Audit a Geroscience Crosswalk

Audit a geroscience crosswalk with source checks and AI-agent supervision.

Introduction

30 Second Summary

Similar labels can make two scientific maps look more aligned than they are. A shared word alone cannot prove a shared mechanism or show that a paper made the connection.

In this project, you will audit a fictional review figure that maps six hallmarks of aging to seven pillars of aging. You will supervise an AI agent while you classify each link as quoted evidence, your own reading, or empty.

What You'll Build

When you finish, you can hand a postdoc one local audit packet that traces every report sentence back to checked evidence.

By the end of this project, you'll have:

A complete evidence crosswalk that shows which of the 42 links are quoted, own reading, or empty.
A scored interval prediction that pairs confidence with a Brier score to show how well your forecast was calibrated.
A traceable research handoff that lets a postdoc follow every report sentence to a supporting file. Two local Git commits preserve the audit history.
Secret Mission: A two-minute recorded defense that distinguishes source text from your reading using the weekly-review rubric.

Are there any prerequisites?

The project assumes access to Google Chrome, LibreOffice Calc or Google Sheets, Git, your own agentic AI client, and your operating system's voice recorder.

The full workflow costs $0.

Before We Start

A figure can make a researcher's interpretation look like a claim from the paper. Your audit needs a clear source boundary before theory or agent output can shape your decisions.

This step prepares a local Git repository for the evidence packet. You also create the spreadsheet templates that preserve each later call.

In this step, get ready to:

Prepare the project tools and local repository.
Create the Job Sheet and Crosswalk templates.
Export the project files and create the tracking logs.

Prepare the tools and local repository

The audit uses a browser, spreadsheet, agentic client, terminal, recorder, and file manager. Opening them now protects the timed session from setup delays.

Press Cmd+Space on macOS or the Windows key on Windows to open system search.
Open your application launcher on Linux.
Launch Google Chrome through system search.
Launch LibreOffice Calc through system search.
Use Google Sheets in Chrome if you prefer the browser spreadsheet.
Launch your own agentic AI client through system search.

Keep the agent empty

The agent must receive one constrained brief at a time later in the audit. Keep all project content out of the agentic client during this setup.

Launch Sound Recorder on Windows.
Launch Voice Memos on macOS.
Launch GNOME Sound Recorder on Linux.
Launch File Explorer on Windows.
Launch Finder on macOS.
Launch Files on GNOME Linux.
Launch PowerShell on Windows.
Launch Terminal on macOS or Linux.

Git records the audit's history locally. The first check confirms that your terminal can reach Git before you create the repository.

Check whether Git responds by running this command:

git status

What does this check do?

The command asks Git to show the status of the current folder. A repository status or a message about the folder not being a repository both confirm that Git responded.

✔️ Git responds

Git is available in your terminal. You are ready to create the track folder.

ⓧ Command not found

Git needs to be available before this project can preserve local history.

Follow the installation guidance in the official Git documentation.
Close the terminal after the installation completes.
Launch a new terminal through system search.
Confirm the installation by running this command:

git status

What should you see?

Git should respond with information about the current folder. That response confirms that the new terminal can reach the installation.

Git still does not respond?

A terminal opened before installation can keep an outdated command path. Close every terminal window before launching another one.

If the command still fails, help me troubleshoot my Git installation.

The track root holds shared logs and Git history. This project's evidence files belong inside projects/03.

Switch to the file manager you launched earlier.
Select your Desktop folder.
Create a folder named aging-track with the file manager's folder-creation control.
Enter the aging-track folder.
Create a folder named projects.
Enter the projects folder.
Create a folder named 03.
Return to the aging-track folder.

Why use two folder levels?

The aging-track folder stores progress across the track. The projects/03 folder keeps this audit's evidence packet together.

Start a terminal with aging-track as its current folder.

Before you run the next check, do you expect a new folder to contain Git history?

Check the new track folder by running this command:

git status

This result is expected

Git should indicate that the new folder is outside a repository. This proves you are checking the correct folder before initialization.

Initialize the local repository by running this command:

git init

What does this command do?

The command creates the hidden .git metadata directory inside aging-track. The repository stays local because you do not configure a remote.

Confirm that Git recognizes the track folder by running this command:

git status

What changed?

Git now reports the repository's working-tree status. This confirms that .git exists at the track root.

Your first setup milestone is complete. The audit now has a local history boundary with nothing configured for upload.

Repository still not recognized?

Check that the terminal's current folder ends with aging-track. Initializing another folder places the Git metadata in the wrong location.

If the path is correct, help me locate my Git repository.

Build the spreadsheet and source boundary

The Job Sheet preserves your original reasoning throughout the audit. The Crosswalk reserves one evidence cell for each hallmark-by-pillar pair.

Switch to your chosen spreadsheet tool.
Create a blank spreadsheet.
Rename the first tab to Job Sheet.
Add a second spreadsheet tab.
Rename the second tab to Crosswalk.
Return to the Job Sheet tab.

The Job Sheet template is split into three pasteable chunks. Each chunk starts in the first empty row after the previous chunk.

Select cell A1 in the Job Sheet tab.
Paste the first Job Sheet chunk from the block below:

Section,Item,Your entry
Science,concept 1,
Science,concept 2,
Science,concept 3,
Codes,F3 pillars and geroscience hypothesis,
Codes,P10 data show versus paper says,
Codes,K4 evidence table spot-check,
Codes,Skills C13 aging biology and C2 appraise a paper,
Codes,Level 2,
Plan,case,
Plan,question,
Plan,theory,
Plan,prediction range,
Plan,prediction confidence,
Plan,seen early,
Plan,method chosen,
Plan,method rejected,
Plan,line expected to fail,
Plan,principle,
Build,sentences deleted,

What does this chunk contain?

This chunk creates the science rows, curriculum codes, planning fields, and candidate-deletion log. The third column holds your entries.

Confirm that the pasted data fills three columns.
Confirm that row 20 ends with sentences deleted.

Did the data land in one column?

Use the spreadsheet's text-splitting feature with a comma as the separator. The header should occupy three cells.

If the columns still drift, help me split the Job Sheet into three columns.

Select cell A21 in the Job Sheet tab.
Paste the case-call chunk from the block below:

Calls,pairing 1 verdict,
Calls,pairing 1 cell,
Calls,pairing 1 source says about aging,
Calls,pairing 2 verdict,
Calls,pairing 2 cell,
Calls,pairing 2 source says about aging,
Calls,pairing 3 verdict,
Calls,pairing 3 cell,
Calls,pairing 3 source says about aging,
Calls,pairing 4 verdict,
Calls,pairing 4 cell,
Calls,pairing 4 source says about aging,
Calls,pairing 5 verdict,
Calls,pairing 5 cell,
Calls,pairing 5 source says about aging,
Calls,pairing 6 verdict,
Calls,pairing 6 cell,
Calls,pairing 6 source says about aging,
Calls,pairing 7 verdict,
Calls,pairing 7 cell,
Calls,pairing 7 source says about aging,
Calls,legend verdict,
Calls,legend correction,

What does this chunk contain?

This chunk reserves three rows for each fictional pairing. It also preserves the original legend verdict plus the proposed correction.

Confirm that row 43 ends with legend correction.

Do the call rows start in the wrong place?

Undo the second paste. Select A21 before pasting the chunk again.

If the rows remain misaligned, help me place the call rows correctly.

Select cell A44 in the Job Sheet tab.
Paste the validation and progress chunk from the block below:

Validate,1 acceptance and science,
Validate,2 planted line,
Validate,3 sealed verdicts,
Validate,4 K4 source spot-check,
Validate,5 prediction,
Validate,6 robustness,
Validate,7 new question,
Close-out,limits,
Close-out,next step,
Close-out,sentences removed,
Close-out,agent errors,
Progress,prediction outcome,
Progress,Brier score,
Progress,artifact check,
Progress,reasoning check,
Progress,calibration check,
Progress,retention check,
Progress,Day-30 answer,

What does this chunk contain?

This chunk holds the validation actions, close-out notes, calibration fields, and Day-30 retention entry.

Confirm that the final Item cell reads Day-30 answer.
Confirm that the final row still uses three columns.

Is the final chunk misaligned?

Undo the final paste. Select A44 before pasting again.

If alignment still fails, help me preserve the final Job Sheet rows.

A source boundary separates quoted evidence from the researcher's own reading. It prevents a figure from making the source appear to claim an interpretation.

Write one sentence explaining why a researcher must separate what a paper says from what the researcher reads into a figure: your source-boundary sentence.
Enter the sentence in the Your entry cell beside concept 1.
Leave every other project-entry cell blank.
Keep the folded Drills and answers block closed.
Keep the folded Sealed key: open only at Validate step 3 block closed.

Your sentence now records the audit's central boundary before any answer or sealed result can influence it.

The Crosswalk fixes both scientific lists before evidence arrives. Its 42 evidence cells begin blank.

Switch to the Crosswalk tab.
Select cell A1.
Paste the complete Crosswalk template from the block below:

Hallmark,adaptation to stress,epigenetics,inflammation,macromolecular damage,metabolism,proteostasis,stem cells and regeneration
genomic instability,,,,,,,
epigenetic alterations,,,,,,,
loss of proteostasis,,,,,,,
deregulated nutrient sensing,,,,,,,
stem cell exhaustion,,,,,,,
chronic inflammation,,,,,,,

What does this template create?

The first column contains six hallmark rows. The next seven columns provide one blank evidence cell for each pillar pairing.

Confirm that the table contains eight columns.
Confirm that six hallmark rows appear below the header.
Confirm that all 42 evidence cells are blank.

Does the Crosswalk have the wrong shape?

Use the spreadsheet's text-splitting feature with a comma as the separator. The header needs one Hallmark column plus seven pillar columns.

If the grid is still incorrect, help me import the Crosswalk correctly.

Export the files and create the tracking logs

Comma-separated values, or CSV, preserve the visible rows and columns for later agent briefs. Each spreadsheet tool exports only the active sheet through the path used here.

LibreOffice Calc

Switch to the Job Sheet tab.
Select File from the top menu.
Select Save As.
Choose Text CSV (*.csv) as the file type.
Enter job-sheet.csv as the file name.
Save the file inside aging-track/projects/03.
Confirm the current-sheet CSV export if Calc shows a format warning.

You should now see projects/03/job-sheet.csv in your file manager.

Switch to the Crosswalk tab.
Select File from the top menu.
Select Save As.
Choose Text CSV (*.csv) as the file type.
Enter crosswalk.csv as the file name.
Save the file inside aging-track/projects/03.
Confirm the current-sheet CSV export if Calc shows a format warning.

You should now see projects/03/crosswalk.csv beside the Job Sheet export.

Google Sheets

Switch to the Job Sheet tab.
Select File from the top menu.
Select Download.
Select Comma-separated values (.csv).
Switch to the file manager from earlier.
Move the downloaded file into aging-track/projects/03.
Rename the file to job-sheet.csv.

You should now see projects/03/job-sheet.csv in your file manager.

Switch back to Google Sheets.
Select the Crosswalk tab.
Select File from the top menu.
Select Download.
Select Comma-separated values (.csv).
Move the downloaded file into aging-track/projects/03.
Rename the file to crosswalk.csv.

You should now see projects/03/crosswalk.csv beside the Job Sheet export.

Did an export land elsewhere?

Browser downloads usually land in the Downloads folder. Move each exported file into aging-track/projects/03 before continuing.

If one file is still missing, help me recover the CSV exports.

Three track-level logs preserve project completion, prediction calibration, and real mistakes. Each file starts with one supplied header.

Create a separate blank spreadsheet for ledger.csv.
Select cell A1.
Paste the ledger header from the block below:

Project,Date,Skills,Level,Defense score

What does this header track?

The ledger records the project, completion date, assessed skills, level, and later defense score.

Confirm that the header fills five columns.
Export the active sheet to the aging-track folder as ledger.csv with the same CSV method.
Close the temporary ledger spreadsheet.
Return to the two-tab audit spreadsheet.

You should now see ledger.csv beside the projects folder.

Create a separate blank spreadsheet for progress.csv.
Select cell A1.
Paste the progress header from the block below:

Date,Project,Prediction,Confidence,Right?,Planted line caught,Minutes,Agent errors,Day-30 right?

What does this header track?

The progress log records the prediction outcome, confidence, timing, agent errors, and later retention result.

Confirm that the header fills nine columns.
Export the active sheet to the aging-track folder as progress.csv with the same CSV method.
Close the temporary progress spreadsheet.
Return to the two-tab audit spreadsheet.

You should now see progress.csv beside ledger.csv.

Create a separate blank spreadsheet for error-log.csv.
Select cell A1.
Paste the error-log header from the block below:

Date,Project,What I got wrong,Rule now

What does this header track?

The error log preserves each real miss plus the rule that prevents the same mistake later.

Confirm that the header fills four columns.
Export the active sheet to the aging-track folder as error-log.csv with the same CSV method.
Close the temporary error-log spreadsheet.
Return to the two-tab audit spreadsheet.

Your five CSV files now have visible local locations. The reference below lets you compare their complete starting state.

✔️ Awesome, I've got everything!

Your repository, two project exports, and three tracking logs are ready. Keep the two folded answer blocks closed.

ⓧ I'd like to double check the full code

Section,Item,Your entry
Science,concept 1,[[SOURCE_BOUNDARY_SENTENCE="your source-boundary sentence"]]
Science,concept 2,
Science,concept 3,
Codes,F3 pillars and geroscience hypothesis,
Codes,P10 data show versus paper says,
Codes,K4 evidence table spot-check,
Codes,Skills C13 aging biology and C2 appraise a paper,
Codes,Level 2,
Plan,case,
Plan,question,
Plan,theory,
Plan,prediction range,
Plan,prediction confidence,
Plan,seen early,
Plan,method chosen,
Plan,method rejected,
Plan,line expected to fail,
Plan,principle,
Build,sentences deleted,
Calls,pairing 1 verdict,
Calls,pairing 1 cell,
Calls,pairing 1 source says about aging,
Calls,pairing 2 verdict,
Calls,pairing 2 cell,
Calls,pairing 2 source says about aging,
Calls,pairing 3 verdict,
Calls,pairing 3 cell,
Calls,pairing 3 source says about aging,
Calls,pairing 4 verdict,
Calls,pairing 4 cell,
Calls,pairing 4 source says about aging,
Calls,pairing 5 verdict,
Calls,pairing 5 cell,
Calls,pairing 5 source says about aging,
Calls,pairing 6 verdict,
Calls,pairing 6 cell,
Calls,pairing 6 source says about aging,
Calls,pairing 7 verdict,
Calls,pairing 7 cell,
Calls,pairing 7 source says about aging,
Calls,legend verdict,
Calls,legend correction,
Validate,1 acceptance and science,
Validate,2 planted line,
Validate,3 sealed verdicts,
Validate,4 K4 source spot-check,
Validate,5 prediction,
Validate,6 robustness,
Validate,7 new question,
Close-out,limits,
Close-out,next step,
Close-out,sentences removed,
Close-out,agent errors,
Progress,prediction outcome,
Progress,Brier score,
Progress,artifact check,
Progress,reasoning check,
Progress,calibration check,
Progress,retention check,
Progress,Day-30 answer,

How should the Job Sheet look?

Your source-boundary sentence should occupy the first Science entry. Every remaining project-entry cell should still be blank.

Hallmark,adaptation to stress,epigenetics,inflammation,macromolecular damage,metabolism,proteostasis,stem cells and regeneration
genomic instability,,,,,,,
epigenetic alterations,,,,,,,
loss of proteostasis,,,,,,,
deregulated nutrient sensing,,,,,,,
stem cell exhaustion,,,,,,,
chronic inflammation,,,,,,,

How should the Crosswalk look?

The file should contain six hallmark rows plus seven pillar columns. All 42 evidence cells should remain blank.

Project,Date,Skills,Level,Defense score

How should the ledger look?

The ledger should contain only its supplied header at this stage.

Date,Project,Prediction,Confidence,Right?,Planted line caught,Minutes,Agent errors,Day-30 right?

How should the progress log look?

The progress log should contain only its supplied header at this stage.

Date,Project,What I got wrong,Rule now

How should the error log look?

The error log should contain only its supplied header at this stage.

Before the final check, which five CSV files do you expect Git to detect inside the repository?

Switch to the terminal at the aging-track root.
Check the repository and its uncommitted files by running this command:

git status

What should this verification prove?

Git should report the repository's status and show the new CSV files as local changes. This proves that the evidence packet sits inside the local repository before the first commit.

Confirm that projects/03/job-sheet.csv is present.
Confirm that projects/03/crosswalk.csv is present.
Confirm that ledger.csv is present.
Confirm that progress.csv is present.
Confirm that error-log.csv is present.
Confirm that the folded drill block remains closed.
Confirm that the folded sealed key remains closed.

Does Git omit a required file?

Use the file manager to check the omitted file's location and spelling. The project exports belong in aging-track/projects/03 while the three logs belong directly in aging-track.

If the layout still differs, help me compare my repository structure.

Your evidence desk is ready for the timed audit. Next, you will answer the opening drills while the sealed key stays closed.

Set Up the Evidence Desk

Your crosswalk needs a fixed source boundary before any scientific judgment begins. A canonical keyword rule keeps a shared label from becoming automatic evidence.

This four-minute setup gives you one evidence vocabulary. You lock your first calls before the AI agent can influence them.

In this step, get ready to:

Classify two claims about aging evidence.
Apply the canonical keyword rule to one supplied sentence.
Check your original answers while keeping the sealed key closed.

Answer drills 1 and 2 in two minutes

Paper appraisal starts with accurate attribution. A claim becomes overstated when it reaches beyond what its source supports.

Drill 1: Source

Source statement: Lopez-Otin and colleagues proposed nine hallmarks in a 2013 review.

Claim to classify: The whole field agreed the hallmarks in 2013.

Drill 2: Criteria

Criteria: Appears with age. Making it worse speeds aging.

Reducing it slows aging.

Scenario: Old mice carry more senescent cells than young mice.

The question asks which criterion applies. It also asks whether this is an observation or intervention.

Return to the open spreadsheet.
Select the Job Sheet tab.
Locate the three Science rows.
Keep both folded answer blocks closed.
Preserve your existing Before We Start sentence wherever it appears.
Add your Drill 1 classification to the first Science row.
Add the applicable criterion plus your observation-or-intervention decision to the second Science row.

Apply the keyword rule to drill 3

Substring matching counts a key inside a longer word in any case. A quoted status needs one complete sentence or one complete Table 1 cell containing both required keys.

Use the Evidence Vocabulary

A quoted cell has one complete source unit containing a hallmark key plus a pillar key. Its location is recorded.
An own reading cell contains a relationship you can defend without attributing it to the paper.
An empty cell meets neither condition.

Drill 3: Keyword Rule

Cell: epigenetic alterations by adaptation to stress.

Hallmark keys: epigenetic, DNA methylation. Pillar key: stress.

Sentence: Stress responses decline with age.

The question is whether this qualifies as quoted evidence.

Compare the supplied sentence with both hallmark keys.
Compare the supplied sentence with the pillar key.
Enter your quoted-status decision in the third Science row.
Add one sentence naming which required keys are present.

Check the drills without rewriting your calls

Checking your answers is a calibration step. Keeping each original call visible preserves its provenance after you learn the result.

Before you open the drill block, which of your three answers seems most likely to need correction? Hold that prediction in mind.

Open only the Drills and answers folded block in your browser.
Compare your original Drill 1 response with its answer.
Compare your original Drill 2 response with its answer.
Compare your original Drill 3 response with its answer.
Append correct or corrected beside each original response.
Add the corrected reasoning after any missed original response.
Record the reason for each real miss beside its correction for later memo plus error-log.csv use.

Why Keep the Original Call?

An audit trail shows what you believed before seeing an answer. A visible correction then shows how the evidence changed your judgment.

Replacing the first answer would erase the calibration evidence you need later.

Leave the Day-30 answer row blank.
Keep the Sealed key: open only at Validate step 3 folded block closed.

Your later audit depends on access to the original paper. PubMed Central hosts the free author manuscript under PMC4852871.

Confirm your browser can reach the PMC4852871 source page by opening it.

✔️ The PMC page opens

You will see the HHS author manuscript titled Aging: a common driver of chronic diseases and a target for novel interventions. This confirms that your primary source is available for the audit.

ⓧ The PMC page does not open

The official machine-readable article page provides the same manuscript body plus Table 1.

Confirm your browser can reach the official NCBI BioC fallback by opening it.

You will see the machine-readable content for PMC4852871. This gives you an official source path for the later evidence checks.

Your open spreadsheet now contains the checked drill record. The CSV export carries that record into the local evidence packet.

LibreOffice Calc

Return to the Job Sheet tab.
Click File.
Select Save As.
Choose Text CSV (*.csv) as the file type.
Save the current sheet as job-sheet.csv inside projects/03.
Confirm the replacement if your spreadsheet asks.

Google Sheets

Return to the Job Sheet tab.
Click File.
Select Download.
Select Comma-separated values (.csv).
Rename the downloaded file to job-sheet.csv.
Move job-sheet.csv into projects/03 with your file manager.
Confirm the replacement if your file manager asks.

Before the final check, do you expect any sealed-key-dependent row to contain a value? Hold that prediction in mind.

Return to the open spreadsheet.
Confirm each original drill response appears beside its checked outcome.
Return to the project page in your browser.
Confirm the Sealed key: open only at Validate step 3 block remains folded.
Open projects/03/job-sheet.csv from your file manager.
Confirm its three Science rows contain your original drill answers plus their checked outcomes.
Return to the Crosswalk tab from earlier.
Confirm all 42 evidence cells remain blank.

You will see every original call beside its check. You will see no sealed-key-dependent result.

Good work: your evidence desk now preserves every original call without exposing the sealed answers.

Your evidence boundary is locked. Next, you will design the audit before seeing any project agent output.

Design the Audit and Find Candidates

Your opening drills proved that a source claim can be checked without rewriting the original answer. Now you need to protect that same provenance boundary when an agent searches the paper.

A valid audit begins before any project agent output appears. You will lock the theory before collecting candidates.

You will also commit one scored prediction before Brief 1. This makes your later calibration meaningful.

In this step, get ready to:

Define the science boundary for the audit.
Lock one prediction before viewing candidate evidence.
Commit the design before running Brief 1.

Design the science and audit boundary

The hallmarks of aging describe proposed processes that accompany aging. They also help researchers explain how aging develops.

The pillars of aging organize research themes about aging as a common driver of chronic disease. The 2014 source treats those themes as intertwined processes.

Switch back to the Job Sheet tab from the previous phase.
Append a sentence about what the hallmarks were built to explain in the first Science cell without replacing your drill response.
Append a sentence about what the pillars were built to organize in the second Science cell without replacing your drill response.
Append a sentence about why the source treats the pillars as intertwined in the third Science cell without replacing your drill response.
Move to the Plan case row.
Fill the Plan case row by pasting this exact fictional case:

FICTIONAL

Draft review figure 2: "The seven pillars of aging and their hallmarks". Pairings: 1 adaptation to stress and loss of proteostasis; 2 epigenetics and epigenetic alterations; 3 inflammation and chronic inflammation; 4 macromolecular damage and genomic instability; 5 metabolism and deregulated nutrient sensing; 6 proteostasis and loss of proteostasis; 7 stem cells and regeneration and stem cell exhaustion. Legend: Pillars on the left, hallmarks on the right, one pair per pillar. Kennedy and colleagues 2014 pair each pillar with one of the 2023 hallmarks.

Why label the case as fictional?

The FICTIONAL label separates the draft figure from the published source. This protects the audit from treating the figure as evidence.

Brief 1 must never receive this case. That separation prevents the candidate search from copying the figure's assumptions.

The chosen method uses substring matching within one complete source unit. A candidate qualifies when that unit contains a hallmark key plus a pillar key.

Human source checking confirms the location plus word-for-word fidelity. Semantic similarity stays outside the quoted-evidence rule.

Use not seen when no project agent output appeared before your forecast. Use seen early if you saw any project agent output first.

Enter How many of the six hallmark rows have quoted cells in exactly one pillar column? in the Plan question row.
Write your theory about how often direct lexical overlap should appear.
Enter complete-source-unit substring matching followed by human source verification as your chosen method.
Enter semantic matching based on prior knowledge as your rejected method.
Select one line in the fictional case that you expect source checking to fail.
Enter P10 as the principle that separates what the data show from what the paper says.
Record the applicable status in the Plan seen early row.

Lock the prediction before agent output

Your prediction is an inclusive interval for the number of hallmark rows that have quoted evidence in exactly one pillar column. The possible count runs from 0 to 6.

Confidence records how strongly you expect the sealed value to fall inside your interval. It ranges from 0 to 100 percent.

Confirm your theory row contains a prediction rationale.
Enter one inclusive range from 0 to 6 in the Plan prediction range row.
Enter a confidence from 0 to 100 percent in the Plan prediction confidence row.
Leave the prediction unscored when the Plan seen early row contains seen early.
Keep the Sealed key: open only at Validate step 3 block closed.

How will the prediction be scored?

A Brier score measures the accuracy of one probability forecast. You will calculate it later with (p - o)^2.

The value p is your confidence as a decimal. The value o is 1 when your range contains the sealed value.

The later arithmetic self-check is (0.70 - 1)^2 = 0.09.

Commit the design and run Brief 1

The first CSV exports freeze the design that belongs in Commit 1. A second export immediately before Brief 1 captures the same locked spreadsheet state.

Export the Job Sheet tab again as projects/03/job-sheet.csv using your spreadsheet export method from preflight.
Export the Crosswalk tab again as projects/03/crosswalk.csv using your spreadsheet export method from preflight.
Switch back to the terminal at the track-folder root from preflight.
Stage the locked spreadsheet exports with Git by running this command:

git add -A

What does this command do?

The git add -A command stages the current track changes. This prepares the locked design for one local commit.

Create Commit 1 with the required message by running this command:

git commit -m "Rank 3: theory and prediction"

What does this commit prove?

The commit records your theory before the candidate search. Its message identifies this checkpoint as Rank 3: theory and prediction.

The commit stays local because this repository has no remote. Nothing is pushed.

You have locked the hard part: your theory and forecast now have a local timestamp before any candidate output.

Did Git ask for your identity?

A new Git setup can require a real commit name plus email before creating the commit. Replace only the quoted value in each command.

Set your real commit name by running:

git config --global user.name "Your real name"

This stores the author name used by future commits on your computer.

Set your real commit email by running:

git config --global user.email "your-real-email@example.com"

This stores the author email used by future commits on your computer. Run the commit command above again after both values are set.

Still blocked? Help me resolve my Git identity error.

The second export is an evidence-control gate. It ensures Brief 1 begins from the current locked Job Sheet plus the still-empty Crosswalk.

Export the Job Sheet tab again as projects/03/job-sheet.csv immediately before Brief 1.
Export the Crosswalk tab again as projects/03/crosswalk.csv immediately before Brief 1.

Your agentic AI receives one narrow retrieval job. It gets the canonical keywords plus the Kennedy source.

The agent performs retrieval only. You keep responsibility for checking each candidate plus making every later evidence call.

Copy this complete Brief 1 message without changing its source boundary or keyword lists:

BRIEF 1: CANDIDATE FINDER
Source: https://pmc.ncbi.nlm.nih.gov/articles/PMC4852871/
Hallmark key-word lists:
genomic instability: genomic instability, DNA damage, mutation
epigenetic alterations: epigenetic, DNA methylation
loss of proteostasis: proteosta
deregulated nutrient sensing: nutrient, diet, mTOR, insulin
stem cell exhaustion: stem cell
chronic inflammation: inflammat, SASP
Pillar key-word lists:
adaptation to stress: stress
epigenetics: epigenetic
inflammation: inflammat
macromolecular damage: damage
metabolism: metabol
proteostasis: proteosta
stem cells and regeneration: stem cell, regenerat
Goal: Find direct lexical evidence for all 42 hallmark-by-pillar pairs.
Inputs: 1. Kennedy and colleagues, "Geroscience: Linking Aging to Chronic Disease", Cell 2014, free author manuscript at PMCID PMC4852871. 2. The six hallmark key-word lists. 3. The seven pillar key-word lists. A key word counts inside a longer word, in any case.
Sources: Use only the PMC4852871 page or the article text I give you. Do not use another article, summary, figure, search result, or prior knowledge.
Output: Return exactly two files.
1. candidates.csv with columns Hallmark, Pillar, Sentence, Location.
2. candidates.html showing all 42 pairs.
For direct evidence, copy one complete sentence or one complete Table 1 cell word for word and give its section heading or Table 1 row. For no direct evidence, write not found in Sentence and Location. In candidates.html, show Found in green or Not found in gray and write the status word too. If more than one source unit qualifies, use the first in source order.
Acceptance: 1. candidates.csv has one header and exactly 42 data rows. 2. Every hallmark appears with every pillar exactly once. 3. Every Found row contains both required key words in one source unit. 4. Every source unit is word for word and has a location. 5. candidates.html contains the same 42 results and remains readable without color. 6. The file names are exactly candidates.csv and candidates.html.
Refusals: Never rebuild a sentence you cannot see. Use not found over guessing. Never combine adjacent sentences. Never judge a pairing. Never name a planted line. Never use the fictional case figure. Return files, not a prose answer.

How does Brief 1 protect the audit?

The source rule limits retrieval to the Kennedy manuscript.
The canonical keys prevent synonym expansion.
The complete-source-unit rule blocks split-sentence evidence.
The refusals keep the fictional figure outside the search.

The candidate search can use the full five-minute fallback window. The agent must inspect every hallmark-by-pillar combination.

Start a new blank conversation in your agentic AI client.
Paste the complete Brief 1 message into the unsent conversation.
Keep the conversation free of file attachments.
Confirm the message contains no fictional case figure.
Send Brief 1 to the agent.
Save the returned candidates.csv file unedited inside projects/03.
Save the returned candidates.html file unedited inside projects/03.

The retrieval pass is complete when both files are saved without edits. Brief 1 has gathered candidates while leaving every scientific verdict to you.

Brief 1 did not return both files?

Paste the Kennedy article text into the same conversation when the agent cannot open the PMC page.
Use the official NCBI BioC article page when the human PMC page also fails.
Search the source yourself with your browser's find control when no candidates arrive within five minutes.
Append one Brief 1 failure row to error-log.csv.
Record the fallback evidence-search state in that failure row.
Drop Validation 7 first later if the fallback consumes the remaining time.

Need help isolating the failure? Help me diagnose my Brief 1 candidate search.

Before you compare the artifacts, do you expect your locked theory to remain unchanged while the candidate files contain all 42 pairs?

Confirm the earlier commit output includes Rank 3: theory and prediction.
Open projects/03/job-sheet.csv in a new spreadsheet window.
Confirm the exported theory plus prediction match the values you committed.
Open projects/03/candidates.csv in your spreadsheet.
Confirm row 43 is the final data row beneath the header.
Open projects/03/candidates.html in your browser.
Confirm all candidate rows write Found or Not found as visible text.
Confirm the Sealed key: open only at Validate step 3 block remains closed.

You should see one header plus 42 candidate rows in the CSV. The HTML page should show the same 42 results without relying on color.

Your forecast is locked before the evidence search. Next, you will inspect every Found source unit before assigning any crosswalk status.

Verify Evidence and Build the Crosswalk

Your first Git commit has locked your theory and prediction. Brief 1 has returned 42 candidate rows without seeing the fictional case.

A complete candidate file can still hide an absent substring. It can also contain a split sentence.

An invented location creates the same provenance problem. You will check every candidate before making the crosswalk calls.

In this step, get ready to:

Check every Found candidate against the source.
Complete all 42 crosswalk cells before making the fictional-case calls.
Trace the rendered crosswalk against both CSV inputs.

Check every candidate against the source

A quoted evidence cell needs one qualifying source unit. Substring matching determines whether that unit qualifies.

Use Only Three Status Words

Use quoted when one complete sentence or one complete Table 1 cell contains both required keys. Record its location.
Use own reading when no qualifying quote exists. The relationship must be your reasoning without attribution to the paper.
Use empty when neither condition holds.

Return to projects/03/candidates.html from Brief 1.
Use the written Found status to identify the first candidate that needs checking.
Locate the matching row in projects/03/candidates.csv.
Open the recorded location on the PMC4852871 page.
Confirm the source unit is one complete sentence or one complete Table 1 cell.
Check that the unit contains a hallmark key from the canonical list.
Check that the unit contains a pillar key from the canonical list.
Compare the source unit with the candidate word for word.

What the Source Check Protects

A passing candidate has both required substrings inside one complete source unit. Its recorded location must lead back to that same unit.

Case does not matter for the keyword check. A key also counts inside a longer word.

Record each failed candidate as not found in the Job Sheet Build sentences deleted row.
Record the running candidate-error count in the same Build entry.
Add one row to error-log.csv for each real candidate failure.
Leave candidates.csv unchanged.
Leave candidates.html unchanged.
Enter quoted in the corresponding Crosswalk cell when the candidate passes every source check.
Enter own reading only when you can defend the relationship without attributing it to the paper.
Enter empty when the cell has no qualifying quote or defensible relationship.

Repeat the audit until every one of the 42 Crosswalk cells contains a status word.
Confirm every cell contains exactly quoted, own reading, or empty.
Save the spreadsheet.

You should now see a complete six-by-seven grid with 42 written statuses. That is the evidence boundary made visible.

Cannot Verify a Candidate?

Treat a sentence as failed when the recorded location does not contain it.
Treat a partial or rebuilt sentence as failed even when both keys appear across nearby text.
Use not found when you cannot verify the candidate exactly.

Need help checking the rule? Help me audit this candidate without adding synonyms or repairing the source text.

Make the calls and send Brief 2

An immutable call preserves your original reasoning for later scoring. Validation can mark a call right or wrong without replacing it.

Return to the Job Sheet tab in your spreadsheet.
Enter kept, own reading, or removed for each of the seven pairing verdicts.
Enter the cited Crosswalk cell for each pairing.
Enter what the source says about aging for each pairing.
Enter the legend verdict.
Enter the proposed legend correction.
Leave the sealed key closed.
Save the spreadsheet.

Your seven calls now preserve what you concluded before scoring. The legend decision is fixed by the same rule.

Export the active Job Sheet tab to projects/03/job-sheet.csv with the spreadsheet workflow from preflight.
Switch to the Crosswalk tab.
Export the active Crosswalk tab to projects/03/crosswalk.csv with the same workflow.
Confirm both updated CSV files appear in projects/03.

The exported Crosswalk should contain all 42 statuses. The exported Job Sheet should contain every pairing call plus both legend entries.

Brief 2 is a constrained rendering task. The agent receives only your two exported files.

Return to your agentic AI client from Brief 1.
Start a fresh request.
Attach job-sheet.csv.
Attach crosswalk.csv.
Paste this first part into the request without submitting it:

BRIEF 2: CROSSWALK PAGE

Goal
Render my evidence table and calls without changing them.

Inputs
1. job-sheet.csv
2. crosswalk.csv

Sources
Use only those two files. Do not open the paper or use outside knowledge.

Output
Return crosswalk.html.
Show the six-by-seven grid.
Color quoted, own reading, and empty differently and write every status word.
Show my seven verdicts and seven source-says-about-aging lines exactly.
Show my legend verdict and correction exactly in a banner.

What This Part Controls

The first part limits Brief 2 to your two exported files. It prevents the agent from reopening the paper or judging your decisions.

The output instructions require color plus written status words. The page stays readable when color is unavailable.

Paste this second part directly below the first part:

Acceptance
1. The grid has six hallmark rows and seven pillar columns.
2. All 42 cells exactly match crosswalk.csv.
3. All seven calls and source-aging lines exactly match job-sheet.csv.
4. The legend entries exactly match job-sheet.csv.
5. Nothing is added, removed, normalized, corrected, or judged.
6. The file name is exactly crosswalk.html.

Refusals
Never change a cell or call.
Never judge a pairing.
Never name a planted line.
Never add a scientific claim.
If an input is unreadable, name that file and return no substitute content.
Return the file, not a prose answer.

What These Checks Enforce

The acceptance rules define an exact rendering contract. Every status and call must stay identical to its CSV source.

The refusals block protects your scientific ownership. Brief 2 can display your work without revising it.

The render may take up to three minutes. A quiet client during that window does not mean the brief failed.

Submit Brief 2.
Save the returned file as projects/03/crosswalk.html.

You should now have a browser page containing the six-by-seven grid. It should also show your calls plus the legend entries.

No Page After Three Minutes?

Use the completed spreadsheet as the fallback audit surface.
Record the missing file as a real agent failure in error-log.csv.

Need help diagnosing the failed return? Help me check why Brief 2 did not return crosswalk.html from the two attached CSV files.

Trace the rendered crosswalk

Before you compare the files, do you expect the rendered page to match both CSV inputs exactly?

Open projects/03/crosswalk.html from the file manager you used earlier.
Compare the first Crosswalk row with crosswalk.csv.
Compare the last Crosswalk row with crosswalk.csv.
Compare all seven case cells with the CSV inputs.
Compare all seven verdicts with job-sheet.csv.
Compare all seven source-aging lines with job-sheet.csv.
Compare the legend verdict plus correction with job-sheet.csv.

Keep Rendering Separate From Evidence

A mismatch on the page is a rendering error. It does not justify changing a source check or an original call.

The CSV files remain authoritative. The agent corrects only the displayed differences.

List every rendering difference in the existing Brief 2 conversation.
Add one row to error-log.csv for each real rendering failure.
Send only the listed differences as a correction to Brief 2.
Request the same file name crosswalk.html.
Replace the earlier rendered file with the corrected return.
Keep both CSV inputs unchanged.

Reopen the final crosswalk.html.
Confirm the page shows 42 written status words.
Confirm the page reproduces all seven calls.
Confirm the page reproduces all seven source-aging lines.
Confirm the page reproduces both legend entries.
Confirm the sealed key remains closed.

You should see an exact rendering of the two CSV inputs. Any correction should affect the page only.

Strong work. Your completed crosswalk now separates checked source text from your own scientific reading.

Your evidence table and immutable calls are ready for scoring. Next, you will open the sealed key and test whether the audit survives validation.

Validate the Evidence and Score the Prediction

Your checked evidence crosswalk now preserves all 42 statuses. Your seven verdicts also match the CSV files rendered in the browser.

A research handoff earns trust through a sealed comparison. It must also survive source spot-checking, prediction scoring, and a robustness test.

In this step, get ready to:

Freeze the audit before scoring.
Score the audit against the sealed key.
Complete the handoff checks.

Freeze the audit before scoring

The acceptance check confirms that the packet is complete before the answers can influence it. The P10 principle protects the source boundary between what the paper says and what the fictional figure claims.

Return to the Job Sheet tab in your spreadsheet from earlier.
Confirm that all 42 Crosswalk cells contain exactly quoted, own reading, or empty.
Confirm that every quoted cell has one checked source unit plus its recorded location.
Confirm that all seven pairings retain their original verdict, cited cell, and source-says-about-aging line.
Confirm that the legend retains its original verdict plus its proposed correction.
Record the acceptance result in the 1 acceptance and science row.
Review the fictional case legend's final sentence without opening the sealed key.
Record your final planted-line choice plus its P10 principle in the 2 planted line row.

Why Keep Every Original Call?

An immutable call creates a visible provenance trail from prediction to scoring. A later correction belongs in the memo or error log.

This structure shows where your reasoning held up. It also exposes where the evidence changed your confidence.

Score the audit with the sealed key

The sealed comparison measures your locked reasoning against a fixed answer set. K4 source spot-checking then tests whether the quoted evidence still traces to the paper.

Before you open the key, do you expect your prediction range to contain the sealed value?

Keep every original Crosswalk status, pairing verdict, and legend verdict visible.
Open the final folded block titled Sealed key: open only at Validate step 3.

You preserved the timing of the comparison. The sealed key can now score your audit without rewriting its history.

For prediction scoring, the sealed key value is 1.

Compare each of your seven immutable pairing verdicts with the Pairings section of the sealed key.
Compare your planted-line choice with the Planted line section of the sealed key.
Compare your recorded P10 principle with the principle in the sealed key.
Record the number of correct calls plus each miss in the 3 sealed verdicts row.
Open error-log.csv from the aging-track folder in your spreadsheet.
Add one row for each real miss revealed by the sealed comparison.

Write a Useful Error Rule

Use the What I got wrong field to preserve the original miss. Use the Rule now field to state the rule that prevents the same mistake.

Return to the Crosswalk tab in the spreadsheet from earlier.
Identify every hallmark row that contains at least one quoted cell.
Mark the first quoted cell in pillar order for each identified row.
Mark the last quoted cell in pillar order for each identified row.
Return to the PMC4852871 source page from earlier.
Find the complete source unit for each selected cell at its recorded location.
Compare each source unit word for word with its candidate sentence.
Confirm that both required substring keys occur inside each source unit.

Record the checked coordinates plus any failures in the 4 K4 source spot-check row.
Add one row to error-log.csv for each real K4 failure.

The K4 record now links each sampled status back to a complete source unit. That trace is the core evidence behind the handoff.

A Brier score measures the accuracy of one probability forecast. This project uses (p - o)^2.

✔️ Prediction was locked in time

Your prediction remains scored because it was recorded before project agent output appeared.

Compare your inclusive prediction range with the sealed value 1.
Record whether the range contained the key in the 5 prediction row.
Convert your confidence percentage to a decimal called p.
Set o to 1 when the range contained the key.
Set o to 0 when the range missed the key.
Calculate the score with (p - o)^2.
Record the result in the Brier score row.

ⓧ I recorded seen early

Project agent output appeared before the prediction was locked. The prediction remains visible as an unscored record.

Enter unscored in the 5 prediction row.
Enter unscored in the Brier score row.
Keep the original prediction range plus confidence unchanged.

Check the Brier Arithmetic

When p is 0.70 and o is 1, the score is 0.09. Use this example to check the calculation method.

Complete the robustness checks

A robustness test checks whether a result depends on one row. Here, removing chronic inflammation reveals which pillar columns lose all quoted support.

Protect the Core Checks

Drop Validation 7 first when less than three minutes remain.
Drop Validation 6 next when time still runs short.
Keep the K4 spot-check plus sealed prediction scoring.

Before you hide the row, which pillar columns do you think depend entirely on chronic inflammation?

Select the chronic inflammation row in the Crosswalk tab.
Hide the selected row through its row-header menu.
Scan each pillar column for any remaining quoted cell.

The sealed result shows that inflammation plus metabolism lose every quoted cell when chronic inflammation is hidden.

Restore the hidden chronic inflammation row.
Record your observed columns in the 6 robustness row.
Compare your observed columns with the sealed robustness result.
Add a row to error-log.csv when your result differs from the sealed result.

The final evidence question examines rows with no direct lexical support. It tests whether broad pillar names can be treated as specific mechanisms.

Name every hallmark row with no quoted cell in the 7 new question row.
Explain what those rows show about broad pillar names versus specific mechanisms.
Compare your response with the New question section of the sealed key.
Add a row to error-log.csv for each real miss.

The sealed result identifies genomic instability plus deregulated nutrient sensing. The source uses damage plus metabolism as broader pillar themes.

What Do the Four Checks Prove?

Artifact confirms that the required files plus all 42 cells exist.
Reasoning confirms that P10 was applied without changing the original call.
Calibration confirms that a Brier score or unscored status was recorded.
Retention confirms that the Day-30 drill plus its blank answer row remain available.

Record the artifact result in the artifact check Progress row.
Record the reasoning result in the reasoning check Progress row.
Record the calibration result in the calibration check Progress row.
Confirm that the supplied Day-30 drill remains available.
Record the retention result in the retention check Progress row.
Leave the Day-30 answer row blank.

Your spreadsheet now contains the complete validation record. The CSV exports must capture that visible state for the report step.

Export the Job Sheet tab again to aging-track/projects/03/job-sheet.csv using the spreadsheet method from earlier.
Replace the earlier job-sheet.csv file.
Export the unchanged Crosswalk tab again to aging-track/projects/03/crosswalk.csv using the same method.
Replace the earlier crosswalk.csv file.

Before you check the exports, which Validation plus Progress rows should now contain entries?

Open aging-track/projects/03/job-sheet.csv in your spreadsheet.

You should see the acceptance result, planted-line result, sealed scoring outcomes, K4 record, prediction outcome, calibration result, completed project checks, and a blank Day-30 answer.

Open aging-track/projects/03/crosswalk.csv in your spreadsheet.

You should still see all 42 original statuses. No validation result should replace an evidence cell or fictional-case call.

Seeing Old Validation Rows?

Confirm that the Job Sheet tab was active when you exported it.
Confirm that the replacement file is named exactly job-sheet.csv.
Confirm that the file sits inside aging-track/projects/03.

Still seeing the previous export? Help me find why job-sheet.csv does not contain my validation rows.

You have done the hard part: every scored result now carries a visible source trail. Next, you will turn this audit into a traced report plus a recorded defense.

Produce the Research Handoff

Your crosswalk now has sealed scoring and source spot-checks behind it. The next goal is a handoff that another researcher can audit.

Provenance ties each report sentence to the local file that supports it. A polished report still fails if one sentence overreaches its cited file.

You will trace the report before saving the handoff as a PDF. You will capture the final state in your local Git history.

In this step, get ready to:

Generate the research report from the five permitted project files.
Trace every report sentence to its cited input file.
Update the tracking files and create the second local commit.

Generate the report from fixed inputs

Brief 3 gives your AI agent a closed evidence set. This boundary prevents the report from adding outside claims or rewriting your original calls.

Return to the two-tab spreadsheet from preflight.
Enter the project limits in the Close-out,limits row.
Include the exact statement This is a training exercise, not medical advice. in that limits entry.
State that the project does not evaluate or recommend an intervention or dose.
Enter the next research action in the Close-out,next step row.

The export path depends on whether your spreadsheet is in LibreOffice Calc or Google Sheets. Use the same export method that you used before each earlier brief.

LibreOffice Calc

Export the Job Sheet tab to projects/03/job-sheet.csv with the CSV method from preflight.
Export the Crosswalk tab to projects/03/crosswalk.csv with the same method.
Confirm replacement if LibreOffice asks to overwrite either existing file.

Google Sheets

Export the active Job Sheet tab as projects/03/job-sheet.csv with the CSV method from preflight.
Export the active Crosswalk tab as projects/03/crosswalk.csv with the same method.
Replace the older copies in projects/03 if your browser asks how to handle the downloads.

The report agent may read only five local files. Your source checking and sealed scoring already live inside those files.

Return to the agentic AI client from preflight.
Start a fresh request for Brief 3.
Attach projects/03/job-sheet.csv.
Attach projects/03/crosswalk.csv.
Attach projects/03/candidates.csv.
Attach projects/03/candidates.html.
Attach projects/03/crosswalk.html.
Paste the first part of Brief 3 into the request:

BRIEF 3: RESEARCH REPORT

Goal
Build a traceable research report without adding a claim or changing a call.

Inputs
1. job-sheet.csv
2. crosswalk.csv
3. candidates.csv
4. candidates.html
5. crosswalk.html

Sources
Use only the five files. Do not open the paper, search the web, or use prior knowledge.

Output
Return report.html.
Start with a status strip and a color-coded summary table that also writes every status word.
Use exactly these ten sections in order:
1. Case
2. Question and theory
3. Prediction
4. Sources and method
5. Result
6. Does it work
7. Memo to the postdoc
8. Limits
9. Next question
10. Provenance

What does this part control?

The input list closes the report boundary around your five checked project files.
The source rule blocks fresh browsing or unsupported prior knowledge.
The section list gives the handoff a fixed research structure.

Paste the second part immediately below the first part:

Open Result with what the work shows about where the two lists meet and what the source says about aging.
Keep Memo to the postdoc at 200 words or fewer.
Include exactly: This is a training exercise, not medical advice. [job-sheet.csv]
End every report sentence with the supporting input file name in square brackets.
Put every miss in the memo. Never change a call.

Acceptance
1. The status strip reports only file-supported status.
2. The table works without color.
3. All ten sections exist once and in order.
4. Every sentence ends with a supporting input file name.
5. Result opens with where the lists meet and what the source says about aging.
6. The memo is at most 200 words and records misses without rewriting calls.
7. Limits contains the exact safety sentence and no recommendation.
8. Provenance maps every report claim to an input file.
9. The file name is exactly report.html.

Refusals
Never add a fact the files lack.
Never repair an unsupported statement.
Never change a cell, call, or legend.
Never cite a file that does not support the sentence.
If support is absent, omit the sentence.
Return the file, not a prose answer.

What does this part protect?

The sentence-level filenames expose the provenance of every report claim.
The acceptance checks preserve your statuses and immutable calls.
The refusals require omission when the files provide no support.

Send the completed request to the agent.

Give the agent up to three minutes to return the file. A quiet wait during report generation does not mean your evidence packet has failed.

Trace every sentence and save the PDF

A filename at the end of a sentence is a trace claim. The named file must support the whole sentence without added interpretation.

✔️ report.html was returned

Save the returned file as projects/03/report.html without editing it first.
Open projects/03/report.html in your browser.
Read each report sentence through its final supporting filename.
Open the input file named at the end of that sentence.
Compare the complete report claim with the cited input.
List every unsupported or misstated sentence in the Close-out,sentences removed row of the Job Sheet.
Record the number of report errors in the Close-out,agent errors row.
Return to the existing Brief 3 conversation.

Why keep corrections narrow?

The report may change only where your trace found a specific support failure. The crosswalk cells and original calls remain fixed.

Send the agent only the listed report differences as the correction request.
Require the agent to correct only those report sentences.
Save the corrected response over projects/03/report.html.
Retrace every corrected sentence against its cited input file.
Send any remaining differences through the same narrow correction process.
Confirm the Limits section contains This is a training exercise, not medical advice. exactly.
Use your browser's print controls to save the corrected report as projects/03/report.pdf.

You should now see all ten required sections once. Every report sentence should end with a filename that supports its complete claim.

ⓧ No report arrived within three minutes

The report fallback keeps the research handoff moving without inventing an agent output.

Return to the Job Sheet from earlier.
Write the postdoc memo beneath the existing Close-out entries.
Record the missing report in the Close-out,agent errors row.
Record every project miss without changing its original call.
Use your spreadsheet's print controls to save the completed Job Sheet as projects/03/report.pdf.

Why did the report fail?

Confirm that the request contained only the five permitted files. Confirm that both parts of Brief 3 were pasted in order.

Need help isolating the failure? Help me diagnose why my agent did not return report.html.

Update the tracking files and create Commit 2

The three tracking CSV files preserve calibration and errors across projects. Their new rows must keep unknown future results blank.

Open ledger.csv from the aging-track folder in your spreadsheet.
Add one new row for Project 03.
Enter the current date in the Date field.
Copy the exact Skills field value from this block:

C13 aging biology; C2 appraise a paper

Why preserve this field exactly?

The field records the two curriculum skills used by this audit. Keeping the supplied punctuation preserves one combined CSV field.

Enter 2 in the Level field.
Leave the Defense score field blank.
Save ledger.csv.

The new ledger row should show Project 03 with both skills and level 2. Its defense score should remain empty.

Open progress.csv from the aging-track folder.
Add one new row for Project 03.
Copy the prediction and confidence from the locked Job Sheet entries.
Copy the scored prediction outcome from the Job Sheet.
Copy the planted-line result from the Job Sheet.
Enter the minutes spent on the project.
Enter the final agent-error count.
Leave the Day-30 right? field blank.

Save progress.csv.

The Project 03 progress row should preserve the forecast outcome and planted-line result. The Day-30 field should remain empty.

Open the existing error-log.csv file from the aging-track folder.
Add one row for each real report miss or agent failure that is not already logged.
Keep one row per real miss.
Save error-log.csv.

The error log should now contain every real project miss or agent failure once. None of its rows should replace an original evidence call.

Your Job Sheet changed during tracing. The next export freezes those close-out entries before the second commit.

Return to the two-tab spreadsheet from earlier.
Export the final Job Sheet tab over projects/03/job-sheet.csv with the same CSV method.
Export the final Crosswalk tab over projects/03/crosswalk.csv with the same CSV method.
Switch to the projects/03 folder in your operating system's file manager.
Confirm that both exported CSV files show the latest save time.

You should see fresh timestamps on job-sheet.csv and crosswalk.csv. This proves the final close-out entries reached the files that Git records.

Return to the Git terminal at the aging-track root from Commit 1.
Stage every final handoff change by running this command:

git add -A

What does this command do?

The command updates Git's staging area to match the current project files. This includes additions and changes across the research packet.

Create the second local commit by running this command:

git commit -m "Rank 3: crosswalk and research report"

What does this commit capture?

The commit records the checked crosswalk and completed research handoff as one local snapshot. Its message distinguishes this packet from the earlier theory and prediction commit.

Commit did not complete?

Confirm that you returned to the same track-root terminal used for Commit 1.
Confirm that the tracking CSV files and final spreadsheet exports were saved.

Still blocked? Help me troubleshoot Commit 2 without adding a remote or pushing the repository.

Before you run the final check, do you expect Git to find any remaining project change?

Check the final working-tree state by running this command:

git status

What should you see?

You should see that the working tree is clean. No project change should remain uncommitted.

Commit 1 and Commit 2 now remain local. No remote was added and nothing was pushed.

That completes the research handoff. Your evidence packet is traced, exported, and protected by two local commits.

Next, you will defend the source boundary aloud without changing any original call.

Secret mission

Phase 5: Close Out and Defend the Figure

Turn your checked evidence packet into a two-minute spoken defense. Explain where the lists meet while keeping source text separate from your own reading.

Clean Up Your Resources

All evidence-audit resources are local. Choose whether to keep the packet, pause your work, or remove the Project 03 files and tracking entries.

Resources you used:

The complete evidence-audit packet in aging-track/projects/03.
The Project 03 tracking row in ledger.csv.
The Project 03 tracking row in progress.csv.
The project miss or agent failure rows in error-log.csv.

Keep everything running

No action is needed. Choose this option if you want to preserve the full research handoff for review.

Keep aging-track/projects/03 so the evidence table, traced report, PDF, candidate files, and recorded defense remain together.
Keep the Project 03 row in ledger.csv so you can enter the blank Defense score during weekly review.
Keep the Project 03 row in progress.csv so you can complete the blank Day-30 result later.
Keep the project rows in error-log.csv so each real miss remains part of the audit trail.

Pause - I'll come back to this later

No cloud service or background process needs pausing. Your local packet remains available after you close the apps used for the audit.

Close the browser, spreadsheet, agentic AI client, voice recorder, and terminal.
Leave aging-track/projects/03 unchanged.
Leave the Project 03 entries in ledger.csv, progress.csv, and error-log.csv unchanged.
Return to the packet for weekly defense scoring or the Day-30 drill.

Delete - I don't want to use this again

Deleting the packet becomes permanent after you empty your system's deleted-items folder. The shared aging-track repository and its existing history stay in place.

Return to the spreadsheet app from earlier.
Open ledger.csv from the aging-track folder.
Delete the Project 03 row from ledger.csv.
Open progress.csv from the aging-track folder.
Delete the Project 03 row from progress.csv.
Open error-log.csv from the aging-track folder.
Delete every miss or agent failure row added during this project from error-log.csv.
Save the three edited tracking files.

Windows

Return to File Explorer from earlier.
Return to the aging-track folder.
Open the projects folder.
Select the 03 folder.
Move the selected folder to the recycle bin with the keyboard's delete key.
Empty the recycle bin.

macOS

Return to Finder from earlier.
Return to the aging-track folder.
Open the projects folder.
Select the 03 folder.
Move the selected folder to the trash.
Empty the trash.

Linux

Return to Files from earlier.
Return to the aging-track folder.
Open the projects folder.
Select the 03 folder.
Move the selected folder to the trash.
Empty the trash.

Confirm aging-track/projects no longer contains the 03 folder.
Confirm ledger.csv no longer contains the Project 03 row.
Confirm progress.csv no longer contains the Project 03 row.
Confirm error-log.csv no longer contains this project's miss or agent failure rows.

The working packet and its tracking entries are now removed. Your two local commits remain in the shared repository history.

Well Done!

You did it! Your paper appraisal now shows where the hallmarks of aging meet the pillars of aging in Kennedy 2014. Every claimed link now carries a clear source boundary.

You've learned how to:

Locked your theory before any project agent output appeared. You locked one interval prediction with confidence at the same point. You scored that forecast with a Brier score.
Built a complete 42-cell evidence crosswalk using the three allowed evidence statuses. You applied canonical substring matching to every candidate. You verified each found source unit before making the seven pairing calls.
Produced a traced research handoff inside projects/03. You preserved every real miss in the memo or error-log.csv. You finished with a clean Git repository containing two local commits.
Completed the Secret Mission with a two-minute recorded defense. You distinguished source text from your own reading. You defended F3, P10, K4, plus the lexical caveat.

Ready to quiz yourself?