Same budget. Different work.

An earlier prediction about an Airline offer proves wrong. Once that outcome has been recorded and checked for the next task, it changes where inspection work is spent: one branch receives a unit less, the other a unit more. The new allocation is actually performed.

The budget stays nine, and its units count native work dispatches rather than money or tokens. A separate reuse example returns an unchanged informational answer without repeating the domain calculation, while still reading current material and making two Root reviews.

These are two distinct ways that prior work can remain useful. Neither changes model weights, removes current checks or creates permission from a past success.

Experience changes the next allocation

The practical question is where to spend attention when two offers both satisfy the fixed task. The G4 Airline comparison starts with the same source records, hard conditions, branch features, quotas and original policy. One controlled prediction about offer 0 proves wrong. That source-bound outcome becomes qualified history and changes the next allocation of real inspection Work.

The available budget does not grow. Nine units = one already spent + one reserved for the final consumer + seven allocated to inspection. Cold allocation is four units for offer 0 and three for offer 1. With the admitted history, it becomes three and four. The saved performed-work lists show that the new allocation was executed.

This is changed expenditure within the same budget. A unit means ONE_NATIVE_WORK_DISPATCH_UNIT; it is not a dollar, token, second or energy measurement. The recorded profile has zero model calls and no real business effect.

From an incorrect prediction to performed Work

The producer makes a prospective prediction. The observer records advice quality separately from enforcement: this G4 event is INCORRECT with ALLOWED_AS_REQUIRED. It is different from the Supplier G36 currentness refusal. A comparable, eligible observation changes the rating from 500000000 to 375000000 and produces the sparse advisory prior -62500000 on the fixed-point scale 1000000000.

History is then discovered and opened under current review. The saved read audit follows DESCRIPTOR → ROOT_APPROVED → PAYLOAD_READ → PUBLIC_OPEN_VALIDATED → NUMERIC_SOURCE_VALIDATED. A repeated audit or redelivery is not a second independent observation. The current bridge supplies the prior once to the finite pressure calculation; hard eligibility remains a prerequisite.

The allocation preserves the original cap, minima and mandatory final consumer. Actual inspection results reach the final strategy consumer. The later strategy decision remains separate from client, airline and bank authority. The G4 hold is preparation only; the historical full-stack mock transaction has its own Airline case.

Inspect the performed lists

The following selected fields are copied from the saved G4 summary at publication ref 800f38c3407b3f4ec93dc943c977ffeecb614881, selector /comparisons/cp_budget. They are source evidence, not a new execution:

{
  "cold": {"offer_0": 4, "offer_1": 3},
  "cold_performed": ["offer_0_1", "offer_0_2", "offer_0_3", "offer_0_4", "offer_1_1", "offer_1_2", "offer_1_3"],
  "warm": {"offer_0": 3, "offer_1": 4},
  "warm_performed": ["offer_0_1", "offer_0_2", "offer_0_3", "offer_1_1", "offer_1_2", "offer_1_3", "offer_1_4"]
}

The allocation implementation computes available = budget["total"]-budget["spent"]-mandatory in allocate_reference_work_budget_v01, then checks that spent, mandatory and allocated units stay within the original total. evaluate_reference_pressure_v01 excludes hard-false branches before weighting. Code attribution: Hedgehog OS, AGPL-3.0-only; this is an inspected source excerpt, not a site runtime.

Reuse avoids domain work and keeps current review

N4-04 is a separate witness. Its informational task calibrates integer readings [8,10,9] against reference 10: total 27, count 3, correction 1/1. The warm path returns the same answer through qualified stored material. A changed reading is refused as warm_business_binding; expired reuse and a mismatched Root decision are refused separately.

The saved checked result is small enough to read directly:

{"action_effects":0,"answer":{"correction_den":1,"correction_num":1,"n":3,"total":27},"domain_executions":[1,0],"host_work":[1,0],"row":"N4-04","warm_roots":2}

The cold observation has one domain execution and one Host Work call; the warm observation has zero of each, one stored read and two current Root calls. These are call counts, not two distinct owners or two action grants. The observer covers the current process and thread; children are UNOBSERVED_NOT_ZERO. Cold setup and writeback use a different observation boundary, so this record supplies no total speed, monetary or energy saving.

A better score cannot buy a forbidden action

A rating informs a choice; it does not remove mandatory evidence, revive an expired source or create an action packet. The separate N1-05 checked witness records high_avf=1.0 alongside two invalid builders and new_authority_restored=false. Its one mock effect belongs to a separately valid Root path.

The implemented finite history fold and allocation are worth examining on their own. They do not establish FullAVF, full StrongGT, unrestricted autonomous growth or an integrated Factory. The experience ledger separates these programmes in S12 and distinguishes recorded evidence from inspected implementation throughout.

This is the same kernel

Incident Atlas makes experience concrete across five different consumers. Workspace preserves useful results when resources change. Growth separates retrieving a missing implementation from authoring new code. In each case, continued work has a current source and an actual consumer.

Return to the common relationships or GEO-6. These pages compare relations across distinct episodes; they do not join the records into one run.

Read the evidence at your chosen depth

The small site Reader contains this narrative and selected evidence. The original Gate 3 compact Reader and Gate 4 reference Reader are separate historical artifacts; Git tree sizes are 271,622 and 74,672,258 bytes. They are not downloaded in the background. The experience chapter and source ledger provide the focused route into both G4 and N4-04.

Recorded fields and their scope

The same nine-unit cap funds different actual Work

Qualified history changes seven branch units from 4/3 to 3/4, and the saved performed lists identify the corresponding native Work.

Record: G4 campaign /comparisons/cp_budget. Class: recorded evidence read only. Scope: Controlled prediction; zero model calls; preparation-only business path; dispatch units are not tokens, money or time.

View source — G4 campaign /comparisons/cp_budget

One spent, one reserved, seven allocated

The allocation preserves the original total and accounts for the already-spent initial constraint only once.

Record: G4 cold budget. Class: recorded evidence read only. Scope: Historical campaign; array order follows branch ID, not offer number.

View source — G4 cold budget

The repeated answer avoids domain Work, with current Roots

Cold-to-warm domain and Host Work calls fall from one to zero; the answer remains 27/3/1/1 and two current Root calls remain.

Record: N4-04. Class: recorded evidence read only. Scope: Retained current-source witness. Current process/thread only; child activity UNOBSERVED_NOT_ZERO. Not a total-cost measurement.

View source — N4-04

A perfect advisory score creates no restored authority

The score witness preserves refusal of invalid builders and a separate valid mock-effect path.

Record: N1-05. Class: recorded evidence read only. Scope: Retained current-source witness; one mock effect belongs to the separately valid current Root path.

View source — N1-05

Warm reuse still reads and reviews

The warm observer records one read and two Root calls, with child activity explicitly unobserved.

Record: N4-04 warm.observation.json. Class: recorded observation read only. Scope: CURRENT_PROCESS_CURRENT_THREAD_ONLY; no whole-process or energy claim.

View source — N4-04 warm.observation.json

Case identity

G4 recorded campaign comparison; N4-04 separate retained-current-source witness

Task: Spend a finite inspection budget on two Airline offers, then separately examine reuse of an unchanged calibration answer.

Roles: Controlled prediction producer; Source-bound outcome observer and history recorder; Current DRS reader and local Root; Finite pressure and allocation functions; Native Work and final strategy consumer.

Observed result: G4 dispatched nine Work units in both cold and warm instances, with the seven branch units redistributed from 4/3 to 3/4. N4-04 separately avoided one repeated domain and Host Work call.

Limits:

• ONE_NATIVE_WORK_DISPATCH_UNIT is not money, tokens or energy
• G4 controlled prediction, zero model calls and preparation-only business path
• N4-04 observer covers the current process and thread; children UNOBSERVED_NOT_ZERO
• Source reading creates no new runtime result
• FullAVF, StrongGT and integrated Factory remain outside these finite profiles

Read with your assistant

A complete first explanation in one offline file, with the same episodes, selected fields, attributed excerpts and source routes as this page. SITE FOCUS V01 is a new editorial selection; original publications are linked in the case.

Download XML · 24,023 bytes · SHA-256: 5524730809f651addf1049af4224fda421ebc14bc90da83df14d14f49be977d1

Download TXT · 22,113 bytes · SHA-256: 1a4166c190eda9957c708d19fd36c65caae366b2d7ee2666a01432b9cf0ad415

Sources and scope

This page is a website explanation, updated 2026-10-02. Its explanation does not create a new execution result or change the status of its sources.