[jehrr]

Fixes from the 2026-10-14 audit of main at 9f66b72.

Problems, and the behaviour now

Finding	Before	Now
F2 served page parses to 0 rows (high)	Real Goa capture with each hotel's id renamed: classified content, 4 properties stated, 0 rows. --pages 1/3 gave exit 4; good p1 + broken p2 gave exit 0, complete	stop_reason: parser_found_nothing: exit 5 with no rows, exit 6 partial, pages_failed: [2] after page 1. The response is saved as <out>_pageN_debug.html. "No Hotels Found" is still exit 4
F3 .env.example URL (high)	cp .env.example .env then --city Goa gave exit 2 ("--url already carries the query")	MAKEMYTRIP_URL ships commented out. Each engine records typed_url before env_config.apply, and an environment URL gives way to typed query flags. A typed --url with --city is still exit 2
F1 atomic writes (high)	A failure injected in json.dump left out.json/out.meta.json as 10-byte invalid fragments	_atomic(): temp file in the target's dir, flush, fsync, os.replace, unlink on error. New file gets 0o666 & ~umask, an existing file keeps its mode. Used by json, csv (newline="") and the sidecar
F4 CSV formulas	Written unchanged	String cells starting = + - @ TAB CR LF get ' after list joining. Numbers and JSON are untouched. csv_cells_escaped is in the sidecar, merged under the caller's extra
F5 solve budget	Stub solver: 3 solves for one page, 5 for a run by city name, against SOLVES_PER_PAGE = 1	page_flow.SolveBudget, one per page, checked before each purchase and never reset by retry or relaunch. The city lookup spends page 1's allowance. Now 1 and 1
F6 private citations	96 lines in 15 tracked files	0. Rules are stated in plain words. New guard: check_no_private_citations_in_shipped_files (tokens assembled from pieces)
F7 "credentials last about a day"	In the 401 hint, README, TROUBLESHOOTING, CONTRIBUTING, .env.example, bug template and canary	The text now says the credentials were refused (expired, revoked or mistyped). The suite check was rewritten to assert no lifetime is claimed
F8 / section 12 canary	Green on a datacentre refusal (exit 3). The data job was dispatch-only and skipped green without its secret	The datacentre job is documented as "datacentre refusal expected" and still treats exit 3 as a notice. The scraping-browser job now runs on the schedule, fails with ::error:: naming MAKEMYTRIP_CDP_ENDPOINT when the secret is absent, and fails on exit 3/4/5/6 or fewer than 60 rows
F9 --retries text	Described another repo's loop and measurement	Describes this repo's loop (while attempt < args.retries after the landing)
F10 .dockerignore	binance_rows.*	makemytrip_hotels.*, .env.*, canary*.json, *.page[0-9]*
F12 negative values	--delay, --retry-delay and --proxy-block-retries took negatives	argparse error naming the flag, exit 2. 0 is still legal
F14 stale example date	Active .env value	Moot: the line is commented out and says to put your own dates in

Evidence

• Offline suite results, no driver installed for the core run:

  Venv	Passed	Failed	Notes
  core	404	0	was 361
  playwright	407	0
  selenium	408	0
  puppeteer	406	0

  Each engine venv skipped only the other engines, and pip check was clean in each.

• ci_checks.py --secret-check: 44 files, clean.

• Through playwright_scraper.parse_args with MAKEMYTRIP_URL set:

  • --city Goa builds a Goa query;
  • --delay -1 gives error: argument --delay: --delay must be 0 or more;
  • a typed --url plus --city is refused.

• Planted-fault controls, run with PYTHONDONTWRITEBYTECODE=1 and __pycache__ removed. Each one went RED on its named check, and the restored tree was green:

  • budget reset per landing
  • lookup with its own budget
  • non-atomic json, csv and meta
  • temp cleanup removed
  • flat 0o644
  • no chmod (0600)
  • no CSV escape
  • escape before list join
  • housekeeping merged over extra
  • env URL treated as typed
  • parser-failure detection removed
  • plain float for --delay
  • a section-sign citation in CHANGELOG
  • a private-notes citation in page_flow.py
  • MAKEMYTRIP_URL uncommented

  The last control first crashed the env check instead of failing it by name. The check was fixed to report by name. A first version of the budget control looped forever, so the solver stub now raises after 20 calls.

• No paid calls, no proxies, no credentials and no live requests.

Not changed

• F11 (refuse --proxy/--fingerprint with --cdp-endpoint instead of warning): owner decision, because it changes behaviour for existing scripts.
• F13 (engines badge): covered by Audit follow-ups: honest engines badge, dated live checks, canary env report #2, and so are its env report and dated checks. This PR conflicts with Audit follow-ups: honest engines badge, dated live checks, canary env report #2 on CHANGELOG.md [Unreleased] and on canary.yml. Audit follow-ups: honest engines badge, dated live checks, canary env report #2 was not touched.
• F10 comments in captcha_solver.py that record measurements on sibling sites. They are shared-solver provenance, not residue in this repo's behaviour.
• No version bump, tag or release. The repo description and topics needed no edit. Release notes for v0.1.0 and v0.1.1 contain no private citations.

Owner action

The MAKEMYTRIP_CDP_ENDPOINT secret (set 2026-09-25) has to be valid, or the scheduled canary goes red. That is intended: red means nothing tested the site.

🤖 Generated with Claude Code