Finance needs a monthly report it can recalculate. A maintainer needs to update it after handoff. A manager needs access to the right figures. Define that deliverable before deciding whether you need an editable workbook, a data dashboard or more model capacity.

This guide uses fictional transactions. We ran an integer reconciliation program locally to check the expected result and a conflict case. The Claude prompts are for readers to try in an authorized environment. Account, model, export and sharing workflows still need product acceptance. Check your local purchase page for current terms.

1. Two releases create two purchase questions

Anthropic released Fable 5.2 and Mythos 5.2 on November 9. Fable is available on paid plans under their usage rules. Mythos has separate approved access routes, including verification programs. Measure Fable’s output quality and consumption as part of a subscription comparison, and verify Mythos eligibility separately.

Anthropic releases Fable 5.2 and Mythos 5.2 · Claude Platform Fable 5.2 model guide · Claude Help Fable models by plan

On November 10, Claude Sheets and Whiteboards entered beta and the former Dashboards became generally available as Claude Data. For a fixed file, test formulas and export. For a dashboard that reflects source data, test the connection, query definition and recipient access. Give each path an owner and a deadline.

Claude launches Sheets, Whiteboards and Data · Claude Help getting started with Sheets · Claude Help getting started with Data

Use the same monthly report to compare Fable and your regular model. Hold the input, task and acceptance criteria constant. Record errors, human review time and actual usage charges. Defaults may differ across surfaces, so record the surface and selected effort too.

2. Separate the launch week from regular usage

The launch offer includes the Fable family within up to 50% of the usual weekly allowance for Pro and Team Standard from the November 9, 2026 release through November 16 at 11:59 p.m. PT. After that share is exhausted or the offer ends, continued Fable use draws on usage credits if enabled. Your default model stays selected.

Claude Help Fable 5.2 launch offers · Claude Help Fable models by plan

Check actual eligibility and deductions in Settings > Usage
Plan or seatLaunch periodRegular use
Pro and Team StandardUp to 50% of regular weekly allowance, with 1 reset if eligibleFable uses usage credits
Max, Ultra and Team PremiumEligible subscribers receive 1–3 resets based on the first paid billFable can use up to 50% of regular weekly usage
Eligible legacy seat-based Enterprise PremiumEligible seats can also get 1–3 resetsOrganization settings and regular allowance govern use

Resets have a separate billing cutoff before November 9 at 10:00 a.m. PT. For seats, check organization billing and member join time. Official guidance describes trial exceptions. The Pro and Team Standard reset expires November 16 at 11:59 p.m. PT. Eligible Max, Ultra and Premium resets expire March 8, 2027 at 11:59 p.m. PT. Plan around resets visible in the intended account.

Claude Help Fable 5.2 launch offers

A reset restores Fable usage already spent from the weekly allowance and refills any existing five-hour window. Label your log “launch offer,” “reset used” or “normal cycle,” then check again in a normal cycle. A report completed during the offer verifies that delivery. Repeat work also depends on post-offer billing and waiting time.

Claude Help Fable 5.2 launch offers · Claude Help usage and length limits

3. Measure Fable’s benefit against an accepted deliverable

First, use your regular model on the fixed sample. Require recalculable results, an explanation of duplicates and traceable amounts. Then give Fable 5.2 the same input and look for earlier detection of ambiguity or less rework. If both outputs pass, weigh regular cost, your existing workflow and the time colleagues need to take over.

Claude Platform lists standard Fable 5.2 API prices of $8 input and $40 output per million tokens. Opus 5.5 lists $4 and $20. With the same assumed 60,000 input and 12,000 output tokens, the Fable illustration is 60000/1000000×8 + 12000/1000000×40 = $0.96, versus $0.48 for Opus. This unexecuted token example excludes caching, tools, retries and fast mode. Actual token counts vary.

Claude Platform Fable 5.2 model guide · Anthropic plans and API pricing

In-app use depends on model, task length, tools and effort. Keep app usage, usage credits and the Console API ledger separate. Divide actual spending by accepted reports. A message estimate for a personal plan requires its own observation of that plan.

Claude Help usage and length limits · Claude Help app usage credits

4. Pick an editable workbook or a refreshable dashboard

Choose a deliverable by what the recipient needs
Delivery needProduct and checksSuggested acceptance
Colleagues edit formulas, tabs and detailsClaude Sheets beta exports .xlsx or CSVOpen the export, recalculate and inspect formulas and formatting
View results from a connected sourceClaude Data requires a connector. Certified queries and live sharing are in betaCheck the query, recipient identity and refresh time
Assign tasks in a retrospectiveClaude Whiteboards betaSave agreed assignments in the team’s existing system

Claude Sheets supports editing on web and desktop. The detailed guide describes basic read-only mobile access. It suits static imports, formulas and multiple tabs. If the report needs pivots, protected ranges or Power Query, test in the recipient’s tool. Use CSV for data snapshots. Deliver formulas in .xlsx and reopen the file to recalculate them.

Claude launches Sheets, Whiteboards and Data · Claude Help getting started with Sheets

For a live dashboard, connect an approved source. Sheets cannot pull live connector data or feed a Claude Data dashboard. Claude Data refreshes when someone opens a dashboard and a refresh is due. If an unattended month-end job must deliver figures at a fixed time, validate scheduling and archiving separately.

Claude Help getting started with Sheets · Claude Help getting started with Data

Include template switches in purchase acceptance. Team enables Sheets and Data by default, subject to organization Artifacts settings. Enterprise has Sheets off by default. Data is scheduled to turn on by default November 19, subject to previously saved settings. The Sheets beta has documented HIPAA, CMEK and ZDR limits. Check the exact feature against your organization’s data requirements.

Claude Help getting started with Sheets · Claude Help getting started with Data · Claude Help Artifacts administration · Claude Help Covered Models

5. Prepare a repeatable synthetic report

The CSV uses integer cents and negative refunds. Count a repeated row once when transaction ID, region, kind and amount match. If an ID has different values, stop and list the conflict. Keep the original table, deduplicated details and summary traceable. The maintainer should see which IDs contribute to each regional total.

Fixed synthetic input to copy as CSV
id,region,kind,cents
t1,East,sale,12000
t2,East,refund,-2000
t3,West,sale,8500
t4,West,sale,3000
t4,West,sale,3000
t5,East,refund,-1500
Report creation prompt, unexecuted
Create a monthly report from the synthetic CSV. Keep the raw table and add deduplicated details and a summary by region. Use integer cents and retain negative refunds. Collapse only identical duplicate IDs. If matching IDs have different fields, stop and list the conflict. Return raw and unique row counts, regional and overall net totals, and each formula’s input range.

Reuse the same CSV for both model runs and keep the answer key out of the prompt. Save the first output before comparing the second. Log amount errors, missing formula ranges, files that fail to reopen and unclear rule explanations separately. This distinguishes reasoning problems from spreadsheet and delivery problems.

6. Check amounts and conflicts locally

The Python program uses the standard library and fictional rows in memory. We ran it locally and confirmed 6 raw records, 5 unique records, East 8,500 cents, West 11,500 cents and 20,000 net cents. It also checks that the same ID with a different amount raises a conflict. Use these results to verify an export.

Synthetic reconciliation program, run locally on the fixed input
import csv
import io
import json
from collections import defaultdict

SAMPLE = 'id,region,kind,cents\nt1,East,sale,12000\nt2,East,refund,-2000\nt3,West,sale,8500\nt4,West,sale,3000\nt4,West,sale,3000\nt5,East,refund,-1500'


def reconcile(rows):
    seen = {}
    totals = defaultdict(int)
    for row in rows:
        key = row["id"]
        region = row["region"]
        kind = row["kind"]
        cents = int(row["cents"])
        value = (region, kind, cents)
        if key in seen:
            if seen[key] != value:
                raise ValueError("conflicting_duplicate")
            continue
        if (kind, cents > 0) not in {("sale", True), ("refund", False)} or cents == 0:
            raise ValueError("unexpected_sign_or_kind")
        seen[key] = value
        totals[region] += cents
    return {"unique_rows": len(seen), "by_region_cents": dict(totals),
            "net_cents": sum(totals.values())}


rows = list(csv.DictReader(io.StringIO(SAMPLE)))
result = reconcile(rows)
assert len(rows) == 6
assert result == {"unique_rows": 5, "by_region_cents": {"East": 8500, "West": 11500},
                  "net_cents": 20000}
try:
    reconcile(rows + [dict(rows[-1], cents="-1501")])
except ValueError as error:
    assert str(error) == "conflicting_duplicate"
else:
    raise AssertionError("conflicting_duplicate_was_accepted")
print(json.dumps(result, sort_keys=True))
Revision and review prompt, unexecuted
List the transaction IDs included in each regional total and identify the duplicate row. Save another version, changing only t3 from 8500 cents to 9000 cents. Recalculate regional and overall net totals, show the affected formula ranges and keep the old version for review.

Expected values after the edit are East 8,500 cents, West 12,000 cents and 20,500 total. Give the export to the actual maintainer. Ask them to recalculate from the file and record manual fixes. Passing both runs establishes these criteria for the small sample. Production reports need their own currency, period, tax and access checks.

7. Log errors and rework on one acceptance sheet

Use one record for both runs. Log input version, model, effort, extra usage status, start and finish time and human review minutes. Preserve each original output before editing. Point errors to a tab or query. A change to the transaction ID rule starts a new test. Retain the old result for comparison.

Record the action to change after a failed check
CheckPass conditionInvestigate after failure
NumbersBoth versions match regional and overall totalsDeduplication key, refund signs and formula ranges
FileRecipient opens, recalculates and finds raw rowsExport format, formula compatibility and delivery method
AccessColleagues view or edit by role with traceable sharing scopeAccount, workspace role and connector permission
Regular costActual usage and review time are availablePromotions, resets, extra usage and tool charges

Resolve a disputed business rule before polishing the prompt. Finance might merge exact rows for the same ID while sales groups orders by customer. If the roles disagree, assign an owner and add the agreed metric rule to the input. Fluency cannot settle that decision.

For a future live dashboard, start with approved test data, one query and two audience roles. Log the timestamp and reconciled value after refresh. Decide who takes over if the sharer leaves or connector credentials change. Keep the accepted static report as a fallback.

8. Check who can view, edit and refresh

Use an approved recipient to verify three actions. The viewer should see the total and date, the editor should make an approved change, and the recipient should open the export. Artifacts start private. Team and Enterprise external sharing depends on organization settings. Verify invitations and access levels and record the owner.

Claude Help sharing Artifacts · Claude Help Artifacts administration

Claude Data live sharing serves approved query results using the sharer’s connector access. Recipients can see those results without source-system access. Review totals, filters and refresh scope with a test identity that has minimal privileges. A certified badge identifies a definition. Check the query version and metric rule too.

Claude Help getting started with Data

Manage workspaces, chats and attachments separately for the new and original accounts. Before switching, retain source files, exports and the review sheet. Account exports depend on role and settings and omit original uploads under official guidance. Test with files the maintainer can open and the intended organization permissions.

Claude Help account data exports · ClaudeMax English pricing

9. Turn one success into a renewal decision

Pass numerical and handoff checks, then observe a normal usage cycle. Change one variable at a time to learn whether the benefit came from the model, workspace or added capacity.

  1. Fix rules, input, recipient and budget
  2. Create a recalculable baseline in the current surface
  3. Compare Fable under the same conditions
  4. Reopen the export and check recipient access
  5. Record usage and blockers outside the launch offer
  6. Choose an own-account subscription or workspace seat for the bottleneck
Use one synthetic input to keep purchase evidence comparable

10. Open the local offer that fits your results

If you mainly deliver static workbooks and Pro with Sheets and your usual model passes, observe regular charges first. If session or weekly capacity regularly blocks the same account, compare Max and Ultra. Verified Ultra accounts currently have no five-hour window. Weekly and Fable limits remain. Check Settings > Usage in the account.

Claude Help usage and length limits · Claude Help Ultra plan · Anthropic plans and API pricing

If the bottleneck is handoff or consistent sharing and connector permissions, confirm the organization and role. The store’s ready-made Standard and Premium seat accounts are a separate delivery path. Check the actual workspace, member role, renewal and exit. The 1.5x and 6.25x labels locate store cards. Confirm the official plan and allowance in the delivered account.

Claude Help Fable models by plan · Claude Help Artifacts administration · ClaudeMax English pricing

ClaudeMax English pricing

Before ordering, record account eligibility, seat needs, destination organization and billing together. Check current descriptions, price, availability and service terms. The links open specific offers. Check Anthropic benefits and store delivery terms separately.

ClaudeMax English pricing

11. Recheck delivery in a normal usage cycle

Raise the budget after two consecutive reports can be recalculated, opened and handed over. Record offer status, model, reset time, file version, review minutes and failures. Classify a problem as capacity when a limit interrupts work. Resolve data conflicts, permission gaps and missing formats at their own checkpoints.

Compare accepted reports per month, the actual bill and staff time. Use launch results to assess model fit and normal-cycle data for renewal. If moving to API automation later, check retention, eligibility and the ledger independently. Anthropic defines retention and ZDR exceptions for Fable 5.2 separately.

Claude Platform Fable 5.2 model guide · Claude Help Fable 5.2 launch offers · Claude Help Covered Models