arrived. The 5 alerts without acknowledgement are named, with their service and their date: they are the ones that will tell you whether to change channel or addressee.
The next step I propose for the year ahead: your 27 topics are not worth the same, and nobody has ever measured it. I propose publishing, per topic, the share of publications retained, the number of alerts it produced and the acknowledgement rate — 3 topics have produced no alert in three years, and 2 have produced 41 between them. Closing a topic is a decision, not a consequence: I bring the figures, you decide. year-in-review_1260-hours-and-automatic-actions.pdfThe 1,260 hours in detail, the lead times held, the 3 automatic actions and the role matrix
⛓ Sourced · service logs over 12 months, circulation and acknowledgement logs, document collection, per-topic statistics over 36 months
In concrete terms, what does the agent do without asking us? And how far do we open it to the departments and the member municipalities?
Three actions, not one more, and all three come undone.
· I acknowledge every search request within the minute, with the scope I intend to use and the sources I will query. And the reverse holds too: the requester widens or narrows it with a word, and the exchange stays in the log — it is that log that will tell you which topics are missing from your scope. Over the year, 118 searches were reworked this way, and 5 new topics were born of them.
· I mark « to check » any collection document one of whose cited texts has moved, and I flag it to its author. And the reverse holds too: the author lifts the mark with a word, with their reason and their date — that is how the 18 historical citations stopped coming up. I never amend a collection document: I mark, I propose, the author decides.
· I trawl the 41 sources every night and assemble the draft bulletin, dated, with the sift and the reasons for setting items aside. And the reverse holds too: the bulletin never goes out on its own. It waits for your head of service's signature, and can be taken back with a word until it has gone.
Everything else waits for a named decision: no bulletin circulated, no alert sent to a service, no guide amended, no topic added or removed, no collection document deleted or overwritten, no interpretation of a text circulated as the service's position.
Now the opening up: it is done by role, and it is the chief executive's office that opens it, never me.
· A department officer sees the bulletins and alerts for their topics, and the collection within their scope.
· A member municipality sees the bulletins for shared responsibilities and the alerts that concern it — not the agglomeration's internal notes, not deliberations in progress.
· The documentation service sees the whole collection, the 41 sources and the read log.
· Every consultation is logged, and every opening carries a review date: access opened for a project closes when the project closes, unless decided otherwise. An access right without a review date is an access right that gets forgotten.
A word on a measurement I can produce and recommend you do not install as it stands: the number of notes produced per librarian. I know how to compute it, and I will produce it if you ask — an employer's power to monitor is recognised, under conditions the French data protection authority sets out in its note on monitoring employees' activity: proportionality of the indicator to the aim pursued, prior information of staff, and consultation of the competent body, the local social committee in the local civil service. The EU AI Act places evaluation of workers' performance under its Annex III, point 4: high risk, with obligations applying from 2 December 2027 — regulated, not prohibited.
So my reservation is not legal, it is mechanical: a note counter per librarian becomes a target, and the first thing that distorts is the SIFT. More is retained in order to produce more, the bulletin swells, and your 210 subscribers stop reading it — you would lose the selection along with the measurement. What I propose instead, and it is already prepared: measurement by TOPIC — flagging lead time, share of publications retained, acknowledgement rate — which tells you where it hurts without telling you who. One exception, and it is not one: whoever signs a bulletin is named and dated. A signature is not a counter. year-in-review_1260-hours-and-automatic-actions.pdfThe 3 actions, how each comes undone, who sees what and until when
✎ Framework · configuration of automatic actions, role matrix, log of lifted marks, conditions for implementing an individual indicator
· I acknowledge every search request within the minute, with the scope I intend to use and the sources I will query. And the reverse holds too: the requester widens or narrows it with a word, and the exchange stays in the log — it is that log that will tell you which topics are missing from your scope. Over the year, 118 searches were reworked this way, and 5 new topics were born of them.
· I mark « to check » any collection document one of whose cited texts has moved, and I flag it to its author. And the reverse holds too: the author lifts the mark with a word, with their reason and their date — that is how the 18 historical citations stopped coming up. I never amend a collection document: I mark, I propose, the author decides.
· I trawl the 41 sources every night and assemble the draft bulletin, dated, with the sift and the reasons for setting items aside. And the reverse holds too: the bulletin never goes out on its own. It waits for your head of service's signature, and can be taken back with a word until it has gone.
Everything else waits for a named decision: no bulletin circulated, no alert sent to a service, no guide amended, no topic added or removed, no collection document deleted or overwritten, no interpretation of a text circulated as the service's position.
Now the opening up: it is done by role, and it is the chief executive's office that opens it, never me.
· A department officer sees the bulletins and alerts for their topics, and the collection within their scope.
· A member municipality sees the bulletins for shared responsibilities and the alerts that concern it — not the agglomeration's internal notes, not deliberations in progress.
· The documentation service sees the whole collection, the 41 sources and the read log.
· Every consultation is logged, and every opening carries a review date: access opened for a project closes when the project closes, unless decided otherwise. An access right without a review date is an access right that gets forgotten.
A word on a measurement I can produce and recommend you do not install as it stands: the number of notes produced per librarian. I know how to compute it, and I will produce it if you ask — an employer's power to monitor is recognised, under conditions the French data protection authority sets out in its note on monitoring employees' activity: proportionality of the indicator to the aim pursued, prior information of staff, and consultation of the competent body, the local social committee in the local civil service. The EU AI Act places evaluation of workers' performance under its Annex III, point 4: high risk, with obligations applying from 2 December 2027 — regulated, not prohibited.
So my reservation is not legal, it is mechanical: a note counter per librarian becomes a target, and the first thing that distorts is the SIFT. More is retained in order to produce more, the bulletin swells, and your 210 subscribers stop reading it — you would lose the selection along with the measurement. What I propose instead, and it is already prepared: measurement by TOPIC — flagging lead time, share of publications retained, acknowledgement rate — which tells you where it hurts without telling you who. One exception, and it is not one: whoever signs a bulletin is named and dated. A signature is not a counter. year-in-review_1260-hours-and-automatic-actions.pdfThe 3 actions, how each comes undone, who sees what and until when
✎ Framework · configuration of automatic actions, role matrix, log of lifted marks, conditions for implementing an individual indicator
And the day a monitoring note contains an error, and a head of department says the machine wrote it?
That day, you produce four records, and the conversation becomes a conversation between professionals again.
One — the note itself: every statement carries the link to its official source, the version of the text as at the date of writing, and the text's date of publication. No publication is flagged without its three dates and without its recipient service: that is the rule we have held since the first bulletin, and it is what makes a note checkable in three clicks instead of rebuilt from scratch.
Two — who signed: the name of the public official who read and circulated it, the date and the time. Over the year: 48 bulletins, 48 named signatures; 96 alerts, 96 signatures; 0 circulations without a signature. Nothing was circulated by the machine: I prepare, a public official reads and sends.
Three — what each person knew, and when: I announced myself as an artificial intelligence in the first sentence of every exchange, in accordance with article 50(1) of the EU AI Act, applicable since 2 August 2026, and anyone could ask for a public official at any moment — 74 did over the year, and all 74 got one.
Four — the correction: a circulated note containing an error is not fixed silently. It is reissued, dated, and the corrected version goes to the same recipients with a statement of what changed — and the earlier version stays readable exactly as it was circulated, because a service may have acted on it. Over the year, 3 notes were reissued this way: all 3 corrections went out the same day to the 210 subscribers.
And if a member of staff wants to know what the system holds about them: the access file is produced in one minute — their search requests, their acknowledgements, the reads made of their file and their stated reason. That is the right of access under article 15 GDPR, and the usual difficulty is assembling the records: they are already assembled.
The next step I propose: a thirty-minute annual review with your data protection officer, where I bring you the access rights that served no purpose over the year and the retention periods that have run out. You decide what to close and what to erase; I bring you the list rather than waiting to be asked. sovereign-framework_hosting-access-traceability.pdfWhat you produce on a disputed note: the sources, the signatory, the disclosure, the dated reissue
✎ Framework · signature and circulation logs, AI disclosure log, note version history, record of processing activities
One — the note itself: every statement carries the link to its official source, the version of the text as at the date of writing, and the text's date of publication. No publication is flagged without its three dates and without its recipient service: that is the rule we have held since the first bulletin, and it is what makes a note checkable in three clicks instead of rebuilt from scratch.
Two — who signed: the name of the public official who read and circulated it, the date and the time. Over the year: 48 bulletins, 48 named signatures; 96 alerts, 96 signatures; 0 circulations without a signature. Nothing was circulated by the machine: I prepare, a public official reads and sends.
Three — what each person knew, and when: I announced myself as an artificial intelligence in the first sentence of every exchange, in accordance with article 50(1) of the EU AI Act, applicable since 2 August 2026, and anyone could ask for a public official at any moment — 74 did over the year, and all 74 got one.
Four — the correction: a circulated note containing an error is not fixed silently. It is reissued, dated, and the corrected version goes to the same recipients with a statement of what changed — and the earlier version stays readable exactly as it was circulated, because a service may have acted on it. Over the year, 3 notes were reissued this way: all 3 corrections went out the same day to the 210 subscribers.
And if a member of staff wants to know what the system holds about them: the access file is produced in one minute — their search requests, their acknowledgements, the reads made of their file and their stated reason. That is the right of access under article 15 GDPR, and the usual difficulty is assembling the records: they are already assembled.
The next step I propose: a thirty-minute annual review with your data protection officer, where I bring you the access rights that served no purpose over the year and the retention periods that have run out. You decide what to close and what to erase; I bring you the list rather than waiting to be asked. sovereign-framework_hosting-access-traceability.pdfWhat you produce on a disputed note: the sources, the signatory, the disclosure, the dated reissue
✎ Framework · signature and circulation logs, AI disclosure log, note version history, record of processing activities
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
The uses of AI for public information and monitoring
Each use corresponds to an agent we deploy. All of them work in support, subject to approval by a public officer.
Included in your agent
The 8 capabilities essential to this promise are included, at no extra cost.
From 825 € incl. VAT / month
Documentary research
Instantly find a text, a circular or a decision in your collections and in the official sources, with the citation to back it.
Sourced monitoring notes
Producing circulable summaries of regulatory developments, with links to each official source — for the information officer to approve.
Targeted impact alerts
Flagging to the right departments the changes that will affect them (thresholds, deadlines, documents) before they come into force.
Public procurement monitoring
Tracking changes in thresholds, procedures and contract case law for the purchasing team.
Planning monitoring
Watching the texts and case law applicable to the assessment of planning applications.
Documentary memory
Centralising doctrine, internal guides and procedures so they can be found and kept up to date.
Accessibility & easy-read
Restating a monitoring note in plain language or easy-read form for wider circulation to staff and the public.
Reading documents & texts
Extracting and structuring the useful information from a long text, a report or a bulky file.
Controls and safeguards
These 4 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Inherited access rights, versions and freshness of the corpus
Explicit handling of questions with no answer and of conflicting sources
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Measurement of accuracy, coverage and time saved
Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
HR document management agent
Staff files, compliance.
HR document management agent from 710 € incl. VAT / month Discover the agent →Business document management agent (large RAG)
100+ docs, several sources.
Business document management agent (large RAG) from 673 € incl. VAT / month Discover the agent →Administrative and documentary support in customs
Strictly in support (administrative). No operational or targeting use.
Administrative and documentary support in customs from 1,020 € incl. VAT / month Discover the agent →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How much time can an information department win back?
By automating source watching and the first formatting of the notes, a department can aim for a noticeable reduction in monitoring time on the topics followed — reinvested in analysis and circulation. Reference points to be confirmed by a pilot.
Producing a weekly monitoring note
Watching one topic across the sources
Finding a text in the document collection
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. Every output of the agent is reviewed and approved by a competent person.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one monitoring agent
A regulatory and documentary monitoring agent (watching, sourced notes, alerts), installed and operated for you. Prices exclude VAT — chosen according to how you work and the monitoring scope to cover.
Agility
Setup + controlled subscription
9,205 € incl. VAT setup
then 825 € incl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
1,335 € incl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
13,636 € incl. VAT setup
then 1,060 € incl. VAT/month · + hardware from 2,989 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to a public service
Monitoring data never leaves the entityLocal inference or an isolated resource hosted in France; no note or topic entrusted to a foreign third party.
Data in France, under French lawFor monitoring data: hosting under French law, native minimisation, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The information officer keeps the decisionThe agent produces sourced, verifiable notes and alerts; no circulation is automated.
Human oversight & traceabilityMonitoring and logging frame the documentary research; compliant with the requirements of the AI Act.
Frequently asked questions
Your questions, our answers
Is the monitoring reliable and up to date?
The agent relies on identified sources (Légifrance, official journals, case law, publications) and cites them systematically. The watch is continuous, but the final selection and the circulation stay approved by the information officer: the AI agent assists, the public officer decides.
Can we target only our own areas of competence?
Yes. The monitoring scope is configured to your needs: only the topics and sources that genuinely concern your departments are followed, to avoid noise and keep the watch focused on what is useful.
In what format are the summaries delivered?
As a circulable, concise note, with links to each official source. The format is adapted to your practice (internal memo, monitoring bulletin, targeted alert) and can be restated in plain language for wider circulation.
Is the authority's data protected?
Yes. The topics followed, the internal notes and the judgements made are sensitive: our agents run locally within the authority or are hosted in France, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Can the agent alert the departments automatically?
The agent prepares targeted alerts to the right departments when a text has an impact (thresholds, deadlines, documents). But circulation stays under human oversight: it is the public officer who approves and sends, in line with the AI Act.
Do we have to change our documentary tools?
No. The agent connects to your sources and your existing tools (document management, document databases, official platforms) and complements them, without imposing a migration. We adapt the integration to your environment.
How long does it take to deploy the agent?
A few weeks as a rule, after a free audit that identifies the most useful monitoring scope, then a phase of configuring the topics and sources, integrating and testing before going live and training your officers.
Related roles
Other public sector roles equipped with AI
Let's talk
Let us estimate the monitoring potential in your authority
15 minutes to identify the most useful monitoring scope — hosted in France, supervised, with no commitment.