AILSA SubSwitch 1.0.0(l)

AILSA SubSwitch 1.0.0(l)

Build 2026102003. Reference API prices for DeepSeek V4.1 Flash, Hy3 and
Step 5 Preview now appear in usage dashboards, historical views and exported
Daily, Weekly and Monthly usage images, including the WorkBuddy routes.

USD per million tokens, checked against official documentation on October 20, 2026:

Model	Uncached input	Cached input	Output
DeepSeek V4.1 Flash, off-peak	$0.15	$0.003	$0.60
DeepSeek V4.1 Flash, peak	$0.30	$0.006	$1.20
Hy3	$0.132	$0.033	$0.528
Step 5 Preview	$1.00	$0.10	$2.70

• Recognize WorkBuddy's deepseek-v4.1-flash model ID and calculate each
  request from its original timestamp. DeepSeek peak hours remain 01:00–04:00
  and 06:00–10:00 UTC on weekdays, excluding Chinese public holidays.
• Use StepFun's own $0.10 cached-input rate rather than the $0.05 reseller rate.
  Cache-hit input is a subset of input; reasoning is already part of output.
• Recalculate historical reference amounts when records are read, without
  changing raw Token counts, stored routing identities or duplicate handling.
• Show consistent Hy3 and Step 5 Preview names. Priced rows remain visible
  when the optional unpriced-model filter is enabled for sharing images.

These are equivalent API reference amounts, not WorkBuddy subscription charges
or invoices. A free WorkBuddy multiplier does not set the reference API rate to
zero. Unreported usage, invalid token breakdowns and unknown model identities
remain unpriced. Existing account and usage layouts are unchanged.

Sources: DeepSeek,
Tencent Cloud,
StepFun.

The app is ad-hoc signed for Apple Silicon, requires macOS 14 or later, and is
not notarized. Existing accounts and saved preferences are retained.

Validation

• 411 tests passed, 0 failed across 55 classes on M1 using the Command Line Tools asserting XCTest-compatible harness. This is not Apple XCTest; Swift Testing suites are not run by it.
• Pricing tests cover vendor cache rates, reasoning subsets, request-time DeepSeek peak/off-peak pricing, incomplete usage, unchanged Token totals, duplicate records and historical image exports.
• Replayed 132 frozen local ledger records through both production readers. Raw Token totals and route identities were identical; every newly priced amount matched an independent per-request calculation. The sanitized fixture stays local.
• Source hashes, app/widget signatures, ZIP checksums and public/private package boundaries were verified.

AILSA SubSwitch 1.0.0(k)

AILSA SubSwitch 1.0.0(k)

Build 2026102001. Control which model details appear in shared usage images.

• Add Settings → Quota display → Usage sharing images → Hide models without
  a USD amount. The option is off by default and is saved across launches.
• Apply it to Daily, Weekly and Monthly PNG exports. Hide only model rows whose
  USD amount is unavailable; models with an explicit $0.00 stay visible.
• Preserve complete Token totals, reference-USD totals and trends. Images state
  when model details have been hidden, including periods with no priced models.

This is an export presentation preference. It does not change account cards,
usage dashboards, stored records, provider prices or switching behavior.

The macOS package is ad-hoc signed for Apple Silicon, requires macOS 14 or later,
and is not notarized. Existing account data and settings are retained.

Validation

• 406 tests passed, 0 failed across 54 classes on M1 using the Command Line Tools asserting XCTest-compatible harness. This is not Apple XCTest; Swift Testing suites are not run by this harness.
• Native offscreen rendering checked daily, weekly and monthly images with the filter on and off, an all-unpriced period, and the Settings panel.
• All 11 localization files, source hashes, app/widget signatures and the ZIP checksum were verified. Public source and package checks found no private integrations or account configuration.

AILSA SubSwitch 1.0.0(j)

AILSA SubSwitch 1.0.0(j)

Build 2026101801. Faster native Codex history reading and more accurate
Antigravity subscription labels, adapted from relevant CodexBar 0.77.0 changes.

• Skip unrelated conversation bodies before decoding native Codex usage records.
  Bounded byte scanning reduces reader overhead while preserving cumulative
  token totals, cache coverage, Unicode metadata and incomplete-file handling.
• Prefer explicit Antigravity paid-tier metadata. Show PAID when the paid tier
  is known but its specific plan is not. Suppress a conflicting FREE label
  beside measured five-hour quotas without guessing Pro or Ultra.
• Keep measured reset times in widgets and preserve the existing first-request
  evidence rule. Unknown or disabled quota buckets do not invent capacity.

The M1 Command Line Tools asserting compatibility harness passed 404 tests
across 54 classes, with zero failures. Coverage includes the native/OAuth plan
paths, stored-plan fallback, identity mismatch, disabled/unmeasured windows,
widget reset projection, byte-search boundaries and cumulative usage parsing.
This harness is not Apple's XCTest runtime; Swift Testing suites are not run by it.

A 23 MB synthetic reader/parser benchmark completed in about 24 ms, compared
with 157 ms before the change at the median of three paired runs. All runs
returned the same record count and fixture token total. This is not a claim
about overall app responsiveness or production CPU usage.

Account credentials, provider pricing and automatic-switching rules are
unchanged. The app remains ad-hoc signed for Apple Silicon, requires macOS 14
or later, and is not notarized. Live provider behavior depends on the metadata
returned by the provider; injected-response tests do not replace a fresh
provider-side acceptance test.

See the upstream applicability review for the
changes adopted, already covered, or outside ASS's current provider support.

AILSA SubSwitch 1.0.0(i)

AILSA SubSwitch 1.0.0(i) improves usage inspection, native history reading and menu-bar reopening while preserving the existing account and dashboard layouts.

Changes

• Cached-input reuse: usage details now show a token-weighted cache reuse rate and the number of reported cache samples. Missing metadata stays unknown instead of appearing as a measured zero. The statistics follow the selected period or hovered interval.
• Bounded native history reading: native Codex JSONL is read in fixed-size chunks with a bounded line buffer. Cumulative token deltas remain intact; native session IDs are recognized. Oversized context rows cannot inherit a stale model identity.
• Reliable menu-bar anchoring: popup presentation waits for a verified button on an actual display. The repair result reflects usable placement; hidden or off-screen menu-bar items receive recovery guidance.
• Public edition boundary: remove a legacy local-deployment price mapping. Official DeepSeek API pricing stays supported, while machine-specific equivalences remain private.
• SDK packaging: both the app and widget retain the selected macOS SDK in their link metadata. Package creation verifies this while retaining macOS 14 as the minimum.

These changes adapt relevant CodexBar 0.74.1 improvements. See the integration notes for the exact scope.

Validation and download

• Public build: 395 checks passed, 0 failed, across 52 classes on M1 using the asserting XCTest-compatible Command Line Tools harness. This is not Apple's XCTest runtime.
• The public package excludes machine-local extensions, account files and credentials.
• Apple Silicon ZIP, matching app/widget versions, SHA-256 checksum and source manifest. Ad-hoc signed; Developer ID signing and notarization remain unavailable.
• Extract the ZIP, move AILSA SubSwitch.app into Applications, then open it and click the menu-bar icon. Existing account data stays in place when updating.

Build: 2026101301. User-facing version: 1.0.0(i).

AILSA SubSwitch 1.0.0(h) · 2026100502

AILSA SubSwitch 1.0.0(h)

Build: 2026100502. Reliable automatic account and usage refresh.

Account freshness

• Reopening the menu refreshes visible Codex and Antigravity subscriptions,
  even while the background worker is checking the active account. Cards
  already being refreshed reuse that in-flight request.
• Returning to Accounts or selecting Codex / Antigravity refreshes the
  relevant subscriptions without requiring the refresh button.
• Cursor rechecks the current native sign-in and quota on every presentation,
  including when an earlier account snapshot is already available.
• Claim the refresh before the initial store load, preventing simultaneous
  presentation callbacks from requesting each subscription twice.
• Keep the previous quota visible during a read; on failure, retain its last
  successful update time and show the provider error.

Usage Management

• Opening Usage Management or changing its provider reads fresh usage.
• Selecting Today, 7 days or 30 days refreshes the selected provider immediately,
  including reselecting the current range. Hover context resets to the range.
• Automatic Cursor refreshes reuse older cached history and fetch the recent
  tail. The explicit refresh button still reloads the complete history.
• Concurrent refreshes for the same provider coalesce. A failed read retains
  the previous snapshot and timestamp rather than implying fresh data.
• Codex reads recorded local usage; Antigravity shows available quota snapshots.
  A refresh does not invent missing provider records or activate a quota window.

Validation and distribution

The M1 command-line test runner passed 381 tests across 50 classes, with
zero failures. A separate local variant passed 421 tests across 56 classes.

Regression coverage includes reopening with a recent cache, inactive accounts
while an active refresh is running, simultaneous initial presentations,
provider selection, repeated range selection, Cursor quota refreshes and
failure recovery without changing the last successful update time.

This public build excludes the maintainer's private OpenCode usage import,
local ATC variant features, Hub service and machine configuration. Existing
usage accounting, chart layout and pricing are unchanged.

The Apple Silicon package is ad-hoc signed and is not notarized. The portable
asserting test runner is not Apple's XCTest runtime. Clean-machine installation
and other users' live-provider environments are outside this validation scope.

AILSA SubSwitch 1.0.0(g) · 2026100101

AILSA SubSwitch 1.0.0(g)

Build: 2026100101. Add GPT-6.1 Sol usage and equivalent API cost accounting.

GPT-6.1 Sol usage

• Recognize the exact openai/gpt-6.1-sol route and display GPT-6.1 Sol
  separately from GPT-6 Sol. Preserve reported input/output tokens and cache
  details without counting cached or reasoning tokens twice.
• Include its equivalent API amounts in model breakdowns, Today/7-day/30-day
  charts and Daily/Weekly/Monthly usage images.
• Recalculate existing records when read. No re-login, ledger rewrite or
  account reimport is required. Other model prices and audit dates are unchanged.
• Reject out-of-range numeric Token fields safely instead of crashing while
  loading a malformed record.

Official USD reference

The GPT-6.1 Sol model page
and API price table were checked
on October 1, 2026. Prices below are per 1 million tokens:

Context / mode	Input	Cached input	Output
Standard, up to 272K input	$2.00	$0.10	$10.00
Standard, over 272K input	$4.00	$0.20	$15.00
Fast, up to 272K input	$4.00	$0.20	$20.00
Fast, over 272K input	$8.00	$0.40	$30.00

• Exactly 272,000 input tokens uses short-context rates. Above that threshold,
  the full request uses twice the input/cache rates and 1.5 times the output rate.
• Fast is twice the applicable Standard rates. Apply it only with consistent
  vendor-confirmed Fast/priority evidence; a requested mode or bridge echo alone
  does not prove a grant. Unknown tiers keep a Standard reference estimate.
• Official cache-write pricing is $2.50/M for short Standard requests. The v1
  ledger does not establish cache-write overlap semantics, so nonzero cache-write
  records remain unpriced rather than receiving invented charges. Regional,
  Batch and Flex adjustments are not inferred from requested settings.
• Strict confirmed-tier bindings remain separate from reference estimates and
  require exact provider/model/context/tier evidence.

Validation and distribution

The M1 command-line test runner passed 370 tests across 49 classes. A
historical replay of 61,380 unique records preserved all Token totals and
all previously priced model amounts. The newly recognized GPT-6.1 Sol records
also matched an independent price calculation.

The Apple Silicon package is ad-hoc signed and not notarized. Native XCTest and
a clean-machine installation have not been run; see the
release checklist.

Equivalent API amounts are reference estimates, not subscription invoices or
actual charges.

AILSA SubSwitch 1.0.0(f) · 2026093001

AILSA SubSwitch 1.0.0(f)

Build: 2026093001. Add DeepSeek V4.1 Flash Token and equivalent API usage
accounting for the ATC local route.

DeepSeek V4.1 Flash

• Recognize atc-deepseek/DeepSeek-v4.1-Flash-EXL3 and show it as
  DeepSeek V4.1 Flash in model breakdowns and usage exports.
• Keep actual input/output Token counts from reported usage. Cache hits remain
  part of input and reasoning remains part of output; neither is added twice.
• Calculate the equivalent API amount from the official deepseek-flash price
  table, which currently serves DeepSeek V4.1 Flash. The exact local provider/model
  mapping does not establish a vendor charge or a quality equivalence for EXL3.
• Recalculate existing history when it is read. There is no need to sign in again
  or rewrite the original usage ledger. Other model prices are unchanged.

Official USD reference

The DeepSeek price table was
checked on September 30, 2026. Prices below are per 1 million tokens:

Period	Input, cache miss	Input, cache hit	Output
Peak	$0.30	$0.006	$1.20
Off-peak	$0.15	$0.003	$0.60

• Each record uses its own request timestamp. Peak hours are 01:00–04:00
  and 06:00–10:00 UTC, Monday–Friday, excluding Chinese public holidays.
  Window end times are exclusive. All other hours use off-peak rates.
• Fix the existing Flash reference calculation to honor Chinese holidays.
  The official 2026 holiday schedule
  is included. Weekends stay off-peak even when designated as make-up workdays.
• No holiday calendar is invented for an unverified year. A potentially peak
  record in such a year remains unpriced; known off-peak hours still work.
• Missing cache details retain a partial estimate. Missing/estimated usage,
  aborted requests, unknown identities and inconsistent token fields do not
  receive invented costs. The confirmed-tier catalog remains unchanged.

Validation and distribution

The full CLT asserting harness passes 360 tests across 48 classes on M1,
including request-time boundaries, holiday rates, cache accounting and identity
checks. This is not Apple's XCTest runtime. The Apple Silicon package is
ad-hoc signed and not notarized; see the release checklist.

Equivalent API amounts are reference estimates, not actual DeepSeek invoices,
subscription charges, or the operating cost of local hardware.

Historical replay preserved every Token count and changed only the newly supported DeepSeek reference amounts. Public packages exclude local credentials and account data.

AILSA SubSwitch 1.0.0(e) · 2026092803

AILSA SubSwitch 1.0.0(e)

Build: 2026092803. Export shareable usage images and include GPT-6 Sol
and GPT-6 Luna in equivalent API cost totals.

Usage images

• Use Export usage image in the top-right toolbar of Codex or Cursor usage.
  Choose Daily, Weekly (7 days), or Monthly (30 days), plus an ending date.
• The native save dialog lets you choose a folder and filename. After saving,
  Finder reveals the PNG. Suggested filenames include the provider, period
  and date range.
• Images use the app's black-and-white layout, with summary cards, model
  breakdowns and a full-height chart. Daily charts show 30-minute totals;
  weekly and monthly charts show daily totals.
• Exported charts honor the saved Token/USD preference and M/B token unit.
  Current periods are labeled with their cutoff time; older dates retain
  their own minute-level history.
• Images contain usage aggregates only, without accounts, credentials,
  request IDs, conversation text or local paths. The Antigravity interface
  and public promotional images are unchanged.

GPT-6 pricing

Official model pages were checked on September 28, 2026:

Model	Input / 1M	Cached input / 1M	Output / 1M
GPT-6 Sol	$2.00	$0.20	$10.00
GPT-6 Luna	$0.10	$0.01	$0.50

• Above 272,000 input tokens, the full request uses twice the input/cache
  rates and 1.5 times the output rate. Exactly 272,000 stays at short-context rates.
• Existing usage is recalculated when read. Token counts and other model
  prices are unchanged. Confirmed Fast pricing remains a separate pricing
  basis and requires actual vendor confirmation.
• Model-specific catalog additions keep older audit dates intact. Unpriced
  or inconsistent records remain unavailable instead of receiving invented costs.
• Includes the Grok Build Fast mapping introduced in 1.0.0(d).

Validation and distribution

• The full CLT asserting harness passes 348 tests across 47 classes, including
  cache discounts, long-context boundaries, date ranges, daylight-saving time,
  historical minute loading, family grouping and unknown prices. This is not
  Apple's XCTest runtime.
• Native offscreen PNG checks cover all three report periods, Chinese and
  English layouts, and both Token and USD charts. Historical replay verifies
  unchanged Token totals and changes limited to the newly priced models.
• The Apple Silicon package is ad-hoc signed and not notarized. See the
  release checklist.

Equivalent API amounts are reference estimates, not subscription invoices or
actual charges. Exported amounts sum the records that can be priced.

AILSA SubSwitch 1.0.0(d) · 2026092801

AILSA SubSwitch 1.0.0(d)

Build: 2026092801. Grok Build Fast now contributes its equivalent API
reference amount to the Grok 4.7 usage group, including previously recorded usage.

Fixed

• Recognize the resolved grok-4.7-build-fast route as a user-selected Grok 4.7
  reference mapping. Its Token usage and equivalent API amount appear together
  under Grok 4.7 in Today, 7-day and 30-day views.
• Recalculate existing local history when it is read. No sign-in, ledger edit
  or record deletion is needed. Other model totals remain unchanged.
• Retain Grok 4.7's cached-input discount and long-context threshold: prompts
  below 200,000 tokens use $2 input / $0.50 cached input / $6 output per million
  tokens; prompts at or above 200,000 use $4 / $1 / $12 for the entire request.
  Rates were checked against the official pricing page.
• A route name containing Fast does not prove Priority processing. Priority
  rates still require an explicitly confirmed grant. Reference mappings do not
  enter the separate vendor-confirmed pricing catalog.

Validation

• Regression coverage includes the Fast alias, cache discounts, the exact
  long-context boundary, confirmed versus unconfirmed Priority, missing model
  identity, request deduplication and unchanged unrelated model records.
• Historical replay verifies that Token totals remain identical and only the
  newly supported route gains reference pricing. No private usage records or
  personal screenshots are included in this release.
• The full CLT asserting harness passes 339 tests across 45 classes. This is
  not Apple's XCTest runtime.
• The Apple Silicon package is ad-hoc signed and not notarized. See the
  release checklist for validation scope.

Equivalent API amounts are reference estimates, not subscription invoices or
actual charges. Records without usable usage or pricing stay unavailable.

AILSA SubSwitch 1.0.0(c) · 2026092302

AILSA SubSwitch 1.0.0(c)

Build: 2026092302. Choose whether usage charts show Token usage or
equivalent API consumption in USD from Settings.

Choose a chart metric

• Open Settings → Usage Display → Usage chart display and select Token usage
  or Equivalent API consumption (USD). Token usage remains the default.
• The preference is saved between launches and updates open usage windows
  immediately. The selector stays in Settings to keep the usage dashboard clean.
• The choice applies to Today, 7-day and 30-day charts, including daily and weekly
  grouping, bar charts, line charts, axis labels and hover values.
• Model-list sorting is now independent of the chart metric. Changing a table's
  sort order no longer silently changes the historical chart's unit.

Consistent amounts and missing data

• Today's USD chart sums the same minute buckets used by its hover summaries.
  Historical USD charts use the same daily or weekly totals as the summary cards.
• USD values keep their existing pricing basis. API-equivalent amounts are not
  subscription invoices, and separate pricing bases are never added together.
• Records without a price remain unavailable instead of becoming zero-dollar
  usage. Known zero amounts and intervals without requests remain distinguishable.
• Antigravity keeps its quota-percentage charts because its source does not
  report per-request tokens or prices.

Validation and distribution

• The full CLT asserting harness passes 334 tests across 45 classes, including
  interval boundaries, unpriced data, monetary overflow and daily/weekly totals.
  This harness is not Apple's XCTest runtime.
• Native offscreen rendering checks cover both metrics for Today, 7-day and
  30-day charts, both historical chart styles, weekly grouping and Settings.
  The checks use synthetic records and isolated preferences.
• The Apple Silicon package is ad-hoc signed and not notarized. Native XCTest,
  clean-machine installation and other hardware configurations remain outside
  this revision's validation scope. See the release checklist.