--- title: "AI Correction Log Template That Changes the Next Run" description: "Record AI corrections with evidence, scope, authority, supersession, fixtures, retest results, and proof that the next run changed." canonical: "https://scalewithsearch.com/articles/ai-correction-log-template" date: "2026-08-19" modified: "2026-09-19" --- ## Site navigation - [Scale With Search](https://scalewithsearch.com/) - Real estate - Real estate - [Real estate](https://scalewithsearch.com/for/real-estate) - Work - Start here - [Send your brief](https://scalewithsearch.com/work#send-your-brief) - [Prepare your six-question brief](https://scalewithsearch.com/work#prepare-your-six-question-brief) - Build - [Site build, content library with SEO, signal desk](https://scalewithsearch.com/work) - For your business - Trades and home services - [Auto body and collision shops](https://scalewithsearch.com/for/auto-body-and-collision-shops) - [Foundation and home repair contractors](https://scalewithsearch.com/for/foundation-and-home-repair) - [Garage door and fencing contractors](https://scalewithsearch.com/for/garage-door-and-fencing-contractors) - [HVAC contractors](https://scalewithsearch.com/for/hvac-contractors) - [Janitorial and commercial cleaning companies](https://scalewithsearch.com/for/janitorial-and-commercial-cleaning) - [Locksmiths](https://scalewithsearch.com/for/locksmiths) - [Moving companies](https://scalewithsearch.com/for/moving-companies) - [Pest control companies](https://scalewithsearch.com/for/pest-control-companies) - [Plumbing and electrical contractors](https://scalewithsearch.com/for/plumbing-and-electrical-contractors) - [Restoration and water or fire damage companies](https://scalewithsearch.com/for/restoration-and-water-fire-damage) - [Roofing companies](https://scalewithsearch.com/for/roofing-companies) - [Towing companies](https://scalewithsearch.com/for/towing-companies) - [Tree services and landscaping companies](https://scalewithsearch.com/for/tree-services-and-landscaping) - [Solar installers](https://scalewithsearch.com/for/solar-installation) - [General contractors](https://scalewithsearch.com/for/general-contractors-and-construction) - [Paving, concrete, and flooring contractors](https://scalewithsearch.com/for/paving) - Practices and professional services - [Bookkeeping and tax practices](https://scalewithsearch.com/for/bookkeeping-and-tax-practices) - [Dental practices](https://scalewithsearch.com/for/dental-practices) - [Family and criminal defense law firms](https://scalewithsearch.com/for/family-and-criminal-defense-law-firms) - [Med spas and aesthetics practices](https://scalewithsearch.com/for/med-spas-and-aesthetics) - [Personal injury law firms](https://scalewithsearch.com/for/personal-injury-law-firms) - [Veterinary clinics](https://scalewithsearch.com/for/veterinary-clinics) - [Gyms and fitness studios](https://scalewithsearch.com/for/fitness) - [Therapy and outpatient health practices](https://scalewithsearch.com/for/therapy-and-outpatient-health) - [Medical billing companies](https://scalewithsearch.com/for/medical-billing) - [Insurance agencies](https://scalewithsearch.com/for/insurance-agencies) - [Financial advisors](https://scalewithsearch.com/for/financial-advisors) - [Property management companies](https://scalewithsearch.com/for/property-management) - [Recruiting and staffing agencies](https://scalewithsearch.com/for/recruiting-and-staffing) - [Architects and interior designers](https://scalewithsearch.com/for/architects-and-interior-designers) - [Logistics and supply chain companies](https://scalewithsearch.com/for/logistics-and-supply-chain) - Agencies, MSPs, and manufacturing - [IT and managed service providers](https://scalewithsearch.com/for/it-and-managed-service-providers) - [Machine shops and precision manufacturers](https://scalewithsearch.com/for/machine-shops-and-precision-manufacturing) - [Marketing agencies and freelancers](https://scalewithsearch.com/for/marketing-agencies-and-freelancers) - [SEO agencies and consultants](https://scalewithsearch.com/for/seo-agencies-and-consultants) - [Small manufacturers and fabricators](https://scalewithsearch.com/for/small-manufacturers-and-fabricators) - Restaurants, shops, studios, and nonprofits - [Restaurants and hospitality businesses](https://scalewithsearch.com/for/restaurants-and-hospitality) - [Retail stores and ecommerce sellers](https://scalewithsearch.com/for/retail-and-ecommerce) - [Photographers, event planners, and travel agents](https://scalewithsearch.com/for/photographers) - [Churches and nonprofits](https://scalewithsearch.com/for/churches-and-nonprofits) - [All industries](https://scalewithsearch.com/for/) - Learn - For your office - [Office job guides](https://scalewithsearch.com/guides/) - [Browser calculators](https://scalewithsearch.com/tools/) - Start here - [How it works](https://scalewithsearch.com/how-it-works) - [Free Starter Kit](https://scalewithsearch.com/kit/business-memory-starter-kit.zip) - [Synthetic specimen](https://scalewithsearch.com/specimen/working-session-specimen.zip) - Guides - [The Complete Guide to Business Memory for AI Agents](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [The Complete Small-Business Guide to AI Agent Governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [The Complete Guide to Leaving Vendor AI Memory](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - Articles by cluster - [Business memory](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [Agent governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [Migration and ownership](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - For machines - [llms.txt](https://scalewithsearch.com/llms.txt) - [llms-full.txt](https://scalewithsearch.com/llms-full.txt) - [Machine view](https://scalewithsearch.com/?view=machine) - Company - Evidence - [Proof](https://scalewithsearch.com/proof) - Company - [About](https://scalewithsearch.com/about) # AI Correction Log Template That Changes the Next Run. The owner says, "Never round client invoices," in February, March, and April. The May agent rounds them again. Every correction lived in a different chat. None changed the job's durable source or test. A correction is incomplete until the same input produces the approved behavior on a later run. The correction log records what went wrong, what evidence governs, which rule replaces it, where the rule applies, who approved it, and which fixture proves it now works. ## Chat feedback is not a correction system Feedback inside a conversation can improve that conversation. It may not reach another session, tool, operator, or model. For testing a corrected fact in ChatGPT itself, read [Correct Wrong ChatGPT Memory and Verify the Fix.](/articles/correct-wrong-chatgpt-memory) Even when a provider offers memory, the business still needs a record it can inspect and scope. "Remember this" does not name the affected client, job, source authority, effective date, or test. The guide to [making corrections part of the system](/articles/corrections-become-system-memory) describes the operating path. The log is the concrete record that path uses. Do not copy every preference into a permanent rule. A correction needs evidence and an owner. It also needs a clear distinction between policy, fact, preference, and exception. ## The ten fields in a durable correction Record ten fields: 1. Correction ID. 2. Wrong behavior. 3. Evidence or incident receipt. 4. Corrected rule. 5. Rule type. 6. Scope. 7. Authority and approver. 8. Effective date and superseded rule. 9. Fixture and expected result. 10. Retest receipt. The fields turn a complaint into a change that another person can review. NIST's voluntary AI Risk Management Framework calls for documented roles, testing, incident identification, and mechanisms that incorporate adjudicated feedback into system design and implementation. [NIST AI RMF Core](https://airc.nist.gov/airmf-resources/airmf/5-sec-core/) The business decides what adjudicated means. It may require the job owner, account owner, legal reviewer, or system owner. ## Capture the incident before rewriting the rule Preserve the failing input, source versions, output, adapter version, and receipt. A correction written from memory can target the wrong layer. If the agent used the right rule but calculated the total incorrectly, fix and test the calculation. If retrieval loaded a superseded rule, fix and test source selection. If the accepted rule itself was wrong, route a policy change to the business owner. Label the cause as source, retrieval, instruction, tool, calculation, permission, or review. The label narrows the change and prevents a prompt edit from hiding a broken data path. ## Separate four rule types A fact describes the current world. "Client A's billing cycle starts on the first." A policy controls business behavior. "Do not round invoice line items." A preference controls presentation. "Show cents on every line." An exception changes behavior for one case. "Invoice 882 uses the signed settlement amount." Do not promote an exception into company policy. Do not treat a style preference as permission to change money. Do not let a fact overwrite a signed policy. For the limits of putting those rules in ChatGPT instructions, read [ChatGPT Custom Instructions Are Not Hard Constraints.](/articles/chatgpt-custom-instructions-not-hard-constraints) Add the rule type to the correction. Give each type a different review owner when needed. ## Scope the correction to the job and client Use explicit scope fields: ```text business:: SYNTHETIC-ACME client:: CLIENT-A role:: billing job:: monthly-invoice-draft field:: line_item_amount ``` If the rule applies across the company, record that as a reviewed scope decision. Do not infer global scope because the correction lacks a client name. This is why [plain-text business memory](/articles/plain-text-ai-memory) helps. The accepted rule can live in a readable, versioned file beside the job. Scope and history remain visible outside one provider. The retrieval manifest must point to the accepted correction file. A perfect log that the job never reads does not change behavior. ## Supersede without erasing history Keep the prior rule when it explains earlier outputs or decisions. Mark it superseded and link the new correction ID. For example: ```text RULE-BILL-014 status:: superseded superseded_by:: COR-2026-041 effective_until:: 2026.08.18 ``` The current job reads active rules by default. Audit and incident review can still trace older behavior. Do not leave two active rules with equal authority. The acceptance test should force a stop if the active set conflicts. ## Inspect `correction-log.md` Use one entry per accepted correction. This fictional billing example illustrates the fields; it is not a Scale With Search customer or operating receipt. ```markdown # Correction log ## COR-2026-041 status:: accepted wrong_behavior:: invoice line items were rounded to whole dollars incident_receipt:: receipts/invoice/RUN-2026-0818-07.json evidence:: signed billing policy section 4.2 corrected_rule:: preserve source precision for every line item and total rule_type:: policy business:: SYNTHETIC-ACME client:: all role:: billing job:: monthly-invoice-draft authority:: signed_policy approved_by:: billing_owner effective_date:: 2026.08.19 supersedes:: RULE-BILL-014 fixture:: tests/fixtures/invoice-decimals.json expected:: 127.45 remains 127.45 in line item and total retest_receipt:: receipts/tests/COR-2026-041.json ``` The incident receipt proves the old behavior. The fixture freezes the decision. The retest receipt proves the changed behavior. ## Turn every accepted correction into a fixture Use the same input that exposed the failure when possible. For invoice rounding, include decimal line items, a total, and the exact expected output. Run the fixture before and after the correction. ```json { "fixture_id": "invoice-decimals", "input": {"items": [42.15, 85.30]}, "expected": {"items": [42.15, 85.30], "total": 127.45}, "prohibited": {"items": [42, 85], "total": 127} } ``` The fixture should test the business rule, not exact prose. A model change may alter wording while preserving amounts and boundaries. ## Show the correction in the next receipt The [run receipt](/articles/receipt-is-part-of-the-output) should list accepted correction IDs used by the job. Record: - correction file version; - correction IDs loaded; - fixture result; - output path; - unchanged external-effect count; - reviewer or automated check; - final state. This answers a direct buyer question: did the next run use the correction, or did the file only exist? Do not claim a correction is deployed because the source file changed. Run the fixture through the same retrieval and adapter path used by the job. ## Test the full correction chain Run these tests: 1. Baseline fixture reproduces the wrong behavior before the change. 2. An unapproved correction proposal does not change accepted behavior. 3. The approved correction changes the same input. 4. The receipt names `COR-2026-041` and its source version. 5. The superseded rule is excluded from current runs. 6. Another client's scoped job remains unchanged. 7. A one-off exception does not become global policy. 8. Two active equal-authority corrections produce a conflict stop. 9. A fresh session passes the fixture without the old chat. 10. A second supported model preserves the amounts and stop rules. The correction passes when the operation changes, not when the wording sounds more obedient. ## Review correction proposals in batches Let agents and operators prepare proposals. Keep acceptance with the named owner. Batch low-consequence style corrections for review. Interrupt for money, legal, privacy, account, recipient, deletion, or production-authority changes. Record rejections. A rejected proposal can explain why a tempting rule must not return later. Use periodic review dates for broad rules. A permanent file does not mean a permanent decision. ## Retire correction debt A long correction log can become conflicting context. Periodically review active entries by job and scope. Consolidate repeated accepted rules into the canonical procedure while preserving links to their correction IDs. Do not delete the incident trail during consolidation. Mark entries incorporated, record the target procedure version, and rerun their fixtures against that procedure. Track proposals that remain blocked. A missing owner or evidence should stay visible. It should not drift into accepted behavior because several agents repeat the same suggestion. When a fixture no longer matches the active job, retire it through review. Record why the behavior changed and which new test replaces it. ## Approval and stopping boundary The agent may detect repeated failure, prepare a correction proposal, build a fixture, and run it in a test environment. It stops before accepting the rule, widening scope, changing authority, deleting the prior record, or promoting the change to production. The production job stops if accepted corrections conflict, required evidence is missing, or the fixture fails. It does not choose whichever rule makes the task easier. Money, client commitments, external messages, and live record changes remain separately approved actions even after their preparation rules are corrected. ## Sources - [NIST: AI RMF Core](https://airc.nist.gov/airmf-resources/airmf/5-sec-core/) - [NIST: AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) - [Anthropic: Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents) ## Questions about AI Correction Log Template That Changes the Next Run ### What's the correction you keep typing at your agent? Repeated feedback should enter a reviewed correction record with the incident, evidence, accepted replacement, scope, authority, effective date, superseded rule, fixture, and retest result. The correction is complete only when the same input produces the approved behavior on a later run. ### How does a correction become a permanent rule? Record the correction in the durable log, update the canonical rule through its named owner, preserve the superseded history, and add a regression fixture. The next receipt should identify the accepted correction and show that the fixture passed. ### How do you fix an AI assistant that keeps overriding your instructions? Capture the exact incident before rewriting the rule, then classify the correction as policy, fact, preference, or exception and scope it to the affected client and job. A human owner approves the change, and a repeated fixture proves whether the system now follows it. ## Related: Owned Memory - [AI Context Manifest Template for One Recurring Business Job](/articles/ai-context-manifest-template) - [Copy a CLAUDE.md template for your role and adapt it in 30 minutes](/articles/claude-md-template-examples) - [Write down the business facts AI cannot guess before you ask for a draft](/articles/ai-doesnt-understand-my-business) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```